wehub/fastai--fastai
1
0
Fork 0
Code Issues Pull Requests Actions Deployments Models Agent Notes Packages Projects Releases Wiki Activity
Files
main
fastai--fastai/nbs/31_text.data.ipynb
T
wehub-resource-sync 844360907f chore: import upstream snapshot with attribution
2026-07-13 13:21:43 +08:00

2368 lines
100 KiB
Plaintext
Raw Permalink Blame History

{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"id": "ed83a136",
"metadata": {},
"outputs": [],
"source": [
"#| hide\n",
"#| eval: false\n",
"! [ -e /content ] && pip install -Uqq fastai # upgrade fastai on colab"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "21b87e7e",
"metadata": {},
"outputs": [],
"source": [
"#| export\n",
"from fastai.torch_basics import *\n",
"from fastai.data.all import *\n",
"from fastai.text.core import *"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "6eb5f1f8",
"metadata": {},
"outputs": [],
"source": [
"#| hide\n",
"from nbdev.showdoc import *"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "a937268b",
"metadata": {},
"outputs": [],
"source": [
"#| default_exp text.data\n",
"#| default_cls_lvl 3"
]
},
{
"cell_type": "markdown",
"id": "15911aa5",
"metadata": {},
"source": [
"# Text data\n",
"\n",
"> Functions and transforms to help gather text data in a `Datasets`"
]
},
{
"cell_type": "markdown",
"id": "8d740576",
"metadata": {},
"source": [
"## Backwards\n",
"\n",
"Reversing the text can provide higher accuracy with an ensemble with a forward model. All that is needed is a `type_tfm` that will reverse the text as it is brought in:"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "060cd127",
"metadata": {},
"outputs": [],
"source": [
"#| export\n",
"def reverse_text(x): return x.flip(0)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "8d589d35",
"metadata": {},
"outputs": [],
"source": [
"t = tensor([0,1,2])\n",
"r = reverse_text(t)\n",
"test_eq(r, tensor([2,1,0]))"
]
},
{
"cell_type": "markdown",
"id": "3f237dcb",
"metadata": {},
"source": [
"## Numericalizing"
]
},
{
"cell_type": "markdown",
"id": "fccf4f27",
"metadata": {},
"source": [
"Numericalization is the step in which we convert tokens to integers. The first step is to build a correspondence token to index that is called a vocab."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f58cfb2a",
"metadata": {},
"outputs": [],
"source": [
"#| export\n",
"def make_vocab(count, min_freq=3, max_vocab=60000, special_toks=None):\n",
" \"Create a vocab of `max_vocab` size from `Counter` `count` with items present more than `min_freq`\"\n",
" vocab = [o for o,c in count.most_common(max_vocab) if c >= min_freq]\n",
" special_toks = ifnone(special_toks, defaults.text_spec_tok)\n",
" for o in reversed(special_toks): #Make sure all special tokens are in the vocab\n",
" if o in vocab: vocab.remove(o)\n",
" vocab.insert(0, o)\n",
" vocab = vocab[:max_vocab]\n",
" return vocab + [f'xxfake' for i in range(0, 8-len(vocab)%8)]"
]
},
{
"cell_type": "markdown",
"id": "bfc8f052",
"metadata": {},
"source": [
"If there are more than `max_vocab` tokens, the ones kept are the most frequent.\n",
"\n",
":::{.callout-note}\n",
"\n",
"For performance when using mixed precision, the vocabulary is always made of size a multiple of 8, potentially by adding `xxfake` tokens.\n",
"\n",
":::"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c56eff90",
"metadata": {},
"outputs": [],
"source": [
"count = Counter(['a', 'a', 'a', 'a', 'b', 'b', 'c', 'c', 'd'])\n",
"test_eq(set([x for x in make_vocab(count) if not x.startswith('xxfake')]), \n",
" set(defaults.text_spec_tok + 'a'.split()))\n",
"test_eq(len(make_vocab(count))%8, 0)\n",
"test_eq(set([x for x in make_vocab(count, min_freq=1) if not x.startswith('xxfake')]), \n",
" set(defaults.text_spec_tok + 'a b c d'.split()))\n",
"test_eq(set([x for x in make_vocab(count,max_vocab=12, min_freq=1) if not x.startswith('xxfake')]), \n",
" set(defaults.text_spec_tok + 'a b c'.split()))"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3b0e7f5c",
"metadata": {},
"outputs": [],
"source": [
"#| export\n",
"class TensorText(TensorBase): pass\n",
"class LMTensorText(TensorText): pass\n",
"\n",
"TensorText.__doc__ = \"Semantic type for a tensor representing text\"\n",
"LMTensorText.__doc__ = \"Semantic type for a tensor representing text in language modeling\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "0b36808c",
"metadata": {},
"outputs": [],
"source": [
"#| export\n",
"class Numericalize(Transform):\n",
" \"Reversible transform of tokenized texts to numericalized ids\"\n",
" def __init__(self, vocab=None, min_freq=3, max_vocab=60000, special_toks=None):\n",
" store_attr('vocab,min_freq,max_vocab,special_toks')\n",
" self.o2i = None if vocab is None else defaultdict(int, {v:k for k,v in enumerate(vocab)})\n",
"\n",
" def setups(self, dsets):\n",
" if dsets is None: return\n",
" if self.vocab is None:\n",
" count = dsets.counter if getattr(dsets, 'counter', None) is not None else Counter(p for o in dsets for p in o)\n",
" if self.special_toks is None and hasattr(dsets, 'special_toks'):\n",
" self.special_toks = dsets.special_toks\n",
" self.vocab = make_vocab(count, min_freq=self.min_freq, max_vocab=self.max_vocab, special_toks=self.special_toks)\n",
" self.o2i = defaultdict(int, {v:k for k,v in enumerate(self.vocab) if v != 'xxfake'})\n",
"\n",
" def encodes(self, o): return TensorText(tensor([self.o2i [o_] for o_ in o]))\n",
" def decodes(self, o): return L(self.vocab[o_] for o_ in o)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "94eb9b08",
"metadata": {},
"outputs": [],
"source": [
"num = Numericalize(min_freq=2)\n",
"num.setup(L('This is an example of text'.split(), 'this is another text'.split()))"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "2c55f545",
"metadata": {},
"outputs": [],
"source": [
"start = 'This is an example of text '"
]
},
{
"cell_type": "markdown",
"id": "e223241b",
"metadata": {},
"source": [
"If no `vocab` is passed, one is created at setup from the data, using `make_vocab` with `min_freq` and `max_vocab`."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e98d0e4e",
"metadata": {},
"outputs": [],
"source": [
"start = 'This is an example of text'\n",
"num = Numericalize(min_freq=1)\n",
"num.setup(L(start.split(), 'this is another text'.split()))\n",
"test_eq(set([x for x in num.vocab if not x.startswith('xxfake')]), \n",
" set(defaults.text_spec_tok + 'This is an example of text this another'.split()))\n",
"test_eq(len(num.vocab)%8, 0)\n",
"t = num(start.split())\n",
"\n",
"test_eq(t, tensor([11, 9, 12, 13, 14, 10]))\n",
"test_eq(num.decode(t), start.split())"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "4dc2e945",
"metadata": {},
"outputs": [],
"source": [
"num = Numericalize(min_freq=2)\n",
"num.setup(L('This is an example of text'.split(), 'this is another text'.split()))\n",
"test_eq(set([x for x in num.vocab if not x.startswith('xxfake')]), \n",
" set(defaults.text_spec_tok + 'is text'.split()))\n",
"test_eq(len(num.vocab)%8, 0)\n",
"t = num(start.split())\n",
"test_eq(t, tensor([0, 9, 0, 0, 0, 10]))\n",
"test_eq(num.decode(t), f'{UNK} is {UNK} {UNK} {UNK} text'.split())"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "732cbec3",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
"<style>\n",
" /* Turns off some styling */\n",
" progress {\n",
" /* gets rid of default border in Firefox and Opera. */\n",
" border: none;\n",
" /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
" background-size: auto;\n",
" }\n",
" progress:not([value]), progress:not([value])::-webkit-progress-bar {\n",
" background: repeating-linear-gradient(45deg, #7e7e7e, #7e7e7e 10px, #5c5c5c 10px, #5c5c5c 20px);\n",
" }\n",
" .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
" background: #F44336;\n",
" }\n",
"</style>\n"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"#| hide\n",
"df = pd.DataFrame({'texts': ['This is an example of text', 'this is another text']})\n",
"tl = TfmdLists(df, [attrgetter('text'), Tokenizer.from_df('texts'), Numericalize(min_freq=2)])\n",
"test_eq(tl, [tensor([2, 8, 9, 10, 0, 0, 0, 11]), tensor([2, 9, 10, 0, 11])])"
]
},
{
"cell_type": "markdown",
"id": "803bb208",
"metadata": {},
"source": [
"## LM_DataLoader -"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b27dd16b",
"metadata": {},
"outputs": [],
"source": [
"#| export\n",
"def _maybe_first(o): return o[0] if isinstance(o, tuple) else o"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "7fd53e96",
"metadata": {},
"outputs": [],
"source": [
"#| export\n",
"def _get_tokenizer(ds):\n",
" tok = getattr(ds, 'tokenizer', None)\n",
" if isinstance(tok, Tokenizer): return tok\n",
" if isinstance(tok, (list,L)):\n",
" for t in tok:\n",
" if isinstance(t, Tokenizer): return t"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "8e71a626",
"metadata": {},
"outputs": [],
"source": [
"#| export\n",
"def _get_lengths(ds):\n",
" tok = _get_tokenizer(ds)\n",
" if tok is None: return\n",
" return tok.get_lengths(ds.items)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1f66c43d",
"metadata": {},
"outputs": [],
"source": [
"#| export\n",
"#TODO: add backward\n",
"@delegates()\n",
"class LMDataLoader(TfmdDL):\n",
" \"A `DataLoader` suitable for language modeling\"\n",
" def __init__(self, dataset, lens=None, cache=2, bs=64, seq_len=72, num_workers=0, **kwargs):\n",
" self.items = ReindexCollection(dataset, cache=cache, tfm=_maybe_first)\n",
" self.seq_len = seq_len\n",
" if lens is None: lens = _get_lengths(dataset)\n",
" if lens is None: lens = [len(o) for o in self.items]\n",
" self.lens = ReindexCollection(lens, idxs=self.items.idxs)\n",
" # The \"-1\" is to allow for final label, we throw away the end that's less than bs\n",
" corpus = round_multiple(sum(lens)-1, bs, round_down=True)\n",
" self.bl = corpus//bs #bl stands for batch length\n",
" self.n_batches = self.bl//(seq_len) + int(self.bl%seq_len!=0)\n",
" self.last_len = self.bl - (self.n_batches-1)*seq_len\n",
" self.make_chunks()\n",
" super().__init__(dataset=dataset, bs=bs, num_workers=num_workers, **kwargs)\n",
" self.n = self.n_batches*bs\n",
"\n",
" def make_chunks(self): self.chunks = Chunks(self.items, self.lens)\n",
" def shuffle_fn(self,idxs):\n",
" self.items.shuffle()\n",
" self.make_chunks()\n",
" return idxs\n",
"\n",
" def create_item(self, seq):\n",
" if seq is None: seq = 0\n",
" if seq>=self.n: raise IndexError\n",
" sl = self.last_len if seq//self.bs==self.n_batches-1 else self.seq_len\n",
" st = (seq%self.bs)*self.bl + (seq//self.bs)*self.seq_len\n",
" txt = self.chunks[st : st+sl+1]\n",
" return LMTensorText(txt[:-1]),txt[1:]\n",
"\n",
" @delegates(TfmdDL.new)\n",
" def new(self, dataset=None, seq_len=None, **kwargs):\n",
" lens = self.lens.coll if dataset is None else None\n",
" seq_len = self.seq_len if seq_len is None else seq_len\n",
" return super().new(dataset=dataset, lens=lens, seq_len=seq_len, **kwargs)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "66a25ec5",
"metadata": {},
"outputs": [
{
"data": {
"text/markdown": [
"<h2 id=\"LMDataLoader\" class=\"doc_header\"><code>class</code> <code>LMDataLoader</code><a href=\"\" class=\"source_link\" style=\"float:right\">[source]</a></h2>\n",
"\n",
"> <code>LMDataLoader</code>(**`dataset`**, **`lens`**=*`None`*, **`cache`**=*`2`*, **`bs`**=*`64`*, **`seq_len`**=*`72`*, **`num_workers`**=*`0`*, **`shuffle`**:`bool`=*`False`*, **`verbose`**:`bool`=*`False`*, **`do_setup`**:`bool`=*`True`*, **`pin_memory`**=*`False`*, **`timeout`**=*`0`*, **`batch_size`**=*`None`*, **`drop_last`**=*`False`*, **`indexed`**=*`None`*, **`n`**=*`None`*, **`device`**=*`None`*, **`persistent_workers`**=*`False`*, **`pin_memory_device`**=*`''`*, **`wif`**=*`None`*, **`before_iter`**=*`None`*, **`after_item`**=*`None`*, **`before_batch`**=*`None`*, **`after_batch`**=*`None`*, **`after_iter`**=*`None`*, **`create_batches`**=*`None`*, **`create_item`**=*`None`*, **`create_batch`**=*`None`*, **`retain`**=*`None`*, **`get_idxs`**=*`None`*, **`sample`**=*`None`*, **`shuffle_fn`**=*`None`*, **`do_batch`**=*`None`*) :: [`TfmdDL`](/data.core.html#TfmdDL)\n",
"\n",
"A [`DataLoader`](/data.load.html#DataLoader) suitable for language modeling"
],
"text/plain": [
"<IPython.core.display.Markdown object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"show_doc(LMDataLoader, title_level=2)"
]
},
{
"cell_type": "markdown",
"id": "c0f8f541",
"metadata": {},
"source": [
"`dataset` should be a collection of numericalized texts for this to work. `lens` can be passed for optimizing the creation, otherwise, the `LMDataLoader` will do a full pass of the `dataset` to compute them. `cache` is used to avoid reloading items unnecessarily.\n",
"\n",
"The `LMDataLoader` will concatenate all texts (maybe `shuffle`d) in one big stream, split it in `bs` contiguous sentences, then go through those `seq_len` at a time."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b6875691",
"metadata": {},
"outputs": [],
"source": [
"#| hide\n",
"bs,sl = 4,3\n",
"ints = L([0,1,2,3,4],[5,6,7,8,9,10],[11,12,13,14,15,16,17,18],[19,20],[21,22]).map(tensor)\n",
"dl = LMDataLoader(ints, bs=bs, seq_len=sl)\n",
"list(dl)\n",
"test_eq(list(dl),\n",
" [[tensor([[0, 1, 2], [5, 6, 7], [10, 11, 12], [15, 16, 17]]),\n",
" tensor([[1, 2, 3], [6, 7, 8], [11, 12, 13], [16, 17, 18]])],\n",
" [tensor([[3, 4], [8, 9], [13, 14], [18, 19]]),\n",
" tensor([[4, 5], [9, 10], [14, 15], [19, 20]])]])"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "364b499b",
"metadata": {},
"outputs": [],
"source": [
"bs,sl = 4,3\n",
"ints = L([0,1,2,3,4],[5,6,7,8,9,10],[11,12,13,14,15,16,17,18],[19,20],[21,22,23],[24]).map(tensor)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "71a799d4",
"metadata": {},
"outputs": [],
"source": [
"dl = LMDataLoader(ints, bs=bs, seq_len=sl)\n",
"test_eq(list(dl),\n",
" [[tensor([[0, 1, 2], [6, 7, 8], [12, 13, 14], [18, 19, 20]]),\n",
" tensor([[1, 2, 3], [7, 8, 9], [13, 14, 15], [19, 20, 21]])],\n",
" [tensor([[3, 4, 5], [ 9, 10, 11], [15, 16, 17], [21, 22, 23]]),\n",
" tensor([[4, 5, 6], [10, 11, 12], [16, 17, 18], [22, 23, 24]])]])"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "eb803149",
"metadata": {},
"outputs": [],
"source": [
"#| hide\n",
"#Check lens work\n",
"dl = LMDataLoader(ints, lens=ints.map(len), bs=bs, seq_len=sl)\n",
"test_eq(list(dl),\n",
" [[tensor([[0, 1, 2], [6, 7, 8], [12, 13, 14], [18, 19, 20]]),\n",
" tensor([[1, 2, 3], [7, 8, 9], [13, 14, 15], [19, 20, 21]])],\n",
" [tensor([[3, 4, 5], [ 9, 10, 11], [15, 16, 17], [21, 22, 23]]),\n",
" tensor([[4, 5, 6], [10, 11, 12], [16, 17, 18], [22, 23, 24]])]])"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "96cfd5d2",
"metadata": {},
"outputs": [],
"source": [
"dl = LMDataLoader(ints, bs=bs, seq_len=sl, shuffle=True)\n",
"for x,y in dl: test_eq(x[:,1:], y[:,:-1])\n",
"((x0,y0), (x1,y1)) = tuple(dl)\n",
"#Second batch begins where first batch ended\n",
"test_eq(y0[:,-1], x1[:,0]) \n",
"test_eq(type(x0), LMTensorText)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "920d47d8",
"metadata": {},
"outputs": [],
"source": [
"#| hide\n",
"#test new works\n",
"dl = LMDataLoader(ints, bs=bs, seq_len=sl, shuffle=True)\n",
"dl1 = dl.new()\n",
"test_eq(dl1.seq_len, sl)\n",
"dl2 = dl.new(seq_len=2)\n",
"test_eq(dl2.seq_len, 2)"
]
},
{
"cell_type": "markdown",
"id": "74d29be0",
"metadata": {},
"source": [
"### Showing -"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "5e86ad40",
"metadata": {},
"outputs": [],
"source": [
"#| export\n",
"@dispatch\n",
"def show_batch(x: TensorText, y, samples, ctxs=None, max_n=10, trunc_at=150, **kwargs):\n",
" if ctxs is None: ctxs = get_empty_df(min(len(samples), max_n))\n",
" if trunc_at is not None: samples = L((s[0].truncate(trunc_at),*s[1:]) for s in samples)\n",
" ctxs = get_show_batch_func(object)(x, y, samples, max_n=max_n, ctxs=ctxs, **kwargs)\n",
" display_df(pd.DataFrame(ctxs))\n",
" return ctxs"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "33cd4023",
"metadata": {},
"outputs": [],
"source": [
"#| export\n",
"@dispatch\n",
"def show_batch(x: LMTensorText, y, samples, ctxs=None, max_n=10, trunc_at=150, **kwargs):\n",
" samples = L((s[0].truncate(trunc_at), s[1].truncate(trunc_at)) for s in samples)\n",
" return get_show_batch_func(TensorText)(x, None, samples, ctxs=ctxs, max_n=max_n, trunc_at=None, **kwargs)"
]
},
{
"cell_type": "markdown",
"id": "27c243f3",
"metadata": {},
"source": [
"## Classification"
]
},
{
"cell_type": "markdown",
"id": "c623a4ec",
"metadata": {},
"source": [
"For classification, we deal with the fact that texts don't all have the same length by using padding."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "2d6f750e",
"metadata": {},
"outputs": [],
"source": [
"#| export\n",
"class Pad_Input(ItemTransform):\n",
" def encodes(self,samples, pad_idx=1, pad_fields=0, pad_first=False, backwards=False):\n",
" \"Function that collect `samples` and adds padding\"\n",
" self.pad_idx = pad_idx\n",
" pad_fields = L(pad_fields)\n",
" max_len_l = pad_fields.map(lambda f: max([len(s[f]) for s in samples]))\n",
" if backwards: pad_first = not pad_first\n",
" def _f(field_idx, x):\n",
" if field_idx not in pad_fields: return x\n",
" idx = pad_fields.items.index(field_idx) #TODO: remove items if L.index is fixed\n",
" sl = slice(-len(x), sys.maxsize) if pad_first else slice(0, len(x))\n",
" pad = x.new_zeros(max_len_l[idx]-x.shape[0])+pad_idx\n",
" x1 = torch.cat([pad, x] if pad_first else [x, pad])\n",
" if backwards: x1 = x1.flip(0)\n",
" return retain_type(x1, x)\n",
" return [tuple(map(lambda idxx: _f(*idxx), enumerate(s))) for s in samples]\n",
" def decodes(self, o:TensorText):\n",
" pad_idx = self.pad_idx if hasattr(self,'pad_idx') else 1\n",
" return o[o != pad_idx]\n",
"pad_input=Pad_Input()"
]
},
{
"cell_type": "markdown",
"id": "ed36063d",
"metadata": {},
"source": [
"`pad_idx` is used for the padding, and the padding is applied to the `pad_fields` of the samples. The padding is applied at the beginning if `pad_first` is `True`, and if `backwards` is added, the tensors are flipped."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "725cf8c6",
"metadata": {},
"outputs": [],
"source": [
"test_eq(pad_input([(tensor([1,2,3]),1), (tensor([4,5]), 2), (tensor([6]), 3)], pad_idx=0), \n",
" [(tensor([1,2,3]),1), (tensor([4,5,0]),2), (tensor([6,0,0]), 3)])\n",
"test_eq(pad_input([(tensor([1,2,3]), (tensor([6]))), (tensor([4,5]), tensor([4,5])), (tensor([6]), (tensor([1,2,3])))], pad_idx=0, pad_fields=1), \n",
" [(tensor([1,2,3]),(tensor([6,0,0]))), (tensor([4,5]),tensor([4,5,0])), ((tensor([6]),tensor([1, 2, 3])))])\n",
"test_eq(pad_input([(tensor([1,2,3]),1), (tensor([4,5]), 2), (tensor([6]), 3)], pad_idx=0, pad_first=True), \n",
" [(tensor([1,2,3]),1), (tensor([0,4,5]),2), (tensor([0,0,6]), 3)])\n",
"test_eq(pad_input([(tensor([1,2,3]),1), (tensor([4,5]), 2), (tensor([6]), 3)], pad_idx=0, backwards=True), \n",
" [(tensor([3,2,1]),1), (tensor([5,4,0]),2), (tensor([6,0,0]), 3)])\n",
"x = pad_input([(TensorText([1,2,3]),1), (TensorText([4,5]), 2), (TensorText([6]), 3)], pad_idx=0)\n",
"test_eq(x, [(tensor([1,2,3]),1), (tensor([4,5,0]), 2), (tensor([6,0,0]), 3)])\n",
"test_eq(pad_input.decode(x[1][0]), tensor([4,5]))"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "65fbbbe6",
"metadata": {},
"outputs": [],
"source": [
"#| hide\n",
"#Check retain type\n",
"x = [(TensorText([1,2,3]),1), (TensorText([4,5]), 2), (TensorText([6]), 3)]\n",
"y = pad_input(x, pad_idx=0)\n",
"for s in y: test_eq(type(s[0]), TensorText)"
]
},
{
"cell_type": "markdown",
"id": "dff3662f",
"metadata": {},
"source": [
"Pads `x` with `pad_idx` to length `pad_len`. If `pad_first` is false, all padding is appended to `x`, until `x` is len `pad_len`. Otherwise ff `pad_first` is true, then chunks of size `seq_len` are prepended to `x`, the remainder of the padding is appended to `x`. "
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "852c32d5",
"metadata": {},
"outputs": [],
"source": [
"#| export\n",
"def pad_chunk(x,pad_idx=1, pad_first=True, seq_len=72, pad_len=10):\n",
" \"Pad `x` by adding padding by chunks of size `seq_len`\"\n",
" l = pad_len - x.shape[0]\n",
" pad_chunk = x.new_zeros((l//seq_len) * seq_len) + pad_idx\n",
" pad_res = x.new_zeros(l % seq_len) + pad_idx\n",
" x1 = torch.cat([pad_chunk, x, pad_res]) if pad_first else torch.cat([x, pad_chunk, pad_res])\n",
" return retain_type(x1, x)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "86d72f6b",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"pad_first: tensor([0, 0, 0, 1, 2, 3, 0, 0])\n",
"pad_last: tensor([1, 2, 3, 0, 0, 0, 0, 0])\n"
]
}
],
"source": [
"print('pad_first: ',pad_chunk(torch.tensor([1,2,3]),seq_len=3,pad_idx=0,pad_len=8))\n",
"print('pad_last: ',pad_chunk(torch.tensor([1,2,3]),seq_len=3,pad_idx=0,pad_len=8,pad_first=False))"
]
},
{
"cell_type": "markdown",
"id": "2e799b26",
"metadata": {},
"source": [
"`pad_input_chunk` is the version of `pad_chunk` that works over a list of lists. "
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "32faa119",
"metadata": {},
"outputs": [],
"source": [
"#| export\n",
"@delegates(pad_chunk)\n",
"def pad_input_chunk(samples, n_inp=1,**kwargs):\n",
" \"Pad `samples` by adding padding by chunks of size `seq_len`\"\n",
" max_len = max([len(s[n]) for s in samples for n in range(n_inp)])\n",
" padeds = [[pad_chunk(s[n],pad_len=max_len,**kwargs) for n in range(n_inp) ] for s in samples]\n",
" return [(*p, *s[n_inp:]) for p,s in zip(padeds,samples)]"
]
},
{
"cell_type": "markdown",
"id": "9a94664a",
"metadata": {},
"source": [
"The difference with the base `pad_input` is that most of the padding is applied first (if `pad_first=True`) or at the end (if `pad_first=False`) but only by a round multiple of `seq_len`. The rest of the padding is applied to the end (or the beginning if `pad_first=False`). This is to work with `SequenceEncoder` with recurrent models."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "235905c7",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"[(TensorText([1, 2, 3, 4, 5, 6]), TensorText([0, 0, 0, 1, 2, 0]), 1)]"
]
},
"execution_count": null,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"pad_input_chunk([(TensorText([1,2,3,4,5,6]),TensorText([1,2]),1)], pad_idx=0, seq_len=3,n_inp=2)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "0ac9381a",
"metadata": {},
"outputs": [],
"source": [
"test_eq(pad_input_chunk([(tensor([1,2,3,4,5,6]),1), (tensor([1,2,3]), 2), (tensor([1,2]), 3)], pad_idx=0, seq_len=2), \n",
" [(tensor([1,2,3,4,5,6]),1), (tensor([0,0,1,2,3,0]),2), (tensor([0,0,0,0,1,2]), 3)])\n",
"test_eq(pad_input_chunk([(tensor([1,2,3,4,5,6]),), (tensor([1,2,3]),), (tensor([1,2]),)], pad_idx=0, seq_len=2), \n",
" [(tensor([1,2,3,4,5,6]),), (tensor([0,0,1,2,3,0]),), (tensor([0,0,0,0,1,2]),)])\n",
"test_eq(pad_input_chunk([(tensor([1,2,3,4,5,6]),), (tensor([1,2,3]),), (tensor([1,2]),)], pad_idx=0, seq_len=2, pad_first=False), \n",
" [(tensor([1,2,3,4,5,6]),), (tensor([1,2,3,0,0,0]),), (tensor([1,2,0,0,0,0]),)])\n",
"\n",
"test_eq(pad_input_chunk([(TensorText([1,2,3,4,5,6]),TensorText([1,2]),1)], pad_idx=0, seq_len=2,n_inp=2), \n",
" [(TensorText([1,2,3,4,5,6]),TensorText([0,0,0,0,1,2]),1)])"
]
},
{
"cell_type": "markdown",
"id": "b9f168ea",
"metadata": {},
"source": [
"`Transform` version of `pad_input_chunk`. This version supports types, decoding, and the other functionality of `Transform`"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3563df07",
"metadata": {},
"outputs": [],
"source": [
"#| export\n",
"class Pad_Chunk(DisplayedTransform):\n",
" \"Pad `samples` by adding padding by chunks of size `seq_len`\"\n",
" def __init__(self, pad_idx=1, pad_first=True, seq_len=72,decode=True,**kwargs):\n",
" store_attr('pad_idx, pad_first, seq_len,seq_len')\n",
" super().__init__(**kwargs)\n",
" def before_call(self, b):\n",
" \"Set `self.max_len` before encodes\" \n",
" self.max_len = max([x.shape[0] for xs in b for x in xs if isinstance(x,TensorText)])\n",
" def __call__(self, b, **kwargs):\n",
" self.before_call(b)\n",
" return super().__call__(tuple(b), **kwargs)\n",
" def encodes(self, x:TensorText):\n",
" return pad_chunk(x,pad_idx=self.pad_idx, pad_first=self.pad_first, seq_len=self.seq_len, pad_len=self.max_len)\n",
" def decodes(self, o:TensorText):\n",
" return o[o != self.pad_idx] if self.decode else o"
]
},
{
"cell_type": "markdown",
"id": "41a6a83d",
"metadata": {},
"source": [
"Here is an example of `Pad_Chunk`"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c10e1a38",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Inputs: (TensorText([1, 2, 3, 4, 5, 6]), TensorText([1, 2]), 1)\n",
"Encoded: (TensorText([1, 2, 3, 4, 5, 6]), TensorText([0, 0, 0, 1, 2, 0]), 1)\n",
"Decoded: (TensorText([1, 2, 3, 4, 5, 6]), TensorText([1, 2]), 1)\n"
]
}
],
"source": [
"pc=Pad_Chunk(pad_idx=0,seq_len=3)\n",
"out=pc([(TensorText([1,2,3,4,5,6]),TensorText([1,2]),1)])\n",
"print('Inputs: ',*[(TensorText([1,2,3,4,5,6]),TensorText([1,2]),1)])\n",
"print('Encoded: ',*out)\n",
"print('Decoded: ',*pc.decode(out))"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "2ec60d6d",
"metadata": {},
"outputs": [],
"source": [
"pc=Pad_Chunk(pad_idx=0, seq_len=2)\n",
"test_eq(pc([(TensorText([1,2,3,4,5,6]),1), (TensorText([1,2,3]), 2), (TensorText([1,2]), 3)]), \n",
" [(tensor([1,2,3,4,5,6]),1), (tensor([0,0,1,2,3,0]),2), (tensor([0,0,0,0,1,2]), 3)])\n",
"\n",
"pc=Pad_Chunk(pad_idx=0, seq_len=2)\n",
"test_eq(pc([(TensorText([1,2,3,4,5,6]),), (TensorText([1,2,3]),), (TensorText([1,2]),)]), \n",
" [(tensor([1,2,3,4,5,6]),), (tensor([0,0,1,2,3,0]),), (tensor([0,0,0,0,1,2]),)])\n",
"\n",
"pc=Pad_Chunk(pad_idx=0, seq_len=2, pad_first=False)\n",
"test_eq(pc([(TensorText([1,2,3,4,5,6]),), (TensorText([1,2,3]),), (TensorText([1,2]),)]), \n",
" [(tensor([1,2,3,4,5,6]),), (tensor([1,2,3,0,0,0]),), (tensor([1,2,0,0,0,0]),)])\n",
"\n",
"pc=Pad_Chunk(pad_idx=0, seq_len=2)\n",
"test_eq(pc([(TensorText([1,2,3,4,5,6]),TensorText([1,2]),1)]), \n",
" [(TensorText([1,2,3,4,5,6]),TensorText([0,0,0,0,1,2]),1)])"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c61cd6bb",
"metadata": {},
"outputs": [],
"source": [
"#| export\n",
"def _default_sort(x): return len(x[0])\n",
"\n",
"@delegates(TfmdDL)\n",
"class SortedDL(TfmdDL):\n",
" \"A `DataLoader` that goes throught the item in the order given by `sort_func`\"\n",
" def __init__(self, dataset, sort_func=None, res=None, **kwargs):\n",
" super().__init__(dataset, **kwargs)\n",
" self.sort_func = _default_sort if sort_func is None else sort_func\n",
" if res is None and self.sort_func == _default_sort: res = _get_lengths(dataset)\n",
" self.res = [self.sort_func(self.do_item(i)) for i in range_of(self.dataset)] if res is None else res\n",
" if len(self.res) > 0: self.idx_max = np.argmax(self.res)\n",
"\n",
" def get_idxs(self):\n",
" idxs = super().get_idxs()\n",
" if self.shuffle: return idxs\n",
" return sorted(idxs, key=lambda i: self.res[i], reverse=True)\n",
"\n",
" def shuffle_fn(self,idxs):\n",
" idxs = np.random.permutation(len(self.dataset))\n",
" idx_max = np.where(idxs==self.idx_max)[0][0]\n",
" idxs[0],idxs[idx_max] = idxs[idx_max],idxs[0]\n",
" sz = self.bs*50\n",
" chunks = [idxs[i:i+sz] for i in range(0, len(idxs), sz)]\n",
" chunks = [sorted(s, key=lambda i: self.res[i], reverse=True) for s in chunks]\n",
" sort_idx = np.concatenate(chunks)\n",
"\n",
" sz = self.bs\n",
" batches = [sort_idx[i:i+sz] for i in range(0, len(sort_idx), sz)]\n",
" sort_idx = np.concatenate(np.random.permutation(batches[1:-1])) if len(batches) > 2 else np.array([],dtype=int)\n",
" sort_idx = np.concatenate((batches[0], sort_idx) if len(batches)==1 else (batches[0], sort_idx, batches[-1]))\n",
" return iter(sort_idx)\n",
"\n",
" @delegates(TfmdDL.new)\n",
" def new(self, dataset=None, **kwargs):\n",
" if 'val_res' in kwargs and kwargs['val_res'] is not None: res = kwargs['val_res']\n",
" else: res = self.res if dataset is None else None\n",
" return super().new(dataset=dataset, res=res, **kwargs)"
]
},
{
"cell_type": "markdown",
"id": "2d9b1143",
"metadata": {},
"source": [
"`res` is the result of `sort_func` applied on all elements of the `dataset`. You can pass it if available to make the init much faster by avoiding an initial pass over the whole dataset. For example if sorting by text length (as in the default `sort_func`, called `_default_sort`) you should pass a list with the length of each element in `dataset` to `res` to take advantage of this speed-up. \n",
"\n",
"To get the same init speed-up for the validation set, `val_res` (a list of text lengths for your validation set) can be passed to the `kwargs` argument of `SortedDL`. Below is an example to reduce the init time by passing a list of text lengths for both the training set and the validation set:\n",
"\n",
"```\n",
"# Pass the training dataset text lengths to SortedDL\n",
"srtd_dl=partial(SortedDL, res = train_text_lens)\n",
"\n",
"# Pass the validation dataset text lengths \n",
"dl_kwargs = [{},{'val_res': val_text_lens}]\n",
"\n",
"# init our Datasets \n",
"dsets = Datasets(...) \n",
"\n",
"# init our Dataloaders\n",
"dls = dsets.dataloaders(...,dl_type = srtd_dl, dl_kwargs = dl_kwargs)\n",
"```\n",
"\n",
"If `shuffle` is `True`, this will shuffle a bit the results of the sort to have items of roughly the same size in batches, but not in the exact sorted order."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "fa1de447",
"metadata": {},
"outputs": [],
"source": [
"ds = [(tensor([1,2]),1), (tensor([3,4,5,6]),2), (tensor([7]),3), (tensor([8,9,10]),4)]\n",
"dl = SortedDL(ds, bs=2, before_batch=partial(pad_input, pad_idx=0))\n",
"test_eq(list(dl), [(tensor([[ 3, 4, 5, 6], [ 8, 9, 10, 0]]), tensor([2, 4])), \n",
" (tensor([[1, 2], [7, 0]]), tensor([1, 3]))])"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "d7f568cf",
"metadata": {},
"outputs": [],
"source": [
"ds = [(tensor(range(random.randint(1,10))),i) for i in range(101)]\n",
"dl = SortedDL(ds, bs=2, create_batch=partial(pad_input, pad_idx=-1), shuffle=True, num_workers=0)\n",
"batches = list(dl)\n",
"max_len = len(batches[0][0])\n",
"for b in batches: \n",
" assert(len(b[0])) <= max_len \n",
" test_ne(b[0][-1], -1)"
]
},
{
"cell_type": "markdown",
"id": "e0e3034f",
"metadata": {},
"source": [
"## TransformBlock for text"
]
},
{
"cell_type": "markdown",
"id": "f34c171f",
"metadata": {},
"source": [
"To use the data block API, you will need this build block for texts."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "99d34a59",
"metadata": {},
"outputs": [],
"source": [
"#| export\n",
"class TextBlock(TransformBlock):\n",
" \"A `TransformBlock` for texts\"\n",
" @delegates(Numericalize.__init__)\n",
" def __init__(self, tok_tfm, vocab=None, is_lm=False, seq_len=72, backwards=False, **kwargs):\n",
" type_tfms = [tok_tfm, Numericalize(vocab, **kwargs)]\n",
" if backwards: type_tfms += [reverse_text]\n",
" return super().__init__(type_tfms=type_tfms,\n",
" dl_type=LMDataLoader if is_lm else SortedDL,\n",
" dls_kwargs={'seq_len': seq_len} if is_lm else {'before_batch': Pad_Chunk(seq_len=seq_len)})\n",
"\n",
" @classmethod\n",
" @delegates(Tokenizer.from_df, keep=True)\n",
" def from_df(cls, text_cols, vocab=None, is_lm=False, seq_len=72, backwards=False, min_freq=3, max_vocab=60000, **kwargs):\n",
" \"Build a `TextBlock` from a dataframe using `text_cols`\"\n",
" return cls(Tokenizer.from_df(text_cols, **kwargs), vocab=vocab, is_lm=is_lm, seq_len=seq_len,\n",
" backwards=backwards, min_freq=min_freq, max_vocab=max_vocab)\n",
"\n",
" @classmethod\n",
" @delegates(Tokenizer.from_folder, keep=True)\n",
" def from_folder(cls, path, vocab=None, is_lm=False, seq_len=72, backwards=False, min_freq=3, max_vocab=60000, **kwargs):\n",
" \"Build a `TextBlock` from a `path`\"\n",
" return cls(Tokenizer.from_folder(path, **kwargs), vocab=vocab, is_lm=is_lm, seq_len=seq_len,\n",
" backwards=backwards, min_freq=min_freq, max_vocab=max_vocab)"
]
},
{
"cell_type": "markdown",
"id": "cf2bfd7b",
"metadata": {},
"source": [
"For efficient tokenization, you probably want to use one of the factory methods. Otherwise, you can pass your custom `tok_tfm` that will deal with tokenization (if your texts are already tokenized, you can pass `noop`), a `vocab`, or leave it to be inferred on the texts using `min_freq` and `max_vocab`.\n",
"\n",
"`is_lm` indicates if we want to use texts for language modeling or another task, `seq_len` is only necessary to tune if `is_lm=False`, and is passed along to `pad_input_chunk`."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c0e41104",
"metadata": {},
"outputs": [
{
"data": {
"text/markdown": [
"<h4 id=\"TextBlock.from_df\" class=\"doc_header\"><code>TextBlock.from_df</code><a href=\"\" class=\"source_link\" style=\"float:right\">[source]</a></h4>\n",
"\n",
"> <code>TextBlock.from_df</code>(**`text_cols`**, **`vocab`**=*`None`*, **`is_lm`**=*`False`*, **`seq_len`**=*`72`*, **`backwards`**=*`False`*, **`min_freq`**=*`3`*, **`max_vocab`**=*`60000`*, **`tok`**=*`None`*, **`rules`**=*`None`*, **`sep`**=*`' '`*, **`n_workers`**=*`8`*, **`mark_fields`**=*`None`*, **`tok_text_col`**=*`'text'`*, **\\*\\*`kwargs`**)\n",
"\n",
"Build a [`TextBlock`](/text.data.html#TextBlock) from a dataframe using `text_cols`"
],
"text/plain": [
"<IPython.core.display.Markdown object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"show_doc(TextBlock.from_df)"
]
},
{
"cell_type": "markdown",
"id": "2df33e1c",
"metadata": {},
"source": [
"Here is an example using a sample of IMDB stored as a CSV file:"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "cf103b68",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
"<style>\n",
" /* Turns off some styling */\n",
" progress {\n",
" /* gets rid of default border in Firefox and Opera. */\n",
" border: none;\n",
" /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
" background-size: auto;\n",
" }\n",
" .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
" background: #F44336;\n",
" }\n",
"</style>\n"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>text</th>\n",
" <th>category</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>xxbos xxmaj raising xxmaj victor xxmaj vargas : a xxmaj review \\n\\n xxmaj you know , xxmaj raising xxmaj victor xxmaj vargas is like sticking your hands into a big , xxunk bowl of xxunk . xxmaj it 's warm and gooey , but you 're not sure if it feels right . xxmaj try as i might , no matter how warm and gooey xxmaj raising xxmaj victor xxmaj vargas became i was always aware that something did n't quite feel right . xxmaj victor xxmaj vargas suffers from a certain xxunk on the director 's part . xxmaj apparently , the director thought that the ethnic backdrop of a xxmaj latino family on the lower east side , and an xxunk storyline would make the film critic proof . xxmaj he was right , but it did n't fool me . xxmaj raising xxmaj victor xxmaj vargas is</td>\n",
" <td>negative</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>xxbos xxup the xxup shop xxup around xxup the xxup corner is one of the xxunk and most feel - good romantic comedies ever made . xxmaj there 's just no getting around that , and it 's hard to actually put one 's feeling for this film into words . xxmaj it 's not one of those films that tries too hard , nor does it come up with the xxunk possible scenarios to get the two protagonists together in the end . xxmaj in fact , all its charm is xxunk , contained within the characters and the setting and the plot … which is highly believable to xxunk . xxmaj it 's easy to think that such a love story , as beautiful as any other ever told , * could * happen to you … a feeling you do n't often get from other romantic comedies</td>\n",
" <td>positive</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"path = untar_data(URLs.IMDB_SAMPLE)\n",
"df = pd.read_csv(path/'texts.csv')\n",
"\n",
"imdb_clas = DataBlock(\n",
" blocks=(TextBlock.from_df('text', seq_len=72), CategoryBlock),\n",
" get_x=ColReader('text'), get_y=ColReader('label'), splitter=ColSplitter())\n",
"\n",
"dls = imdb_clas.dataloaders(df, bs=64)\n",
"dls.show_batch(max_n=2)"
]
},
{
"cell_type": "markdown",
"id": "b45aa574",
"metadata": {},
"source": [
"`vocab`, `is_lm`, `seq_len`, `min_freq` and `max_vocab` are passed to the main init, the other argument to `Tokenizer.from_df`."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "882188c1",
"metadata": {},
"outputs": [
{
"data": {
"text/markdown": [
"<h4 id=\"TextBlock.from_folder\" class=\"doc_header\"><code>TextBlock.from_folder</code><a href=\"\" class=\"source_link\" style=\"float:right\">[source]</a></h4>\n",
"\n",
"> <code>TextBlock.from_folder</code>(**`path`**, **`vocab`**=*`None`*, **`is_lm`**=*`False`*, **`seq_len`**=*`72`*, **`backwards`**=*`False`*, **`min_freq`**=*`3`*, **`max_vocab`**=*`60000`*, **`tok`**=*`None`*, **`rules`**=*`None`*, **`extensions`**=*`None`*, **`folders`**=*`None`*, **`output_dir`**=*`None`*, **`skip_if_exists`**=*`True`*, **`output_names`**=*`None`*, **`n_workers`**=*`8`*, **`encoding`**=*`'utf8'`*, **\\*\\*`kwargs`**)\n",
"\n",
"Build a [`TextBlock`](/text.data.html#TextBlock) from a `path`"
],
"text/plain": [
"<IPython.core.display.Markdown object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"show_doc(TextBlock.from_folder)"
]
},
{
"cell_type": "markdown",
"id": "4c0bbe8f",
"metadata": {},
"source": [
"`vocab`, `is_lm`, `seq_len`, `min_freq` and `max_vocab` are passed to the main init, the other argument to `Tokenizer.from_folder`."
]
},
{
"cell_type": "markdown",
"id": "1e44e7c8",
"metadata": {},
"source": [
"## TextDataLoaders -"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "c8bda6d7",
"metadata": {},
"outputs": [],
"source": [
"#| export\n",
"class TextDataLoaders(DataLoaders):\n",
" \"Basic wrapper around several `DataLoader`s with factory methods for NLP problems\"\n",
" @classmethod\n",
" @delegates(DataLoaders.from_dblock)\n",
" def from_folder(cls, path, train='train', valid='valid', valid_pct=None, seed=None, vocab=None, text_vocab=None, is_lm=False,\n",
" tok_tfm=None, seq_len=72, splitter=None, backwards=False, **kwargs):\n",
" \"Create from imagenet style dataset in `path` with `train` and `valid` subfolders (or provide `valid_pct`)\"\n",
" if splitter is None:\n",
" splitter = GrandparentSplitter(train_name=train, valid_name=valid) if valid_pct is None else RandomSplitter(valid_pct, seed=seed)\n",
" blocks = [TextBlock.from_folder(path, text_vocab, is_lm, seq_len, backwards, tok=tok_tfm)]\n",
" if not is_lm: blocks.append(CategoryBlock(vocab=vocab))\n",
" get_items = partial(get_text_files, folders=[train,valid]) if valid_pct is None else get_text_files\n",
" dblock = DataBlock(blocks=blocks,\n",
" get_items=get_items,\n",
" splitter=splitter,\n",
" get_y=None if is_lm else parent_label)\n",
" return cls.from_dblock(dblock, path, path=path, seq_len=seq_len, **kwargs)\n",
"\n",
" @classmethod\n",
" @delegates(DataLoaders.from_dblock)\n",
" def from_df(cls, df, path='.', valid_pct=0.2, seed=None, text_col=0, label_col=1, label_delim=None, y_block=None,\n",
" text_vocab=None, is_lm=False, valid_col=None, tok_tfm=None, tok_text_col=\"text\", seq_len=72, backwards=False, **kwargs):\n",
" \"Create from `df` in `path` with `valid_pct`\"\n",
" blocks = [TextBlock.from_df(text_col, text_vocab, is_lm, seq_len, backwards, tok=tok_tfm)]\n",
" if y_block is None and not is_lm:\n",
" blocks.append(MultiCategoryBlock if is_listy(label_col) and len(label_col) > 1 else CategoryBlock)\n",
" if y_block is not None and not is_lm: blocks += (y_block if is_listy(y_block) else [y_block])\n",
" splitter = RandomSplitter(valid_pct, seed=seed) if valid_col is None else ColSplitter(valid_col)\n",
" dblock = DataBlock(blocks=blocks,\n",
" get_x=ColReader(tok_text_col),\n",
" get_y=None if is_lm else ColReader(label_col, label_delim=label_delim),\n",
" splitter=splitter)\n",
" return cls.from_dblock(dblock, df, path=path, seq_len=seq_len, **kwargs)\n",
"\n",
" @classmethod\n",
" def from_csv(cls, path, csv_fname='labels.csv', header='infer', delimiter=None, quoting=csv.QUOTE_MINIMAL, **kwargs):\n",
" \"Create from `csv` file in `path/csv_fname`\"\n",
" df = pd.read_csv(Path(path)/csv_fname, header=header, delimiter=delimiter, quoting=quoting)\n",
" return cls.from_df(df, path=path, **kwargs)\n",
"\n",
"TextDataLoaders.from_csv = delegates(to=TextDataLoaders.from_df)(TextDataLoaders.from_csv)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1e0aedcf",
"metadata": {},
"outputs": [
{
"data": {
"text/markdown": [
"---\n",
"\n",
"[source](https://github.com/fastai/fastai/blob/main/fastai/text/data.py#L248){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
"\n",
"## TextDataLoaders\n",
"\n",
"> TextDataLoaders (*loaders, path:str|pathlib.Path='.', device=None)\n",
"\n",
"*Basic wrapper around several `DataLoader`s with factory methods for NLP problems*\n",
"\n",
"| | **Type** | **Default** | **Details** |\n",
"| -- | -------- | ----------- | ----------- |\n",
"| loaders | VAR_POSITIONAL | | `DataLoader` objects to wrap |\n",
"| path | str \\| pathlib.Path | . | Path to store export objects |\n",
"| device | NoneType | None | Device to put `DataLoaders` |"
],
"text/plain": [
"---\n",
"\n",
"[source](https://github.com/fastai/fastai/blob/main/fastai/text/data.py#L248){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
"\n",
"## TextDataLoaders\n",
"\n",
"> TextDataLoaders (*loaders, path:str|pathlib.Path='.', device=None)\n",
"\n",
"*Basic wrapper around several `DataLoader`s with factory methods for NLP problems*\n",
"\n",
"| | **Type** | **Default** | **Details** |\n",
"| -- | -------- | ----------- | ----------- |\n",
"| loaders | VAR_POSITIONAL | | `DataLoader` objects to wrap |\n",
"| path | str \\| pathlib.Path | . | Path to store export objects |\n",
"| device | NoneType | None | Device to put `DataLoaders` |"
]
},
"execution_count": null,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"show_doc(TextDataLoaders, title_level=2)"
]
},
{
"cell_type": "markdown",
"id": "eed8a9ce",
"metadata": {},
"source": [
"You should not use the init directly but one of the following factory methods. All those factory methods accept as arguments:\n",
"\n",
"- `text_vocab`: the vocabulary used for numericalizing texts (if not passed, it's inferred from the data)\n",
"- `tok_tfm`: if passed, uses this `tok_tfm` instead of the default\n",
"- `seq_len`: the sequence length used for batch\n",
"- `bs`: the batch size\n",
"- `val_bs`: the batch size for the validation `DataLoader` (defaults to `bs`)\n",
"- `shuffle_train`: if we shuffle the training `DataLoader` or not\n",
"- `device`: the PyTorch device to use (defaults to `default_device()`)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "82c3a0c2",
"metadata": {},
"outputs": [
{
"data": {
"text/markdown": [
"---\n",
"\n",
"[source](https://github.com/fastai/fastai/blob/main/fastai/text/data.py#L252){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
"\n",
"### TextDataLoaders.from_folder\n",
"\n",
"> TextDataLoaders.from_folder (path, train='train', valid='valid',\n",
"> valid_pct=None, seed=None, vocab=None,\n",
"> text_vocab=None, is_lm=False, tok_tfm=None,\n",
"> seq_len=72, splitter=None, backwards=False,\n",
"> bs:int=64, val_bs:int=None,\n",
"> shuffle:bool=True, device=None)\n",
"\n",
"*Create from imagenet style dataset in `path` with `train` and `valid` subfolders (or provide `valid_pct`)*\n",
"\n",
"| | **Type** | **Default** | **Details** |\n",
"| -- | -------- | ----------- | ----------- |\n",
"| path | str \\| pathlib.Path | . | Path to put in `DataLoaders` |\n",
"| train | str | train | |\n",
"| valid | str | valid | |\n",
"| valid_pct | NoneType | None | |\n",
"| seed | NoneType | None | |\n",
"| vocab | NoneType | None | |\n",
"| text_vocab | NoneType | None | |\n",
"| is_lm | bool | False | |\n",
"| tok_tfm | NoneType | None | |\n",
"| seq_len | int | 72 | |\n",
"| splitter | NoneType | None | |\n",
"| backwards | bool | False | |\n",
"| bs | int | 64 | Size of batch |\n",
"| val_bs | int | None | Size of batch for validation `DataLoader` |\n",
"| shuffle | bool | True | Whether to shuffle data |\n",
"| device | NoneType | None | Device to put `DataLoaders` |"
],
"text/plain": [
"---\n",
"\n",
"[source](https://github.com/fastai/fastai/blob/main/fastai/text/data.py#L252){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
"\n",
"### TextDataLoaders.from_folder\n",
"\n",
"> TextDataLoaders.from_folder (path, train='train', valid='valid',\n",
"> valid_pct=None, seed=None, vocab=None,\n",
"> text_vocab=None, is_lm=False, tok_tfm=None,\n",
"> seq_len=72, splitter=None, backwards=False,\n",
"> bs:int=64, val_bs:int=None,\n",
"> shuffle:bool=True, device=None)\n",
"\n",
"*Create from imagenet style dataset in `path` with `train` and `valid` subfolders (or provide `valid_pct`)*\n",
"\n",
"| | **Type** | **Default** | **Details** |\n",
"| -- | -------- | ----------- | ----------- |\n",
"| path | str \\| pathlib.Path | . | Path to put in `DataLoaders` |\n",
"| train | str | train | |\n",
"| valid | str | valid | |\n",
"| valid_pct | NoneType | None | |\n",
"| seed | NoneType | None | |\n",
"| vocab | NoneType | None | |\n",
"| text_vocab | NoneType | None | |\n",
"| is_lm | bool | False | |\n",
"| tok_tfm | NoneType | None | |\n",
"| seq_len | int | 72 | |\n",
"| splitter | NoneType | None | |\n",
"| backwards | bool | False | |\n",
"| bs | int | 64 | Size of batch |\n",
"| val_bs | int | None | Size of batch for validation `DataLoader` |\n",
"| shuffle | bool | True | Whether to shuffle data |\n",
"| device | NoneType | None | Device to put `DataLoaders` |"
]
},
"execution_count": null,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"show_doc(TextDataLoaders.from_folder)"
]
},
{
"cell_type": "markdown",
"id": "a4625534",
"metadata": {},
"source": [
"If `valid_pct` is provided, a random split is performed (with an optional `seed`) by setting aside that percentage of the data for the validation set (instead of looking at the grandparents folder). If a `vocab` is passed, only the folders with names in `vocab` are kept.\n",
"\n",
"Here is an example on a sample of the IMDB movie review dataset:"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "745fb609",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>text</th>\n",
" <th>category</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>xxbos xxmaj match 1 : xxmaj tag xxmaj team xxmaj table xxmaj match xxmaj bubba xxmaj ray and xxmaj spike xxmaj dudley vs xxmaj eddie xxmaj guerrero and xxmaj chris xxmaj benoit xxmaj bubba xxmaj ray and xxmaj spike xxmaj dudley started things off with a xxmaj tag xxmaj team xxmaj table xxmaj match against xxmaj eddie xxmaj guerrero and xxmaj chris xxmaj benoit . xxmaj according to the rules of the match , both opponents have to go through tables in order to get the win . xxmaj benoit and xxmaj guerrero heated up early on by taking turns hammering first xxmaj spike and then xxmaj bubba xxmaj ray . a xxmaj german xxunk by xxmaj benoit to xxmaj bubba took the wind out of the xxmaj dudley brother . xxmaj spike tried to help his brother , but the referee restrained him while xxmaj benoit and xxmaj guerrero</td>\n",
" <td>pos</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>xxbos xxmaj okay , so xxmaj i 'm not a big video game buff , but was the game xxmaj house of the xxmaj dead really famous enough to make a movie from ? xxmaj sure , they went as far as to actually put in quick video game clips throughout the movie , as though justifying any particular scene of violence , but there are dozens and dozens of games that look exactly the same , with the hand in the bottom on the screen , supposedly your own , holding whatever weapon and goo - ing all kinds of aliens or walking dead or snipers or whatever the case may be . \\n\\n xxmaj it 's an interesting premise in xxmaj house of the xxmaj dead , with a lot of college kids ( loaded college kids , as it were , kids who are able to pay</td>\n",
" <td>neg</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>xxbos xxup anchors xxup aweigh sees two eager young sailors , xxmaj joe xxmaj brady ( gene xxmaj kelly ) and xxmaj clarence xxmaj doolittle / xxmaj brooklyn ( frank xxmaj sinatra ) , get a special four - day shore leave . xxmaj eager to get to the girls , particularly xxmaj joe 's xxmaj lola , neither xxmaj joe nor xxmaj brooklyn figure on the interruption of little xxmaj navy - mad xxmaj donald ( dean xxmaj stockwell ) and his xxmaj aunt xxmaj susie ( kathryn xxmaj grayson ) . xxmaj unexperienced in the ways of females and courting , xxmaj brooklyn quickly enlists xxmaj joe to help him win xxmaj aunt xxmaj susie over . xxmaj along the way , however , xxmaj joe finds himself falling for the gal he thinks belongs to his best friend . xxmaj how is xxmaj brooklyn going to take</td>\n",
" <td>pos</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"#| slow\n",
"path = untar_data(URLs.IMDB)\n",
"dls = TextDataLoaders.from_folder(path)\n",
"dls.show_batch(max_n=3)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f1338228",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>text</th>\n",
" <th>text_</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>xxbos xxmaj this film is a spicy little piece of film - making from xxmaj sam xxmaj fuller which gives xxmaj richard xxmaj widmark the chance to show of some of his best , most edgy acting in the role of xxmaj skip mccoy , a small - time thief who stumbles onto a military secret while picking beautiful xxmaj candy 's ( jean xxmaj peters ) pocket on a crowded bus</td>\n",
" <td>xxmaj this film is a spicy little piece of film - making from xxmaj sam xxmaj fuller which gives xxmaj richard xxmaj widmark the chance to show of some of his best , most edgy acting in the role of xxmaj skip mccoy , a small - time thief who stumbles onto a military secret while picking beautiful xxmaj candy 's ( jean xxmaj peters ) pocket on a crowded bus .</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>text</th>\n",
" <th>category</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>xxbos xxmaj match 1 : xxmaj tag xxmaj team xxmaj table xxmaj match xxmaj bubba xxmaj ray and xxmaj spike xxmaj dudley vs xxmaj eddie xxmaj guerrero and xxmaj chris xxmaj benoit xxmaj bubba xxmaj ray and xxmaj spike xxmaj dudley started things off with a xxmaj tag xxmaj team xxmaj table xxmaj match against xxmaj eddie xxmaj guerrero and xxmaj chris xxmaj benoit . xxmaj according to the rules of the match , both opponents have to go through tables in order to get the win . xxmaj benoit and xxmaj guerrero heated up early on by taking turns hammering first xxmaj spike and then xxmaj bubba xxmaj ray . a xxmaj german xxunk by xxmaj benoit to xxmaj bubba took the wind out of the xxmaj dudley brother . xxmaj spike tried to help his brother , but the referee restrained him while xxmaj benoit and xxmaj guerrero</td>\n",
" <td>pos</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"#| hide\n",
"# Test that `tok_tfm` parameter works\n",
"path = untar_data(URLs.IMDB)\n",
"tknzer = WordTokenizer()\n",
"dls = TextDataLoaders.from_folder(path, tok_tfm=tknzer, is_lm=True)\n",
"dls.show_batch(max_n=1)\n",
"dls = TextDataLoaders.from_folder(path, tok_tfm=tknzer, is_lm=False)\n",
"dls.show_batch(max_n=1)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "5c072481",
"metadata": {},
"outputs": [
{
"data": {
"text/markdown": [
"---\n",
"\n",
"[source](https://github.com/fastai/fastai/blob/main/fastai/text/data.py#L268){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
"\n",
"### TextDataLoaders.from_df\n",
"\n",
"> TextDataLoaders.from_df (df, path='.', valid_pct=0.2, seed=None,\n",
"> text_col=0, label_col=1, label_delim=None,\n",
"> y_block=None, text_vocab=None, is_lm=False,\n",
"> valid_col=None, tok_tfm=None,\n",
"> tok_text_col='text', seq_len=72,\n",
"> backwards=False, bs:int=64, val_bs:int=None,\n",
"> shuffle:bool=True, device=None)\n",
"\n",
"*Create from `df` in `path` with `valid_pct`*\n",
"\n",
"| | **Type** | **Default** | **Details** |\n",
"| -- | -------- | ----------- | ----------- |\n",
"| df | | | |\n",
"| path | str \\| pathlib.Path | . | Path to put in `DataLoaders` |\n",
"| valid_pct | float | 0.2 | |\n",
"| seed | NoneType | None | |\n",
"| text_col | int | 0 | |\n",
"| label_col | int | 1 | |\n",
"| label_delim | NoneType | None | |\n",
"| y_block | NoneType | None | |\n",
"| text_vocab | NoneType | None | |\n",
"| is_lm | bool | False | |\n",
"| valid_col | NoneType | None | |\n",
"| tok_tfm | NoneType | None | |\n",
"| tok_text_col | str | text | |\n",
"| seq_len | int | 72 | |\n",
"| backwards | bool | False | |\n",
"| bs | int | 64 | Size of batch |\n",
"| val_bs | int | None | Size of batch for validation `DataLoader` |\n",
"| shuffle | bool | True | Whether to shuffle data |\n",
"| device | NoneType | None | Device to put `DataLoaders` |"
],
"text/plain": [
"---\n",
"\n",
"[source](https://github.com/fastai/fastai/blob/main/fastai/text/data.py#L268){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
"\n",
"### TextDataLoaders.from_df\n",
"\n",
"> TextDataLoaders.from_df (df, path='.', valid_pct=0.2, seed=None,\n",
"> text_col=0, label_col=1, label_delim=None,\n",
"> y_block=None, text_vocab=None, is_lm=False,\n",
"> valid_col=None, tok_tfm=None,\n",
"> tok_text_col='text', seq_len=72,\n",
"> backwards=False, bs:int=64, val_bs:int=None,\n",
"> shuffle:bool=True, device=None)\n",
"\n",
"*Create from `df` in `path` with `valid_pct`*\n",
"\n",
"| | **Type** | **Default** | **Details** |\n",
"| -- | -------- | ----------- | ----------- |\n",
"| df | | | |\n",
"| path | str \\| pathlib.Path | . | Path to put in `DataLoaders` |\n",
"| valid_pct | float | 0.2 | |\n",
"| seed | NoneType | None | |\n",
"| text_col | int | 0 | |\n",
"| label_col | int | 1 | |\n",
"| label_delim | NoneType | None | |\n",
"| y_block | NoneType | None | |\n",
"| text_vocab | NoneType | None | |\n",
"| is_lm | bool | False | |\n",
"| valid_col | NoneType | None | |\n",
"| tok_tfm | NoneType | None | |\n",
"| tok_text_col | str | text | |\n",
"| seq_len | int | 72 | |\n",
"| backwards | bool | False | |\n",
"| bs | int | 64 | Size of batch |\n",
"| val_bs | int | None | Size of batch for validation `DataLoader` |\n",
"| shuffle | bool | True | Whether to shuffle data |\n",
"| device | NoneType | None | Device to put `DataLoaders` |"
]
},
"execution_count": null,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"show_doc(TextDataLoaders.from_df)"
]
},
{
"cell_type": "markdown",
"id": "cea209ca",
"metadata": {},
"source": [
"`seed` can optionally be passed for reproducibility. `text_col`, `label_col` and optionally `valid_col` are indices or names of columns for texts/labels and the validation flag. `label_delim` can be passed for a multi-label problem if your labels are in one column, separated by a particular char. `y_block` should be passed to indicate your type of targets, in case the library did no infer it properly.\n",
"\n",
"Along with this, you can specify the specific column the tokenized text are sent to with `tok_text_col`. By default they are stored in a column named `text` after tokenizing. \n",
"\n",
"Here are examples on subsets of IMDB:"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b85e70c0",
"metadata": {},
"outputs": [],
"source": [
"path = untar_data(URLs.IMDB_SAMPLE)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "392127ca",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>label</th>\n",
" <th>text</th>\n",
" <th>is_valid</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>negative</td>\n",
" <td>Un-bleeping-believable! Meg Ryan doesn't even look her usual pert lovable self in this, which normally makes me forgive her shallow ticky acting schtick. Hard to believe she was the producer on this dog. Plus Kevin Kline: what kind of suicide trip has his career been on? Whoosh... Banzai!!! Finally this was directed by the guy who did Big Chill? Must be a replay of Jonestown - hollywood style. Wooofff!</td>\n",
" <td>False</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>positive</td>\n",
" <td>This is a extremely well-made film. The acting, script and camera-work are all first-rate. The music is good, too, though it is mostly early in the film, when things are still relatively cheery. There are no really superstars in the cast, though several faces will be familiar. The entire cast does an excellent job with the script.&lt;br /&gt;&lt;br /&gt;But it is hard to watch, because there is no good end to a situation like the one presented. It is now fashionable to blame the British for setting Hindus and Muslims against each other, and then cruelly separating them into two countries. There is som...</td>\n",
" <td>False</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>negative</td>\n",
" <td>Every once in a long while a movie will come along that will be so awful that I feel compelled to warn people. If I labor all my days and I can save but one soul from watching this movie, how great will be my joy.&lt;br /&gt;&lt;br /&gt;Where to begin my discussion of pain. For starters, there was a musical montage every five minutes. There was no character development. Every character was a stereotype. We had swearing guy, fat guy who eats donuts, goofy foreign guy, etc. The script felt as if it were being written as the movie was being shot. The production value was so incredibly low that it felt li...</td>\n",
" <td>False</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>positive</td>\n",
" <td>Name just says it all. I watched this movie with my dad when it came out and having served in Korea he had great admiration for the man. The disappointing thing about this film is that it only concentrate on a short period of the man's life - interestingly enough the man's entire life would have made such an epic bio-pic that it is staggering to imagine the cost for production.&lt;br /&gt;&lt;br /&gt;Some posters elude to the flawed characteristics about the man, which are cheap shots. The theme of the movie \"Duty, Honor, Country\" are not just mere words blathered from the lips of a high-brassed offic...</td>\n",
" <td>False</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>negative</td>\n",
" <td>This movie succeeds at being one of the most unique movies you've seen. However this comes from the fact that you can't make heads or tails of this mess. It almost seems as a series of challenges set up to determine whether or not you are willing to walk out of the movie and give up the money you just paid. If you don't want to feel slighted you'll sit through this horrible film and develop a real sense of pity for the actors involved, they've all seen better days, but then you realize they actually got paid quite a bit of money to do this and you'll lose pity for them just like you've alr...</td>\n",
" <td>False</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"</div>"
],
"text/plain": [
" label \\\n",
"0 negative \n",
"1 positive \n",
"2 negative \n",
"3 positive \n",
"4 negative \n",
"\n",
" text \\\n",
"0 Un-bleeping-believable! Meg Ryan doesn't even look her usual pert lovable self in this, which normally makes me forgive her shallow ticky acting schtick. Hard to believe she was the producer on this dog. Plus Kevin Kline: what kind of suicide trip has his career been on? Whoosh... Banzai!!! Finally this was directed by the guy who did Big Chill? Must be a replay of Jonestown - hollywood style. Wooofff! \n",
"1 This is a extremely well-made film. The acting, script and camera-work are all first-rate. The music is good, too, though it is mostly early in the film, when things are still relatively cheery. There are no really superstars in the cast, though several faces will be familiar. The entire cast does an excellent job with the script.<br /><br />But it is hard to watch, because there is no good end to a situation like the one presented. It is now fashionable to blame the British for setting Hindus and Muslims against each other, and then cruelly separating them into two countries. There is som... \n",
"2 Every once in a long while a movie will come along that will be so awful that I feel compelled to warn people. If I labor all my days and I can save but one soul from watching this movie, how great will be my joy.<br /><br />Where to begin my discussion of pain. For starters, there was a musical montage every five minutes. There was no character development. Every character was a stereotype. We had swearing guy, fat guy who eats donuts, goofy foreign guy, etc. The script felt as if it were being written as the movie was being shot. The production value was so incredibly low that it felt li... \n",
"3 Name just says it all. I watched this movie with my dad when it came out and having served in Korea he had great admiration for the man. The disappointing thing about this film is that it only concentrate on a short period of the man's life - interestingly enough the man's entire life would have made such an epic bio-pic that it is staggering to imagine the cost for production.<br /><br />Some posters elude to the flawed characteristics about the man, which are cheap shots. The theme of the movie \"Duty, Honor, Country\" are not just mere words blathered from the lips of a high-brassed offic... \n",
"4 This movie succeeds at being one of the most unique movies you've seen. However this comes from the fact that you can't make heads or tails of this mess. It almost seems as a series of challenges set up to determine whether or not you are willing to walk out of the movie and give up the money you just paid. If you don't want to feel slighted you'll sit through this horrible film and develop a real sense of pity for the actors involved, they've all seen better days, but then you realize they actually got paid quite a bit of money to do this and you'll lose pity for them just like you've alr... \n",
"\n",
" is_valid \n",
"0 False \n",
"1 False \n",
"2 False \n",
"3 False \n",
"4 False "
]
},
"execution_count": null,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"df = pd.read_csv(path/\"texts.csv\"); df.head()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b02677ec",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
"<style>\n",
" /* Turns off some styling */\n",
" progress {\n",
" /* gets rid of default border in Firefox and Opera. */\n",
" border: none;\n",
" /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
" background-size: auto;\n",
" }\n",
" progress:not([value]), progress:not([value])::-webkit-progress-bar {\n",
" background: repeating-linear-gradient(45deg, #7e7e7e, #7e7e7e 10px, #5c5c5c 10px, #5c5c5c 20px);\n",
" }\n",
" .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
" background: #F44336;\n",
" }\n",
"</style>\n"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"#| hide\n",
"path = untar_data(URLs.IMDB_SAMPLE)\n",
"df = pd.read_csv(path/\"texts.csv\")\n",
"df.columns = ['label', 'text_col', 'is_valid'] # to test tok_text_col is working properly\n",
"dls = TextDataLoaders.from_df(df, path=path, text_col='text_col', label_col='label', valid_col='is_valid')\n",
"dl = dls.test_dl([\"This movie was bad\"])\n",
"x, = dl.one_batch()\n",
"test_eq(x.cpu(), TensorText([[2,8,21,29,25,97]]))"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "d85a3f83",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
"<style>\n",
" /* Turns off some styling */\n",
" progress {\n",
" /* gets rid of default border in Firefox and Opera. */\n",
" border: none;\n",
" /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
" background-size: auto;\n",
" }\n",
" .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
" background: #F44336;\n",
" }\n",
"</style>\n"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>text</th>\n",
" <th>category</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>xxbos xxmaj raising xxmaj victor xxmaj vargas : a xxmaj review \\n\\n xxmaj you know , xxmaj raising xxmaj victor xxmaj vargas is like sticking your hands into a big , xxunk bowl of xxunk . xxmaj it 's warm and gooey , but you 're not sure if it feels right . xxmaj try as i might , no matter how warm and gooey xxmaj raising xxmaj victor xxmaj vargas became i was always aware that something did n't quite feel right . xxmaj victor xxmaj vargas suffers from a certain xxunk on the director 's part . xxmaj apparently , the director thought that the ethnic backdrop of a xxmaj latino family on the lower east side , and an xxunk storyline would make the film critic proof . xxmaj he was right , but it did n't fool me . xxmaj raising xxmaj victor xxmaj vargas is</td>\n",
" <td>negative</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>xxbos xxup the xxup shop xxup around xxup the xxup corner is one of the xxunk and most feel - good romantic comedies ever made . xxmaj there 's just no getting around that , and it 's hard to actually put one 's feeling for this film into words . xxmaj it 's not one of those films that tries too hard , nor does it come up with the xxunk possible scenarios to get the two protagonists together in the end . xxmaj in fact , all its charm is xxunk , contained within the characters and the setting and the plot … which is highly believable to xxunk . xxmaj it 's easy to think that such a love story , as beautiful as any other ever told , * could * happen to you … a feeling you do n't often get from other romantic comedies</td>\n",
" <td>positive</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>xxbos xxmaj now that xxmaj che(2008 ) has finished its relatively short xxmaj australian cinema run ( extremely limited xxunk screen in xxmaj xxunk , after xxunk ) , i can xxunk join both xxunk of \" at xxmaj the xxmaj movies \" in taking xxmaj steven xxmaj soderbergh to task . \\n\\n xxmaj it 's usually satisfying to watch a film director change his style / subject , but xxmaj soderbergh 's most recent stinker , xxmaj the xxmaj girlfriend xxmaj xxunk ) , was also missing a story , so narrative ( and editing ? ) seem to suddenly be xxmaj soderbergh 's main challenge . xxmaj strange , after 20 - odd years in the business . xxmaj he was probably never much good at narrative , just xxunk it well inside \" edgy \" projects . \\n\\n xxmaj none of this excuses him this present ,</td>\n",
" <td>negative</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"path = untar_data(URLs.IMDB_SAMPLE)\n",
"df = pd.read_csv(path/\"texts.csv\")\n",
"dls = TextDataLoaders.from_df(df, path=path, text_col='text', label_col='label', valid_col='is_valid')\n",
"dls.show_batch(max_n=3)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "974c56a0",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
"<style>\n",
" /* Turns off some styling */\n",
" progress {\n",
" /* gets rid of default border in Firefox and Opera. */\n",
" border: none;\n",
" /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
" background-size: auto;\n",
" }\n",
" .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
" background: #F44336;\n",
" }\n",
"</style>\n"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>text</th>\n",
" <th>category</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>xxbos xxmaj raising xxmaj victor xxmaj vargas : a xxmaj review \\n\\n xxmaj you know , xxmaj raising xxmaj victor xxmaj vargas is like sticking your hands into a big , xxunk bowl of xxunk . xxmaj it 's warm and gooey , but you 're not sure if it feels right . xxmaj try as i might , no matter how warm and gooey xxmaj raising xxmaj victor xxmaj vargas became i was always aware that something did n't quite feel right . xxmaj victor xxmaj vargas suffers from a certain xxunk on the director 's part . xxmaj apparently , the director thought that the ethnic backdrop of a xxmaj latino family on the lower east side , and an xxunk storyline would make the film critic proof . xxmaj he was right , but it did n't fool me . xxmaj raising xxmaj victor xxmaj vargas is</td>\n",
" <td>negative</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
"\n",
"<style>\n",
" /* Turns off some styling */\n",
" progress {\n",
" /* gets rid of default border in Firefox and Opera. */\n",
" border: none;\n",
" /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
" background-size: auto;\n",
" }\n",
" .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
" background: #F44336;\n",
" }\n",
"</style>\n"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>text</th>\n",
" <th>text_</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>xxbos xxmaj this may have been made for the hell of it , but it was most probably the worst film i 've seen in years , xxmaj the best thing about the entire xxup dvd would be the case xxrep 3 ! xxmaj i 'm surprised that people took the time to make something so rubbish and yet spend money on it too , xxmaj i 'm glad i only rented</td>\n",
" <td>xxmaj this may have been made for the hell of it , but it was most probably the worst film i 've seen in years , xxmaj the best thing about the entire xxup dvd would be the case xxrep 3 ! xxmaj i 'm surprised that people took the time to make something so rubbish and yet spend money on it too , xxmaj i 'm glad i only rented .</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"#| hide\n",
"# Test that \"tok_tfm\" parameter works in `from_df`\n",
"tknzer = WordTokenizer()\n",
"dls = TextDataLoaders.from_df(df, path=path, is_lm=False, tok_tfm=tknzer,\n",
" text_col='text', label_col='label', valid_col='is_valid'\n",
" )\n",
"dls.show_batch(max_n=1)\n",
"dls = TextDataLoaders.from_df(df, path=path, is_lm=True, tok_tfm=tknzer,\n",
" text_col='text', label_col='label', valid_col='is_valid'\n",
" )\n",
"dls.show_batch(max_n=1)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "81bb738a",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
"<style>\n",
" /* Turns off some styling */\n",
" progress {\n",
" /* gets rid of default border in Firefox and Opera. */\n",
" border: none;\n",
" /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
" background-size: auto;\n",
" }\n",
" .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
" background: #F44336;\n",
" }\n",
"</style>\n"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>text</th>\n",
" <th>text_</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>xxbos xxmaj critics need to review what they class as a quality movie . i think the critics have seen too many actions films and have xxunk to the xxmaj matrix style of films . xxmaj xxunk is a breath of fresh air , a film with so many layers that one viewing is not enough to understand or appreciate this outstanding film . xxmaj xxunk von xxmaj xxunk shows that old</td>\n",
" <td>xxmaj critics need to review what they class as a quality movie . i think the critics have seen too many actions films and have xxunk to the xxmaj matrix style of films . xxmaj xxunk is a breath of fresh air , a film with so many layers that one viewing is not enough to understand or appreciate this outstanding film . xxmaj xxunk von xxmaj xxunk shows that old styles</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>xxmaj xxunk is something ) , but noticeable moments of xxunk as he still struggles to find his humanity . xxmaj this xxunk of his for a real life could get boring , and almost did in xxmaj supremacy , but just works better in xxmaj ultimatum ( better script ) . \\n\\n i am reminded of a scene in \" xxunk \" ( the only good xxmaj pierce xxmaj xxunk xxmaj</td>\n",
" <td>xxunk is something ) , but noticeable moments of xxunk as he still struggles to find his humanity . xxmaj this xxunk of his for a real life could get boring , and almost did in xxmaj supremacy , but just works better in xxmaj ultimatum ( better script ) . \\n\\n i am reminded of a scene in \" xxunk \" ( the only good xxmaj pierce xxmaj xxunk xxmaj bond</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>xxmaj mr . xxmaj julia , played his role equally as perfect . xxmaj it was interesting to see how reluctant xxmaj richard xxmaj dreyfuss was in replacing the dictator against his will . xxmaj but he became more confident and comfortable with the role as time passed . xxmaj since everything happens for a reason in life , i believe he was forced to replace the dictator because he was meant</td>\n",
" <td>mr . xxmaj julia , played his role equally as perfect . xxmaj it was interesting to see how reluctant xxmaj richard xxmaj dreyfuss was in replacing the dictator against his will . xxmaj but he became more confident and comfortable with the role as time passed . xxmaj since everything happens for a reason in life , i believe he was forced to replace the dictator because he was meant to</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"dls = TextDataLoaders.from_df(df, path=path, text_col='text', is_lm=True, valid_col='is_valid')\n",
"dls.show_batch(max_n=3)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "2dd59a8d",
"metadata": {},
"outputs": [
{
"data": {
"text/markdown": [
"---\n",
"\n",
"[source](https://github.com/fastai/fastai/blob/main/fastai/text/data.py#L283){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
"\n",
"### TextDataLoaders.from_csv\n",
"\n",
"> TextDataLoaders.from_csv (path, csv_fname='labels.csv', header='infer',\n",
"> delimiter=None, quoting=0, valid_pct=0.2,\n",
"> seed=None, text_col=0, label_col=1,\n",
"> label_delim=None, y_block=None,\n",
"> text_vocab=None, is_lm=False, valid_col=None,\n",
"> tok_tfm=None, tok_text_col='text', seq_len=72,\n",
"> backwards=False, bs:int=64, val_bs:int=None,\n",
"> shuffle:bool=True, device=None)\n",
"\n",
"*Create from `csv` file in `path/csv_fname`*\n",
"\n",
"| | **Type** | **Default** | **Details** |\n",
"| -- | -------- | ----------- | ----------- |\n",
"| path | str \\| pathlib.Path | . | Path to put in `DataLoaders` |\n",
"| csv_fname | str | labels.csv | |\n",
"| header | str | infer | |\n",
"| delimiter | NoneType | None | |\n",
"| quoting | int | 0 | |\n",
"| valid_pct | float | 0.2 | |\n",
"| seed | NoneType | None | |\n",
"| text_col | int | 0 | |\n",
"| label_col | int | 1 | |\n",
"| label_delim | NoneType | None | |\n",
"| y_block | NoneType | None | |\n",
"| text_vocab | NoneType | None | |\n",
"| is_lm | bool | False | |\n",
"| valid_col | NoneType | None | |\n",
"| tok_tfm | NoneType | None | |\n",
"| tok_text_col | str | text | |\n",
"| seq_len | int | 72 | |\n",
"| backwards | bool | False | |\n",
"| bs | int | 64 | Size of batch |\n",
"| val_bs | int | None | Size of batch for validation `DataLoader` |\n",
"| shuffle | bool | True | Whether to shuffle data |\n",
"| device | NoneType | None | Device to put `DataLoaders` |"
],
"text/plain": [
"---\n",
"\n",
"[source](https://github.com/fastai/fastai/blob/main/fastai/text/data.py#L283){target=\"_blank\" style=\"float:right; font-size:smaller\"}\n",
"\n",
"### TextDataLoaders.from_csv\n",
"\n",
"> TextDataLoaders.from_csv (path, csv_fname='labels.csv', header='infer',\n",
"> delimiter=None, quoting=0, valid_pct=0.2,\n",
"> seed=None, text_col=0, label_col=1,\n",
"> label_delim=None, y_block=None,\n",
"> text_vocab=None, is_lm=False, valid_col=None,\n",
"> tok_tfm=None, tok_text_col='text', seq_len=72,\n",
"> backwards=False, bs:int=64, val_bs:int=None,\n",
"> shuffle:bool=True, device=None)\n",
"\n",
"*Create from `csv` file in `path/csv_fname`*\n",
"\n",
"| | **Type** | **Default** | **Details** |\n",
"| -- | -------- | ----------- | ----------- |\n",
"| path | str \\| pathlib.Path | . | Path to put in `DataLoaders` |\n",
"| csv_fname | str | labels.csv | |\n",
"| header | str | infer | |\n",
"| delimiter | NoneType | None | |\n",
"| quoting | int | 0 | |\n",
"| valid_pct | float | 0.2 | |\n",
"| seed | NoneType | None | |\n",
"| text_col | int | 0 | |\n",
"| label_col | int | 1 | |\n",
"| label_delim | NoneType | None | |\n",
"| y_block | NoneType | None | |\n",
"| text_vocab | NoneType | None | |\n",
"| is_lm | bool | False | |\n",
"| valid_col | NoneType | None | |\n",
"| tok_tfm | NoneType | None | |\n",
"| tok_text_col | str | text | |\n",
"| seq_len | int | 72 | |\n",
"| backwards | bool | False | |\n",
"| bs | int | 64 | Size of batch |\n",
"| val_bs | int | None | Size of batch for validation `DataLoader` |\n",
"| shuffle | bool | True | Whether to shuffle data |\n",
"| device | NoneType | None | Device to put `DataLoaders` |"
]
},
"execution_count": null,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"show_doc(TextDataLoaders.from_csv)"
]
},
{
"cell_type": "markdown",
"id": "581eadee",
"metadata": {},
"source": [
"Opens the csv file with `header` and `delimiter`, then pass all the other arguments to `TextDataLoaders.from_df`."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "ee588eb2",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
"<style>\n",
" /* Turns off some styling */\n",
" progress {\n",
" /* gets rid of default border in Firefox and Opera. */\n",
" border: none;\n",
" /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
" background-size: auto;\n",
" }\n",
" .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
" background: #F44336;\n",
" }\n",
"</style>\n"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>text</th>\n",
" <th>category</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>xxbos xxmaj raising xxmaj victor xxmaj vargas : a xxmaj review \\n\\n xxmaj you know , xxmaj raising xxmaj victor xxmaj vargas is like sticking your hands into a big , xxunk bowl of xxunk . xxmaj it 's warm and gooey , but you 're not sure if it feels right . xxmaj try as i might , no matter how warm and gooey xxmaj raising xxmaj victor xxmaj vargas became i was always aware that something did n't quite feel right . xxmaj victor xxmaj vargas suffers from a certain xxunk on the director 's part . xxmaj apparently , the director thought that the ethnic backdrop of a xxmaj latino family on the lower east side , and an xxunk storyline would make the film critic proof . xxmaj he was right , but it did n't fool me . xxmaj raising xxmaj victor xxmaj vargas is</td>\n",
" <td>negative</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>xxbos xxup the xxup shop xxup around xxup the xxup corner is one of the xxunk and most feel - good romantic comedies ever made . xxmaj there 's just no getting around that , and it 's hard to actually put one 's feeling for this film into words . xxmaj it 's not one of those films that tries too hard , nor does it come up with the xxunk possible scenarios to get the two protagonists together in the end . xxmaj in fact , all its charm is xxunk , contained within the characters and the setting and the plot … which is highly believable to xxunk . xxmaj it 's easy to think that such a love story , as beautiful as any other ever told , * could * happen to you … a feeling you do n't often get from other romantic comedies</td>\n",
" <td>positive</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>xxbos xxmaj now that xxmaj che(2008 ) has finished its relatively short xxmaj australian cinema run ( extremely limited xxunk screen in xxmaj xxunk , after xxunk ) , i can xxunk join both xxunk of \" at xxmaj the xxmaj movies \" in taking xxmaj steven xxmaj soderbergh to task . \\n\\n xxmaj it 's usually satisfying to watch a film director change his style / subject , but xxmaj soderbergh 's most recent stinker , xxmaj the xxmaj girlfriend xxmaj xxunk ) , was also missing a story , so narrative ( and editing ? ) seem to suddenly be xxmaj soderbergh 's main challenge . xxmaj strange , after 20 - odd years in the business . xxmaj he was probably never much good at narrative , just xxunk it well inside \" edgy \" projects . \\n\\n xxmaj none of this excuses him this present ,</td>\n",
" <td>negative</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"dls = TextDataLoaders.from_csv(path=path, csv_fname='texts.csv', text_col='text', label_col='label', valid_col='is_valid')\n",
"dls.show_batch(max_n=3)"
]
},
{
"cell_type": "markdown",
"id": "a0583e22",
"metadata": {},
"source": [
"## Export -"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "a03a489e",
"metadata": {},
"outputs": [],
"source": [
"#| hide\n",
"from nbdev import nbdev_export\n",
"nbdev_export()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b0b9c312",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"jupytext": {
"split_at_heading": true
},
"kernelspec": {
"display_name": "python3",
"language": "python",
"name": "python3"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
Reference in New Issue View Git Blame Copy Permalink