wehub-resource-sync
1666 lines
97 KiB
Plaintext
1666 lines
97 KiB
Plaintext
{
|
|
"cells": [
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {
|
|
"colab_type": "text",
|
|
"id": "11xi8CRmVA1d"
|
|
},
|
|
"source": [
|
|
"Deep Learning Models -- A collection of various deep learning architectures, models, and tips for TensorFlow and PyTorch in Jupyter Notebooks.\n",
|
|
"- Author: Sebastian Raschka\n",
|
|
"- GitHub Repository: https://github.com/rasbt/deeplearning-models"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 1,
|
|
"metadata": {
|
|
"colab": {
|
|
"autoexec": {
|
|
"startup": false,
|
|
"wait_interval": 0
|
|
},
|
|
"base_uri": "https://localhost:8080/",
|
|
"height": 119
|
|
},
|
|
"colab_type": "code",
|
|
"executionInfo": {
|
|
"elapsed": 2889,
|
|
"status": "ok",
|
|
"timestamp": 1525034072517,
|
|
"user": {
|
|
"displayName": "Sebastian Raschka",
|
|
"photoUrl": "//lh6.googleusercontent.com/-cxK6yOSQ6uE/AAAAAAAAAAI/AAAAAAAAIfw/P9ar_CHsKOQ/s50-c-k-no/photo.jpg",
|
|
"userId": "118404394130788869227"
|
|
},
|
|
"user_tz": 240
|
|
},
|
|
"id": "WuXDfh6UVA1g",
|
|
"outputId": "80c92b95-76aa-444e-9aa3-9bcf5d000a6e"
|
|
},
|
|
"outputs": [
|
|
{
|
|
"name": "stdout",
|
|
"output_type": "stream",
|
|
"text": [
|
|
"Sebastian Raschka \n",
|
|
"\n",
|
|
"CPython 3.6.8\n",
|
|
"IPython 7.2.0\n",
|
|
"\n",
|
|
"torch 1.0.0\n"
|
|
]
|
|
}
|
|
],
|
|
"source": [
|
|
"%load_ext watermark\n",
|
|
"%watermark -a 'Sebastian Raschka' -v -p torch"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {
|
|
"colab_type": "text",
|
|
"id": "Cii2luqnVA1s"
|
|
},
|
|
"source": [
|
|
"- Runs on CPU (not recommended here) or GPU (if available)"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {
|
|
"colab_type": "text",
|
|
"id": "EYAtjwgyVA1t"
|
|
},
|
|
"source": [
|
|
"# Model Zoo -- Convolutional Autoencoder with Nearest-neighbor Interpolation (Trained on 10 categories of the Quickdraw dataset)"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {
|
|
"colab_type": "text",
|
|
"id": "Ke9a_LDUVA1v"
|
|
},
|
|
"source": [
|
|
"A convolutional autoencoder using nearest neighbor upscaling layers that compresses 768-pixel Quickdraw images down to a 7x7x8 (392 pixel) representation."
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {
|
|
"colab_type": "text",
|
|
"id": "1iZAQwueVA1x"
|
|
},
|
|
"source": [
|
|
"## Imports"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 2,
|
|
"metadata": {
|
|
"colab": {
|
|
"autoexec": {
|
|
"startup": false,
|
|
"wait_interval": 0
|
|
}
|
|
},
|
|
"colab_type": "code",
|
|
"id": "CO_0yUH6VA1z"
|
|
},
|
|
"outputs": [],
|
|
"source": [
|
|
"import os\n",
|
|
"import time\n",
|
|
"\n",
|
|
"import numpy as np\n",
|
|
"import pandas as pd\n",
|
|
"from PIL import Image\n",
|
|
"import matplotlib.pyplot as plt\n",
|
|
"\n",
|
|
"import torch\n",
|
|
"import torch.nn as nn\n",
|
|
"import torch.nn.functional as F\n",
|
|
"\n",
|
|
"from torch.utils.data import Dataset\n",
|
|
"from torch.utils.data import DataLoader\n",
|
|
"from torchvision import transforms\n",
|
|
"\n",
|
|
"\n",
|
|
"if torch.cuda.is_available():\n",
|
|
" torch.backends.cudnn.deterministic = True"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {
|
|
"colab_type": "text",
|
|
"id": "uhGdecraVA1-"
|
|
},
|
|
"source": [
|
|
"## Dataset"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {},
|
|
"source": [
|
|
"This notebook is based on Google's Quickdraw dataset (https://quickdraw.withgoogle.com). In particular we will be working with an arbitrary subset of 10 categories in png format:\n",
|
|
"\n",
|
|
" label_dict = {\n",
|
|
" \"lollipop\": 0,\n",
|
|
" \"binoculars\": 1,\n",
|
|
" \"mouse\": 2,\n",
|
|
" \"basket\": 3,\n",
|
|
" \"penguin\": 4,\n",
|
|
" \"washing machine\": 5,\n",
|
|
" \"canoe\": 6,\n",
|
|
" \"eyeglasses\": 7,\n",
|
|
" \"beach\": 8,\n",
|
|
" \"screwdriver\": 9,\n",
|
|
" }\n",
|
|
" \n",
|
|
"(The class labels 0-9 can be ignored in this notebook). \n",
|
|
"\n",
|
|
"For more details on obtaining and preparing the dataset, please see the\n",
|
|
"\n",
|
|
"- [custom-data-loader-quickdraw.ipynb](custom-data-loader-quickdraw.ipynb)\n",
|
|
"\n",
|
|
"notebook."
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 3,
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"name": "stdout",
|
|
"output_type": "stream",
|
|
"text": [
|
|
"(28, 28)\n"
|
|
]
|
|
},
|
|
{
|
|
"data": {
|
|
"image/png": "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\n",
|
|
"text/plain": [
|
|
"<Figure size 432x288 with 1 Axes>"
|
|
]
|
|
},
|
|
"metadata": {
|
|
"needs_background": "light"
|
|
},
|
|
"output_type": "display_data"
|
|
}
|
|
],
|
|
"source": [
|
|
"df = pd.read_csv('quickdraw_png_set1_train.csv', index_col=0)\n",
|
|
"df.head()\n",
|
|
"\n",
|
|
"main_dir = 'quickdraw-png_set1/'\n",
|
|
"\n",
|
|
"img = Image.open(os.path.join(main_dir, df.index[99]))\n",
|
|
"img = np.asarray(img, dtype=np.uint8)\n",
|
|
"print(img.shape)\n",
|
|
"plt.imshow(np.array(img), cmap='binary')\n",
|
|
"plt.show()"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {
|
|
"colab_type": "text",
|
|
"id": "Z1zF9mPwVA2P"
|
|
},
|
|
"source": [
|
|
"### Create a Custom Data Loader"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 4,
|
|
"metadata": {
|
|
"colab": {
|
|
"autoexec": {
|
|
"startup": false,
|
|
"wait_interval": 0
|
|
}
|
|
},
|
|
"colab_type": "code",
|
|
"id": "fV2cRUDx7mqz"
|
|
},
|
|
"outputs": [],
|
|
"source": [
|
|
"class QuickdrawDataset(Dataset):\n",
|
|
" \"\"\"Custom Dataset for loading Quickdraw images\"\"\"\n",
|
|
"\n",
|
|
" def __init__(self, txt_path, img_dir, transform=None):\n",
|
|
" \n",
|
|
" df = pd.read_csv(txt_path, sep=\",\", index_col=0)\n",
|
|
" self.img_dir = img_dir\n",
|
|
" self.txt_path = txt_path\n",
|
|
" self.img_names = df.index.values\n",
|
|
" self.y = df['Label'].values\n",
|
|
" self.transform = transform\n",
|
|
"\n",
|
|
" def __getitem__(self, index):\n",
|
|
" img = Image.open(os.path.join(self.img_dir,\n",
|
|
" self.img_names[index]))\n",
|
|
" \n",
|
|
" if self.transform is not None:\n",
|
|
" img = self.transform(img)\n",
|
|
" \n",
|
|
" label = self.y[index]\n",
|
|
" return img, label\n",
|
|
"\n",
|
|
" def __len__(self):\n",
|
|
" return self.y.shape[0]"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 5,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"# Note that transforms.ToTensor()\n",
|
|
"# already divides pixels by 255. internally\n",
|
|
"\n",
|
|
"\n",
|
|
"BATCH_SIZE = 128\n",
|
|
"\n",
|
|
"custom_transform = transforms.Compose([#transforms.Lambda(lambda x: x/255.),\n",
|
|
" transforms.ToTensor()])\n",
|
|
"\n",
|
|
"train_dataset = QuickdrawDataset(txt_path='quickdraw_png_set1_train.csv',\n",
|
|
" img_dir='quickdraw-png_set1/',\n",
|
|
" transform=custom_transform)\n",
|
|
"\n",
|
|
"train_loader = DataLoader(dataset=train_dataset,\n",
|
|
" batch_size=BATCH_SIZE,\n",
|
|
" shuffle=True,\n",
|
|
" num_workers=4) \n"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 6,
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"name": "stdout",
|
|
"output_type": "stream",
|
|
"text": [
|
|
"Epoch: 1 | Batch index: 0 | Batch size: 128\n",
|
|
"Epoch: 2 | Batch index: 0 | Batch size: 128\n"
|
|
]
|
|
}
|
|
],
|
|
"source": [
|
|
"device = torch.device(\"cuda:3\" if torch.cuda.is_available() else \"cpu\")\n",
|
|
"torch.manual_seed(0)\n",
|
|
"\n",
|
|
"num_epochs = 2\n",
|
|
"for epoch in range(num_epochs):\n",
|
|
"\n",
|
|
" for batch_idx, (x, y) in enumerate(train_loader):\n",
|
|
" \n",
|
|
" print('Epoch:', epoch+1, end='')\n",
|
|
" print(' | Batch index:', batch_idx, end='')\n",
|
|
" print(' | Batch size:', y.size()[0])\n",
|
|
" \n",
|
|
" x = x.to(device)\n",
|
|
" y = y.to(device)\n",
|
|
" break"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {
|
|
"colab_type": "text",
|
|
"id": "wHRUkZFFVA2T"
|
|
},
|
|
"source": [
|
|
"## Settings"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 7,
|
|
"metadata": {
|
|
"colab": {
|
|
"autoexec": {
|
|
"startup": false,
|
|
"wait_interval": 0
|
|
},
|
|
"base_uri": "https://localhost:8080/",
|
|
"height": 34
|
|
},
|
|
"colab_type": "code",
|
|
"executionInfo": {
|
|
"elapsed": 330,
|
|
"status": "ok",
|
|
"timestamp": 1525034084237,
|
|
"user": {
|
|
"displayName": "Sebastian Raschka",
|
|
"photoUrl": "//lh6.googleusercontent.com/-cxK6yOSQ6uE/AAAAAAAAAAI/AAAAAAAAIfw/P9ar_CHsKOQ/s50-c-k-no/photo.jpg",
|
|
"userId": "118404394130788869227"
|
|
},
|
|
"user_tz": 240
|
|
},
|
|
"id": "KthquBjBVA2V",
|
|
"outputId": "4014037d-f8b6-4dcc-e6ec-9db4e4cd38fd"
|
|
},
|
|
"outputs": [
|
|
{
|
|
"name": "stdout",
|
|
"output_type": "stream",
|
|
"text": [
|
|
"Device: cuda:3\n"
|
|
]
|
|
}
|
|
],
|
|
"source": [
|
|
"##########################\n",
|
|
"### SETTINGS\n",
|
|
"##########################\n",
|
|
"\n",
|
|
"# Device\n",
|
|
"device = torch.device(\"cuda:3\" if torch.cuda.is_available() else \"cpu\")\n",
|
|
"print('Device:', device)\n",
|
|
"\n",
|
|
"# Hyperparameters\n",
|
|
"random_seed = 123\n",
|
|
"learning_rate = 0.0005\n",
|
|
"num_epochs = 50"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {
|
|
"colab_type": "text",
|
|
"id": "pnSfNaJrVA2Z"
|
|
},
|
|
"source": [
|
|
"### Model"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 8,
|
|
"metadata": {
|
|
"colab": {
|
|
"autoexec": {
|
|
"startup": false,
|
|
"wait_interval": 0
|
|
}
|
|
},
|
|
"colab_type": "code",
|
|
"id": "aKDo_CM9-5eL"
|
|
},
|
|
"outputs": [],
|
|
"source": [
|
|
"##########################\n",
|
|
"### MODEL\n",
|
|
"##########################\n",
|
|
"\n",
|
|
"\n",
|
|
"class Autoencoder(torch.nn.Module):\n",
|
|
"\n",
|
|
" def __init__(self):\n",
|
|
" super(Autoencoder, self).__init__()\n",
|
|
" \n",
|
|
" # calculate same padding:\n",
|
|
" # (w - k + 2*p)/s + 1 = o\n",
|
|
" # => p = (s(o-1) - w + k)/2\n",
|
|
" \n",
|
|
" ### ENCODER\n",
|
|
" \n",
|
|
" # 28x28x1 => 28x28x4\n",
|
|
" self.conv_1 = torch.nn.Conv2d(in_channels=1,\n",
|
|
" out_channels=4,\n",
|
|
" kernel_size=(3, 3),\n",
|
|
" stride=(1, 1),\n",
|
|
" # (1(28-1) - 28 + 3) / 2 = 1\n",
|
|
" padding=1)\n",
|
|
" # 28x28x4 => 14x14x4 \n",
|
|
" self.pool_1 = torch.nn.MaxPool2d(kernel_size=(2, 2),\n",
|
|
" stride=(2, 2),\n",
|
|
" # (2(14-1) - 28 + 2) / 2 = 0\n",
|
|
" padding=0) \n",
|
|
" # 14x14x4 => 14x14x8\n",
|
|
" self.conv_2 = torch.nn.Conv2d(in_channels=4,\n",
|
|
" out_channels=8,\n",
|
|
" kernel_size=(3, 3),\n",
|
|
" stride=(1, 1),\n",
|
|
" # (1(14-1) - 14 + 3) / 2 = 1\n",
|
|
" padding=1) \n",
|
|
" # 14x14x8 => 7x7x8 \n",
|
|
" self.pool_2 = torch.nn.MaxPool2d(kernel_size=(2, 2),\n",
|
|
" stride=(2, 2),\n",
|
|
" # (2(7-1) - 14 + 2) / 2 = 0\n",
|
|
" padding=0)\n",
|
|
" \n",
|
|
" ### DECODER\n",
|
|
" \n",
|
|
" # 7x7x8 => 14x14x8 \n",
|
|
" \n",
|
|
" ## interpolation\n",
|
|
" \n",
|
|
" # 14x14x8 => 14x14x8\n",
|
|
" self.conv_3 = torch.nn.Conv2d(in_channels=8,\n",
|
|
" out_channels=4,\n",
|
|
" kernel_size=(3, 3),\n",
|
|
" stride=(1, 1),\n",
|
|
" # (1(14-1) - 14 + 3) / 2 = 1\n",
|
|
" padding=1)\n",
|
|
" # 14x14x4 => 28x28x4 \n",
|
|
"\n",
|
|
" ## interpolation\n",
|
|
" \n",
|
|
" # 28x28x4 => 28x28x1\n",
|
|
" self.conv_4 = torch.nn.Conv2d(in_channels=4,\n",
|
|
" out_channels=1,\n",
|
|
" kernel_size=(3, 3),\n",
|
|
" stride=(1, 1),\n",
|
|
" # (1(28-1) - 28 + 3) / 2 = 1\n",
|
|
" padding=1)\n",
|
|
" \n",
|
|
" def forward(self, x):\n",
|
|
" \n",
|
|
" ### ENCODER\n",
|
|
" x = self.conv_1(x)\n",
|
|
" x = F.leaky_relu(x)\n",
|
|
" x = self.pool_1(x)\n",
|
|
" x = self.conv_2(x)\n",
|
|
" x = F.leaky_relu(x)\n",
|
|
" x = self.pool_2(x)\n",
|
|
" \n",
|
|
" ### DECODER\n",
|
|
" x = F.interpolate(x, scale_factor=2, mode='nearest')\n",
|
|
" x = self.conv_3(x)\n",
|
|
" x = F.leaky_relu(x)\n",
|
|
" x = F.interpolate(x, scale_factor=2, mode='nearest')\n",
|
|
" x = self.conv_4(x)\n",
|
|
" x = F.leaky_relu(x)\n",
|
|
" x = torch.sigmoid(x)\n",
|
|
" return x\n",
|
|
"\n",
|
|
" \n",
|
|
"torch.manual_seed(random_seed)\n",
|
|
"model = Autoencoder()\n",
|
|
"model = model.to(device)\n",
|
|
" \n",
|
|
"\n",
|
|
"##########################\n",
|
|
"### COST AND OPTIMIZER\n",
|
|
"##########################\n",
|
|
"\n",
|
|
"cost_fn = torch.nn.BCELoss() # torch.nn.MSELoss()\n",
|
|
"optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {
|
|
"colab_type": "text",
|
|
"id": "PfAT3P8_VA2e"
|
|
},
|
|
"source": [
|
|
"## Training"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 9,
|
|
"metadata": {
|
|
"colab": {
|
|
"autoexec": {
|
|
"startup": false,
|
|
"wait_interval": 0
|
|
},
|
|
"base_uri": "https://localhost:8080/",
|
|
"height": 1224
|
|
},
|
|
"colab_type": "code",
|
|
"executionInfo": {
|
|
"elapsed": 10453399,
|
|
"status": "ok",
|
|
"timestamp": 1525044538453,
|
|
"user": {
|
|
"displayName": "Sebastian Raschka",
|
|
"photoUrl": "//lh6.googleusercontent.com/-cxK6yOSQ6uE/AAAAAAAAAAI/AAAAAAAAIfw/P9ar_CHsKOQ/s50-c-k-no/photo.jpg",
|
|
"userId": "118404394130788869227"
|
|
},
|
|
"user_tz": 240
|
|
},
|
|
"id": "rbZ8ploO_JW2",
|
|
"outputId": "42e24455-31a2-425b-fea4-599e7cc144d3"
|
|
},
|
|
"outputs": [
|
|
{
|
|
"name": "stdout",
|
|
"output_type": "stream",
|
|
"text": [
|
|
"Epoch: 001/050 | Batch 0000/8290 | Cost: 0.2212\n",
|
|
"Epoch: 001/050 | Batch 0500/8290 | Cost: 0.0968\n",
|
|
"Epoch: 001/050 | Batch 1000/8290 | Cost: 0.0681\n",
|
|
"Epoch: 001/050 | Batch 1500/8290 | Cost: 0.0658\n",
|
|
"Epoch: 001/050 | Batch 2000/8290 | Cost: 0.0531\n",
|
|
"Epoch: 001/050 | Batch 2500/8290 | Cost: 0.0411\n",
|
|
"Epoch: 001/050 | Batch 3000/8290 | Cost: 0.0417\n",
|
|
"Epoch: 001/050 | Batch 3500/8290 | Cost: 0.0363\n",
|
|
"Epoch: 001/050 | Batch 4000/8290 | Cost: 0.0338\n",
|
|
"Epoch: 001/050 | Batch 4500/8290 | Cost: 0.0352\n",
|
|
"Epoch: 001/050 | Batch 5000/8290 | Cost: 0.0363\n",
|
|
"Epoch: 001/050 | Batch 5500/8290 | Cost: 0.0334\n",
|
|
"Epoch: 001/050 | Batch 6000/8290 | Cost: 0.0346\n",
|
|
"Epoch: 001/050 | Batch 6500/8290 | Cost: 0.0299\n",
|
|
"Epoch: 001/050 | Batch 7000/8290 | Cost: 0.0320\n",
|
|
"Epoch: 001/050 | Batch 7500/8290 | Cost: 0.0288\n",
|
|
"Epoch: 001/050 | Batch 8000/8290 | Cost: 0.0288\n",
|
|
"Time elapsed: 1.08 min\n",
|
|
"Epoch: 002/050 | Batch 0000/8290 | Cost: 0.0308\n",
|
|
"Epoch: 002/050 | Batch 0500/8290 | Cost: 0.0291\n",
|
|
"Epoch: 002/050 | Batch 1000/8290 | Cost: 0.0292\n",
|
|
"Epoch: 002/050 | Batch 1500/8290 | Cost: 0.0291\n",
|
|
"Epoch: 002/050 | Batch 2000/8290 | Cost: 0.0259\n",
|
|
"Epoch: 002/050 | Batch 2500/8290 | Cost: 0.0274\n",
|
|
"Epoch: 002/050 | Batch 3000/8290 | Cost: 0.0257\n",
|
|
"Epoch: 002/050 | Batch 3500/8290 | Cost: 0.0262\n",
|
|
"Epoch: 002/050 | Batch 4000/8290 | Cost: 0.0261\n",
|
|
"Epoch: 002/050 | Batch 4500/8290 | Cost: 0.0281\n",
|
|
"Epoch: 002/050 | Batch 5000/8290 | Cost: 0.0238\n",
|
|
"Epoch: 002/050 | Batch 5500/8290 | Cost: 0.0236\n",
|
|
"Epoch: 002/050 | Batch 6000/8290 | Cost: 0.0234\n",
|
|
"Epoch: 002/050 | Batch 6500/8290 | Cost: 0.0219\n",
|
|
"Epoch: 002/050 | Batch 7000/8290 | Cost: 0.0225\n",
|
|
"Epoch: 002/050 | Batch 7500/8290 | Cost: 0.0237\n",
|
|
"Epoch: 002/050 | Batch 8000/8290 | Cost: 0.0237\n",
|
|
"Time elapsed: 2.01 min\n",
|
|
"Epoch: 003/050 | Batch 0000/8290 | Cost: 0.0263\n",
|
|
"Epoch: 003/050 | Batch 0500/8290 | Cost: 0.0231\n",
|
|
"Epoch: 003/050 | Batch 1000/8290 | Cost: 0.0231\n",
|
|
"Epoch: 003/050 | Batch 1500/8290 | Cost: 0.0238\n",
|
|
"Epoch: 003/050 | Batch 2000/8290 | Cost: 0.0235\n",
|
|
"Epoch: 003/050 | Batch 2500/8290 | Cost: 0.0218\n",
|
|
"Epoch: 003/050 | Batch 3000/8290 | Cost: 0.0214\n",
|
|
"Epoch: 003/050 | Batch 3500/8290 | Cost: 0.0223\n",
|
|
"Epoch: 003/050 | Batch 4000/8290 | Cost: 0.0230\n",
|
|
"Epoch: 003/050 | Batch 4500/8290 | Cost: 0.0231\n",
|
|
"Epoch: 003/050 | Batch 5000/8290 | Cost: 0.0222\n",
|
|
"Epoch: 003/050 | Batch 5500/8290 | Cost: 0.0240\n",
|
|
"Epoch: 003/050 | Batch 6000/8290 | Cost: 0.0224\n",
|
|
"Epoch: 003/050 | Batch 6500/8290 | Cost: 0.0221\n",
|
|
"Epoch: 003/050 | Batch 7000/8290 | Cost: 0.0225\n",
|
|
"Epoch: 003/050 | Batch 7500/8290 | Cost: 0.0223\n",
|
|
"Epoch: 003/050 | Batch 8000/8290 | Cost: 0.0210\n",
|
|
"Time elapsed: 2.95 min\n",
|
|
"Epoch: 004/050 | Batch 0000/8290 | Cost: 0.0215\n",
|
|
"Epoch: 004/050 | Batch 0500/8290 | Cost: 0.0225\n",
|
|
"Epoch: 004/050 | Batch 1000/8290 | Cost: 0.0235\n",
|
|
"Epoch: 004/050 | Batch 1500/8290 | Cost: 0.0230\n",
|
|
"Epoch: 004/050 | Batch 2000/8290 | Cost: 0.0218\n",
|
|
"Epoch: 004/050 | Batch 2500/8290 | Cost: 0.0228\n",
|
|
"Epoch: 004/050 | Batch 3000/8290 | Cost: 0.0223\n",
|
|
"Epoch: 004/050 | Batch 3500/8290 | Cost: 0.0221\n",
|
|
"Epoch: 004/050 | Batch 4000/8290 | Cost: 0.0225\n",
|
|
"Epoch: 004/050 | Batch 4500/8290 | Cost: 0.0212\n",
|
|
"Epoch: 004/050 | Batch 5000/8290 | Cost: 0.0219\n",
|
|
"Epoch: 004/050 | Batch 5500/8290 | Cost: 0.0216\n",
|
|
"Epoch: 004/050 | Batch 6000/8290 | Cost: 0.0229\n",
|
|
"Epoch: 004/050 | Batch 6500/8290 | Cost: 0.0221\n",
|
|
"Epoch: 004/050 | Batch 7000/8290 | Cost: 0.0237\n",
|
|
"Epoch: 004/050 | Batch 7500/8290 | Cost: 0.0195\n",
|
|
"Epoch: 004/050 | Batch 8000/8290 | Cost: 0.0195\n",
|
|
"Time elapsed: 3.88 min\n",
|
|
"Epoch: 005/050 | Batch 0000/8290 | Cost: 0.0219\n",
|
|
"Epoch: 005/050 | Batch 0500/8290 | Cost: 0.0214\n",
|
|
"Epoch: 005/050 | Batch 1000/8290 | Cost: 0.0217\n",
|
|
"Epoch: 005/050 | Batch 1500/8290 | Cost: 0.0210\n",
|
|
"Epoch: 005/050 | Batch 2000/8290 | Cost: 0.0220\n",
|
|
"Epoch: 005/050 | Batch 2500/8290 | Cost: 0.0215\n",
|
|
"Epoch: 005/050 | Batch 3000/8290 | Cost: 0.0213\n",
|
|
"Epoch: 005/050 | Batch 3500/8290 | Cost: 0.0214\n",
|
|
"Epoch: 005/050 | Batch 4000/8290 | Cost: 0.0232\n",
|
|
"Epoch: 005/050 | Batch 4500/8290 | Cost: 0.0216\n",
|
|
"Epoch: 005/050 | Batch 5000/8290 | Cost: 0.0214\n",
|
|
"Epoch: 005/050 | Batch 5500/8290 | Cost: 0.0228\n",
|
|
"Epoch: 005/050 | Batch 6000/8290 | Cost: 0.0211\n",
|
|
"Epoch: 005/050 | Batch 6500/8290 | Cost: 0.0221\n",
|
|
"Epoch: 005/050 | Batch 7000/8290 | Cost: 0.0216\n",
|
|
"Epoch: 005/050 | Batch 7500/8290 | Cost: 0.0216\n",
|
|
"Epoch: 005/050 | Batch 8000/8290 | Cost: 0.0208\n",
|
|
"Time elapsed: 4.82 min\n",
|
|
"Epoch: 006/050 | Batch 0000/8290 | Cost: 0.0227\n",
|
|
"Epoch: 006/050 | Batch 0500/8290 | Cost: 0.0226\n",
|
|
"Epoch: 006/050 | Batch 1000/8290 | Cost: 0.0214\n",
|
|
"Epoch: 006/050 | Batch 1500/8290 | Cost: 0.0206\n",
|
|
"Epoch: 006/050 | Batch 2000/8290 | Cost: 0.0209\n",
|
|
"Epoch: 006/050 | Batch 2500/8290 | Cost: 0.0208\n",
|
|
"Epoch: 006/050 | Batch 3000/8290 | Cost: 0.0215\n",
|
|
"Epoch: 006/050 | Batch 3500/8290 | Cost: 0.0226\n",
|
|
"Epoch: 006/050 | Batch 4000/8290 | Cost: 0.0203\n",
|
|
"Epoch: 006/050 | Batch 4500/8290 | Cost: 0.0195\n",
|
|
"Epoch: 006/050 | Batch 5000/8290 | Cost: 0.0215\n",
|
|
"Epoch: 006/050 | Batch 5500/8290 | Cost: 0.0204\n",
|
|
"Epoch: 006/050 | Batch 6000/8290 | Cost: 0.0196\n",
|
|
"Epoch: 006/050 | Batch 6500/8290 | Cost: 0.0209\n",
|
|
"Epoch: 006/050 | Batch 7000/8290 | Cost: 0.0197\n",
|
|
"Epoch: 006/050 | Batch 7500/8290 | Cost: 0.0191\n",
|
|
"Epoch: 006/050 | Batch 8000/8290 | Cost: 0.0218\n",
|
|
"Time elapsed: 5.76 min\n",
|
|
"Epoch: 007/050 | Batch 0000/8290 | Cost: 0.0198\n",
|
|
"Epoch: 007/050 | Batch 0500/8290 | Cost: 0.0211\n",
|
|
"Epoch: 007/050 | Batch 1000/8290 | Cost: 0.0218\n",
|
|
"Epoch: 007/050 | Batch 1500/8290 | Cost: 0.0216\n",
|
|
"Epoch: 007/050 | Batch 2000/8290 | Cost: 0.0218\n",
|
|
"Epoch: 007/050 | Batch 2500/8290 | Cost: 0.0203\n",
|
|
"Epoch: 007/050 | Batch 3000/8290 | Cost: 0.0196\n",
|
|
"Epoch: 007/050 | Batch 3500/8290 | Cost: 0.0203\n",
|
|
"Epoch: 007/050 | Batch 4000/8290 | Cost: 0.0202\n",
|
|
"Epoch: 007/050 | Batch 4500/8290 | Cost: 0.0219\n",
|
|
"Epoch: 007/050 | Batch 5000/8290 | Cost: 0.0190\n",
|
|
"Epoch: 007/050 | Batch 5500/8290 | Cost: 0.0204\n",
|
|
"Epoch: 007/050 | Batch 6000/8290 | Cost: 0.0206\n",
|
|
"Epoch: 007/050 | Batch 6500/8290 | Cost: 0.0201\n",
|
|
"Epoch: 007/050 | Batch 7000/8290 | Cost: 0.0210\n",
|
|
"Epoch: 007/050 | Batch 7500/8290 | Cost: 0.0213\n",
|
|
"Epoch: 007/050 | Batch 8000/8290 | Cost: 0.0215\n",
|
|
"Time elapsed: 6.69 min\n",
|
|
"Epoch: 008/050 | Batch 0000/8290 | Cost: 0.0218\n",
|
|
"Epoch: 008/050 | Batch 0500/8290 | Cost: 0.0229\n",
|
|
"Epoch: 008/050 | Batch 1000/8290 | Cost: 0.0194\n",
|
|
"Epoch: 008/050 | Batch 1500/8290 | Cost: 0.0220\n",
|
|
"Epoch: 008/050 | Batch 2000/8290 | Cost: 0.0208\n",
|
|
"Epoch: 008/050 | Batch 2500/8290 | Cost: 0.0209\n",
|
|
"Epoch: 008/050 | Batch 3000/8290 | Cost: 0.0197\n",
|
|
"Epoch: 008/050 | Batch 3500/8290 | Cost: 0.0218\n",
|
|
"Epoch: 008/050 | Batch 4000/8290 | Cost: 0.0214\n",
|
|
"Epoch: 008/050 | Batch 4500/8290 | Cost: 0.0209\n",
|
|
"Epoch: 008/050 | Batch 5000/8290 | Cost: 0.0202\n",
|
|
"Epoch: 008/050 | Batch 5500/8290 | Cost: 0.0202\n",
|
|
"Epoch: 008/050 | Batch 6000/8290 | Cost: 0.0215\n",
|
|
"Epoch: 008/050 | Batch 6500/8290 | Cost: 0.0213\n",
|
|
"Epoch: 008/050 | Batch 7000/8290 | Cost: 0.0215\n",
|
|
"Epoch: 008/050 | Batch 7500/8290 | Cost: 0.0201\n",
|
|
"Epoch: 008/050 | Batch 8000/8290 | Cost: 0.0215\n",
|
|
"Time elapsed: 7.62 min\n",
|
|
"Epoch: 009/050 | Batch 0000/8290 | Cost: 0.0212\n",
|
|
"Epoch: 009/050 | Batch 0500/8290 | Cost: 0.0211\n",
|
|
"Epoch: 009/050 | Batch 1000/8290 | Cost: 0.0208\n",
|
|
"Epoch: 009/050 | Batch 1500/8290 | Cost: 0.0213\n",
|
|
"Epoch: 009/050 | Batch 2000/8290 | Cost: 0.0196\n",
|
|
"Epoch: 009/050 | Batch 2500/8290 | Cost: 0.0199\n",
|
|
"Epoch: 009/050 | Batch 3000/8290 | Cost: 0.0211\n",
|
|
"Epoch: 009/050 | Batch 3500/8290 | Cost: 0.0218\n",
|
|
"Epoch: 009/050 | Batch 4000/8290 | Cost: 0.0201\n",
|
|
"Epoch: 009/050 | Batch 4500/8290 | Cost: 0.0208\n",
|
|
"Epoch: 009/050 | Batch 5000/8290 | Cost: 0.0202\n",
|
|
"Epoch: 009/050 | Batch 5500/8290 | Cost: 0.0202\n",
|
|
"Epoch: 009/050 | Batch 6000/8290 | Cost: 0.0214\n",
|
|
"Epoch: 009/050 | Batch 6500/8290 | Cost: 0.0197\n",
|
|
"Epoch: 009/050 | Batch 7000/8290 | Cost: 0.0205\n",
|
|
"Epoch: 009/050 | Batch 7500/8290 | Cost: 0.0196\n",
|
|
"Epoch: 009/050 | Batch 8000/8290 | Cost: 0.0199\n",
|
|
"Time elapsed: 8.55 min\n",
|
|
"Epoch: 010/050 | Batch 0000/8290 | Cost: 0.0200\n",
|
|
"Epoch: 010/050 | Batch 0500/8290 | Cost: 0.0210\n",
|
|
"Epoch: 010/050 | Batch 1000/8290 | Cost: 0.0203\n",
|
|
"Epoch: 010/050 | Batch 1500/8290 | Cost: 0.0198\n",
|
|
"Epoch: 010/050 | Batch 2000/8290 | Cost: 0.0186\n",
|
|
"Epoch: 010/050 | Batch 2500/8290 | Cost: 0.0205\n",
|
|
"Epoch: 010/050 | Batch 3000/8290 | Cost: 0.0201\n",
|
|
"Epoch: 010/050 | Batch 3500/8290 | Cost: 0.0197\n",
|
|
"Epoch: 010/050 | Batch 4000/8290 | Cost: 0.0211\n",
|
|
"Epoch: 010/050 | Batch 4500/8290 | Cost: 0.0182\n",
|
|
"Epoch: 010/050 | Batch 5000/8290 | Cost: 0.0208\n",
|
|
"Epoch: 010/050 | Batch 5500/8290 | Cost: 0.0182\n",
|
|
"Epoch: 010/050 | Batch 6000/8290 | Cost: 0.0209\n",
|
|
"Epoch: 010/050 | Batch 6500/8290 | Cost: 0.0229\n",
|
|
"Epoch: 010/050 | Batch 7000/8290 | Cost: 0.0180\n",
|
|
"Epoch: 010/050 | Batch 7500/8290 | Cost: 0.0211\n",
|
|
"Epoch: 010/050 | Batch 8000/8290 | Cost: 0.0194\n",
|
|
"Time elapsed: 9.48 min\n",
|
|
"Epoch: 011/050 | Batch 0000/8290 | Cost: 0.0203\n",
|
|
"Epoch: 011/050 | Batch 0500/8290 | Cost: 0.0197\n",
|
|
"Epoch: 011/050 | Batch 1000/8290 | Cost: 0.0212\n",
|
|
"Epoch: 011/050 | Batch 1500/8290 | Cost: 0.0207\n",
|
|
"Epoch: 011/050 | Batch 2000/8290 | Cost: 0.0203\n",
|
|
"Epoch: 011/050 | Batch 2500/8290 | Cost: 0.0194\n",
|
|
"Epoch: 011/050 | Batch 3000/8290 | Cost: 0.0191\n",
|
|
"Epoch: 011/050 | Batch 3500/8290 | Cost: 0.0202\n",
|
|
"Epoch: 011/050 | Batch 4000/8290 | Cost: 0.0206\n",
|
|
"Epoch: 011/050 | Batch 4500/8290 | Cost: 0.0213\n",
|
|
"Epoch: 011/050 | Batch 5000/8290 | Cost: 0.0205\n",
|
|
"Epoch: 011/050 | Batch 5500/8290 | Cost: 0.0221\n",
|
|
"Epoch: 011/050 | Batch 6000/8290 | Cost: 0.0193\n",
|
|
"Epoch: 011/050 | Batch 6500/8290 | Cost: 0.0203\n",
|
|
"Epoch: 011/050 | Batch 7000/8290 | Cost: 0.0205\n",
|
|
"Epoch: 011/050 | Batch 7500/8290 | Cost: 0.0202\n",
|
|
"Epoch: 011/050 | Batch 8000/8290 | Cost: 0.0206\n",
|
|
"Time elapsed: 10.40 min\n",
|
|
"Epoch: 012/050 | Batch 0000/8290 | Cost: 0.0201\n",
|
|
"Epoch: 012/050 | Batch 0500/8290 | Cost: 0.0190\n",
|
|
"Epoch: 012/050 | Batch 1000/8290 | Cost: 0.0190\n",
|
|
"Epoch: 012/050 | Batch 1500/8290 | Cost: 0.0200\n",
|
|
"Epoch: 012/050 | Batch 2000/8290 | Cost: 0.0216\n",
|
|
"Epoch: 012/050 | Batch 2500/8290 | Cost: 0.0199\n",
|
|
"Epoch: 012/050 | Batch 3000/8290 | Cost: 0.0206\n",
|
|
"Epoch: 012/050 | Batch 3500/8290 | Cost: 0.0199\n",
|
|
"Epoch: 012/050 | Batch 4000/8290 | Cost: 0.0202\n",
|
|
"Epoch: 012/050 | Batch 4500/8290 | Cost: 0.0208\n",
|
|
"Epoch: 012/050 | Batch 5000/8290 | Cost: 0.0201\n",
|
|
"Epoch: 012/050 | Batch 5500/8290 | Cost: 0.0206\n",
|
|
"Epoch: 012/050 | Batch 6000/8290 | Cost: 0.0192\n",
|
|
"Epoch: 012/050 | Batch 6500/8290 | Cost: 0.0210\n",
|
|
"Epoch: 012/050 | Batch 7000/8290 | Cost: 0.0228\n",
|
|
"Epoch: 012/050 | Batch 7500/8290 | Cost: 0.0200\n",
|
|
"Epoch: 012/050 | Batch 8000/8290 | Cost: 0.0191\n",
|
|
"Time elapsed: 11.33 min\n",
|
|
"Epoch: 013/050 | Batch 0000/8290 | Cost: 0.0196\n",
|
|
"Epoch: 013/050 | Batch 0500/8290 | Cost: 0.0193\n",
|
|
"Epoch: 013/050 | Batch 1000/8290 | Cost: 0.0188\n",
|
|
"Epoch: 013/050 | Batch 1500/8290 | Cost: 0.0187\n",
|
|
"Epoch: 013/050 | Batch 2000/8290 | Cost: 0.0205\n",
|
|
"Epoch: 013/050 | Batch 2500/8290 | Cost: 0.0195\n",
|
|
"Epoch: 013/050 | Batch 3000/8290 | Cost: 0.0201\n",
|
|
"Epoch: 013/050 | Batch 3500/8290 | Cost: 0.0216\n",
|
|
"Epoch: 013/050 | Batch 4000/8290 | Cost: 0.0205\n",
|
|
"Epoch: 013/050 | Batch 4500/8290 | Cost: 0.0194\n",
|
|
"Epoch: 013/050 | Batch 5000/8290 | Cost: 0.0202\n",
|
|
"Epoch: 013/050 | Batch 5500/8290 | Cost: 0.0213\n",
|
|
"Epoch: 013/050 | Batch 6000/8290 | Cost: 0.0186\n",
|
|
"Epoch: 013/050 | Batch 6500/8290 | Cost: 0.0195\n",
|
|
"Epoch: 013/050 | Batch 7000/8290 | Cost: 0.0181\n",
|
|
"Epoch: 013/050 | Batch 7500/8290 | Cost: 0.0189\n",
|
|
"Epoch: 013/050 | Batch 8000/8290 | Cost: 0.0190\n",
|
|
"Time elapsed: 12.27 min\n",
|
|
"Epoch: 014/050 | Batch 0000/8290 | Cost: 0.0192\n",
|
|
"Epoch: 014/050 | Batch 0500/8290 | Cost: 0.0198\n",
|
|
"Epoch: 014/050 | Batch 1000/8290 | Cost: 0.0209\n",
|
|
"Epoch: 014/050 | Batch 1500/8290 | Cost: 0.0188\n",
|
|
"Epoch: 014/050 | Batch 2000/8290 | Cost: 0.0200\n",
|
|
"Epoch: 014/050 | Batch 2500/8290 | Cost: 0.0190\n",
|
|
"Epoch: 014/050 | Batch 3000/8290 | Cost: 0.0200\n",
|
|
"Epoch: 014/050 | Batch 3500/8290 | Cost: 0.0203\n",
|
|
"Epoch: 014/050 | Batch 4000/8290 | Cost: 0.0205\n",
|
|
"Epoch: 014/050 | Batch 4500/8290 | Cost: 0.0184\n",
|
|
"Epoch: 014/050 | Batch 5000/8290 | Cost: 0.0194\n",
|
|
"Epoch: 014/050 | Batch 5500/8290 | Cost: 0.0193\n",
|
|
"Epoch: 014/050 | Batch 6000/8290 | Cost: 0.0195\n",
|
|
"Epoch: 014/050 | Batch 6500/8290 | Cost: 0.0195\n",
|
|
"Epoch: 014/050 | Batch 7000/8290 | Cost: 0.0214\n",
|
|
"Epoch: 014/050 | Batch 7500/8290 | Cost: 0.0189\n",
|
|
"Epoch: 014/050 | Batch 8000/8290 | Cost: 0.0201\n",
|
|
"Time elapsed: 13.20 min\n",
|
|
"Epoch: 015/050 | Batch 0000/8290 | Cost: 0.0200\n",
|
|
"Epoch: 015/050 | Batch 0500/8290 | Cost: 0.0198\n",
|
|
"Epoch: 015/050 | Batch 1000/8290 | Cost: 0.0181\n",
|
|
"Epoch: 015/050 | Batch 1500/8290 | Cost: 0.0196\n",
|
|
"Epoch: 015/050 | Batch 2000/8290 | Cost: 0.0204\n",
|
|
"Epoch: 015/050 | Batch 2500/8290 | Cost: 0.0186\n",
|
|
"Epoch: 015/050 | Batch 3000/8290 | Cost: 0.0201\n",
|
|
"Epoch: 015/050 | Batch 3500/8290 | Cost: 0.0201\n",
|
|
"Epoch: 015/050 | Batch 4000/8290 | Cost: 0.0193\n",
|
|
"Epoch: 015/050 | Batch 4500/8290 | Cost: 0.0197\n",
|
|
"Epoch: 015/050 | Batch 5000/8290 | Cost: 0.0181\n",
|
|
"Epoch: 015/050 | Batch 5500/8290 | Cost: 0.0194\n",
|
|
"Epoch: 015/050 | Batch 6000/8290 | Cost: 0.0190\n",
|
|
"Epoch: 015/050 | Batch 6500/8290 | Cost: 0.0202\n",
|
|
"Epoch: 015/050 | Batch 7000/8290 | Cost: 0.0188\n",
|
|
"Epoch: 015/050 | Batch 7500/8290 | Cost: 0.0208\n",
|
|
"Epoch: 015/050 | Batch 8000/8290 | Cost: 0.0185\n",
|
|
"Time elapsed: 14.14 min\n",
|
|
"Epoch: 016/050 | Batch 0000/8290 | Cost: 0.0186\n",
|
|
"Epoch: 016/050 | Batch 0500/8290 | Cost: 0.0194\n",
|
|
"Epoch: 016/050 | Batch 1000/8290 | Cost: 0.0217\n",
|
|
"Epoch: 016/050 | Batch 1500/8290 | Cost: 0.0194\n",
|
|
"Epoch: 016/050 | Batch 2000/8290 | Cost: 0.0180\n",
|
|
"Epoch: 016/050 | Batch 2500/8290 | Cost: 0.0194\n",
|
|
"Epoch: 016/050 | Batch 3000/8290 | Cost: 0.0197\n",
|
|
"Epoch: 016/050 | Batch 3500/8290 | Cost: 0.0183\n",
|
|
"Epoch: 016/050 | Batch 4000/8290 | Cost: 0.0198\n",
|
|
"Epoch: 016/050 | Batch 4500/8290 | Cost: 0.0193\n",
|
|
"Epoch: 016/050 | Batch 5000/8290 | Cost: 0.0198\n",
|
|
"Epoch: 016/050 | Batch 5500/8290 | Cost: 0.0177\n",
|
|
"Epoch: 016/050 | Batch 6000/8290 | Cost: 0.0195\n",
|
|
"Epoch: 016/050 | Batch 6500/8290 | Cost: 0.0203\n",
|
|
"Epoch: 016/050 | Batch 7000/8290 | Cost: 0.0181\n",
|
|
"Epoch: 016/050 | Batch 7500/8290 | Cost: 0.0196\n",
|
|
"Epoch: 016/050 | Batch 8000/8290 | Cost: 0.0186\n",
|
|
"Time elapsed: 15.08 min\n",
|
|
"Epoch: 017/050 | Batch 0000/8290 | Cost: 0.0179\n",
|
|
"Epoch: 017/050 | Batch 0500/8290 | Cost: 0.0200\n",
|
|
"Epoch: 017/050 | Batch 1000/8290 | Cost: 0.0202\n",
|
|
"Epoch: 017/050 | Batch 1500/8290 | Cost: 0.0198\n",
|
|
"Epoch: 017/050 | Batch 2000/8290 | Cost: 0.0199\n",
|
|
"Epoch: 017/050 | Batch 2500/8290 | Cost: 0.0201\n",
|
|
"Epoch: 017/050 | Batch 3000/8290 | Cost: 0.0195\n",
|
|
"Epoch: 017/050 | Batch 3500/8290 | Cost: 0.0193\n",
|
|
"Epoch: 017/050 | Batch 4000/8290 | Cost: 0.0191\n",
|
|
"Epoch: 017/050 | Batch 4500/8290 | Cost: 0.0205\n",
|
|
"Epoch: 017/050 | Batch 5000/8290 | Cost: 0.0200\n",
|
|
"Epoch: 017/050 | Batch 5500/8290 | Cost: 0.0200\n",
|
|
"Epoch: 017/050 | Batch 6000/8290 | Cost: 0.0202\n",
|
|
"Epoch: 017/050 | Batch 6500/8290 | Cost: 0.0193\n",
|
|
"Epoch: 017/050 | Batch 7000/8290 | Cost: 0.0192\n",
|
|
"Epoch: 017/050 | Batch 7500/8290 | Cost: 0.0207\n",
|
|
"Epoch: 017/050 | Batch 8000/8290 | Cost: 0.0183\n",
|
|
"Time elapsed: 16.02 min\n",
|
|
"Epoch: 018/050 | Batch 0000/8290 | Cost: 0.0199\n",
|
|
"Epoch: 018/050 | Batch 0500/8290 | Cost: 0.0202\n",
|
|
"Epoch: 018/050 | Batch 1000/8290 | Cost: 0.0203\n",
|
|
"Epoch: 018/050 | Batch 1500/8290 | Cost: 0.0193\n",
|
|
"Epoch: 018/050 | Batch 2000/8290 | Cost: 0.0190\n",
|
|
"Epoch: 018/050 | Batch 2500/8290 | Cost: 0.0205\n",
|
|
"Epoch: 018/050 | Batch 3000/8290 | Cost: 0.0210\n",
|
|
"Epoch: 018/050 | Batch 3500/8290 | Cost: 0.0184\n",
|
|
"Epoch: 018/050 | Batch 4000/8290 | Cost: 0.0197\n",
|
|
"Epoch: 018/050 | Batch 4500/8290 | Cost: 0.0206\n",
|
|
"Epoch: 018/050 | Batch 5000/8290 | Cost: 0.0173\n",
|
|
"Epoch: 018/050 | Batch 5500/8290 | Cost: 0.0177\n",
|
|
"Epoch: 018/050 | Batch 6000/8290 | Cost: 0.0198\n",
|
|
"Epoch: 018/050 | Batch 6500/8290 | Cost: 0.0200\n",
|
|
"Epoch: 018/050 | Batch 7000/8290 | Cost: 0.0203\n",
|
|
"Epoch: 018/050 | Batch 7500/8290 | Cost: 0.0199\n",
|
|
"Epoch: 018/050 | Batch 8000/8290 | Cost: 0.0193\n",
|
|
"Time elapsed: 16.96 min\n",
|
|
"Epoch: 019/050 | Batch 0000/8290 | Cost: 0.0193\n",
|
|
"Epoch: 019/050 | Batch 0500/8290 | Cost: 0.0193\n",
|
|
"Epoch: 019/050 | Batch 1000/8290 | Cost: 0.0202\n",
|
|
"Epoch: 019/050 | Batch 1500/8290 | Cost: 0.0210\n",
|
|
"Epoch: 019/050 | Batch 2000/8290 | Cost: 0.0193\n",
|
|
"Epoch: 019/050 | Batch 2500/8290 | Cost: 0.0193\n",
|
|
"Epoch: 019/050 | Batch 3000/8290 | Cost: 0.0197\n",
|
|
"Epoch: 019/050 | Batch 3500/8290 | Cost: 0.0197\n",
|
|
"Epoch: 019/050 | Batch 4000/8290 | Cost: 0.0208\n",
|
|
"Epoch: 019/050 | Batch 4500/8290 | Cost: 0.0199\n",
|
|
"Epoch: 019/050 | Batch 5000/8290 | Cost: 0.0191\n",
|
|
"Epoch: 019/050 | Batch 5500/8290 | Cost: 0.0204\n",
|
|
"Epoch: 019/050 | Batch 6000/8290 | Cost: 0.0208\n",
|
|
"Epoch: 019/050 | Batch 6500/8290 | Cost: 0.0207\n",
|
|
"Epoch: 019/050 | Batch 7000/8290 | Cost: 0.0184\n",
|
|
"Epoch: 019/050 | Batch 7500/8290 | Cost: 0.0182\n",
|
|
"Epoch: 019/050 | Batch 8000/8290 | Cost: 0.0203\n",
|
|
"Time elapsed: 17.91 min\n",
|
|
"Epoch: 020/050 | Batch 0000/8290 | Cost: 0.0189\n",
|
|
"Epoch: 020/050 | Batch 0500/8290 | Cost: 0.0190\n",
|
|
"Epoch: 020/050 | Batch 1000/8290 | Cost: 0.0204\n",
|
|
"Epoch: 020/050 | Batch 1500/8290 | Cost: 0.0174\n",
|
|
"Epoch: 020/050 | Batch 2000/8290 | Cost: 0.0205\n",
|
|
"Epoch: 020/050 | Batch 2500/8290 | Cost: 0.0191\n",
|
|
"Epoch: 020/050 | Batch 3000/8290 | Cost: 0.0187\n",
|
|
"Epoch: 020/050 | Batch 3500/8290 | Cost: 0.0192\n",
|
|
"Epoch: 020/050 | Batch 4000/8290 | Cost: 0.0207\n",
|
|
"Epoch: 020/050 | Batch 4500/8290 | Cost: 0.0183\n",
|
|
"Epoch: 020/050 | Batch 5000/8290 | Cost: 0.0191\n",
|
|
"Epoch: 020/050 | Batch 5500/8290 | Cost: 0.0202\n",
|
|
"Epoch: 020/050 | Batch 6000/8290 | Cost: 0.0208\n",
|
|
"Epoch: 020/050 | Batch 6500/8290 | Cost: 0.0211\n",
|
|
"Epoch: 020/050 | Batch 7000/8290 | Cost: 0.0202\n",
|
|
"Epoch: 020/050 | Batch 7500/8290 | Cost: 0.0199\n",
|
|
"Epoch: 020/050 | Batch 8000/8290 | Cost: 0.0184\n",
|
|
"Time elapsed: 18.85 min\n",
|
|
"Epoch: 021/050 | Batch 0000/8290 | Cost: 0.0188\n",
|
|
"Epoch: 021/050 | Batch 0500/8290 | Cost: 0.0208\n",
|
|
"Epoch: 021/050 | Batch 1000/8290 | Cost: 0.0190\n",
|
|
"Epoch: 021/050 | Batch 1500/8290 | Cost: 0.0204\n",
|
|
"Epoch: 021/050 | Batch 2000/8290 | Cost: 0.0203\n",
|
|
"Epoch: 021/050 | Batch 2500/8290 | Cost: 0.0188\n",
|
|
"Epoch: 021/050 | Batch 3000/8290 | Cost: 0.0205\n",
|
|
"Epoch: 021/050 | Batch 3500/8290 | Cost: 0.0188\n",
|
|
"Epoch: 021/050 | Batch 4000/8290 | Cost: 0.0195\n",
|
|
"Epoch: 021/050 | Batch 4500/8290 | Cost: 0.0199\n",
|
|
"Epoch: 021/050 | Batch 5000/8290 | Cost: 0.0198\n",
|
|
"Epoch: 021/050 | Batch 5500/8290 | Cost: 0.0208\n",
|
|
"Epoch: 021/050 | Batch 6000/8290 | Cost: 0.0195\n",
|
|
"Epoch: 021/050 | Batch 6500/8290 | Cost: 0.0194\n",
|
|
"Epoch: 021/050 | Batch 7000/8290 | Cost: 0.0187\n",
|
|
"Epoch: 021/050 | Batch 7500/8290 | Cost: 0.0175\n",
|
|
"Epoch: 021/050 | Batch 8000/8290 | Cost: 0.0195\n",
|
|
"Time elapsed: 19.79 min\n",
|
|
"Epoch: 022/050 | Batch 0000/8290 | Cost: 0.0184\n",
|
|
"Epoch: 022/050 | Batch 0500/8290 | Cost: 0.0184\n",
|
|
"Epoch: 022/050 | Batch 1000/8290 | Cost: 0.0206\n",
|
|
"Epoch: 022/050 | Batch 1500/8290 | Cost: 0.0195\n",
|
|
"Epoch: 022/050 | Batch 2000/8290 | Cost: 0.0192\n",
|
|
"Epoch: 022/050 | Batch 2500/8290 | Cost: 0.0184\n",
|
|
"Epoch: 022/050 | Batch 3000/8290 | Cost: 0.0208\n",
|
|
"Epoch: 022/050 | Batch 3500/8290 | Cost: 0.0194\n",
|
|
"Epoch: 022/050 | Batch 4000/8290 | Cost: 0.0206\n",
|
|
"Epoch: 022/050 | Batch 4500/8290 | Cost: 0.0184\n",
|
|
"Epoch: 022/050 | Batch 5000/8290 | Cost: 0.0201\n",
|
|
"Epoch: 022/050 | Batch 5500/8290 | Cost: 0.0196\n",
|
|
"Epoch: 022/050 | Batch 6000/8290 | Cost: 0.0192\n",
|
|
"Epoch: 022/050 | Batch 6500/8290 | Cost: 0.0196\n",
|
|
"Epoch: 022/050 | Batch 7000/8290 | Cost: 0.0198\n",
|
|
"Epoch: 022/050 | Batch 7500/8290 | Cost: 0.0183\n",
|
|
"Epoch: 022/050 | Batch 8000/8290 | Cost: 0.0199\n",
|
|
"Time elapsed: 20.72 min\n",
|
|
"Epoch: 023/050 | Batch 0000/8290 | Cost: 0.0206\n",
|
|
"Epoch: 023/050 | Batch 0500/8290 | Cost: 0.0210\n",
|
|
"Epoch: 023/050 | Batch 1000/8290 | Cost: 0.0193\n",
|
|
"Epoch: 023/050 | Batch 1500/8290 | Cost: 0.0186\n",
|
|
"Epoch: 023/050 | Batch 2000/8290 | Cost: 0.0180\n",
|
|
"Epoch: 023/050 | Batch 2500/8290 | Cost: 0.0201\n",
|
|
"Epoch: 023/050 | Batch 3000/8290 | Cost: 0.0192\n",
|
|
"Epoch: 023/050 | Batch 3500/8290 | Cost: 0.0187\n",
|
|
"Epoch: 023/050 | Batch 4000/8290 | Cost: 0.0195\n",
|
|
"Epoch: 023/050 | Batch 4500/8290 | Cost: 0.0174\n",
|
|
"Epoch: 023/050 | Batch 5000/8290 | Cost: 0.0199\n",
|
|
"Epoch: 023/050 | Batch 5500/8290 | Cost: 0.0193\n",
|
|
"Epoch: 023/050 | Batch 6000/8290 | Cost: 0.0183\n",
|
|
"Epoch: 023/050 | Batch 6500/8290 | Cost: 0.0193\n",
|
|
"Epoch: 023/050 | Batch 7000/8290 | Cost: 0.0182\n",
|
|
"Epoch: 023/050 | Batch 7500/8290 | Cost: 0.0186\n",
|
|
"Epoch: 023/050 | Batch 8000/8290 | Cost: 0.0208\n",
|
|
"Time elapsed: 21.65 min\n",
|
|
"Epoch: 024/050 | Batch 0000/8290 | Cost: 0.0189\n",
|
|
"Epoch: 024/050 | Batch 0500/8290 | Cost: 0.0189\n",
|
|
"Epoch: 024/050 | Batch 1000/8290 | Cost: 0.0191\n",
|
|
"Epoch: 024/050 | Batch 1500/8290 | Cost: 0.0193\n",
|
|
"Epoch: 024/050 | Batch 2000/8290 | Cost: 0.0192\n",
|
|
"Epoch: 024/050 | Batch 2500/8290 | Cost: 0.0193\n",
|
|
"Epoch: 024/050 | Batch 3000/8290 | Cost: 0.0201\n",
|
|
"Epoch: 024/050 | Batch 3500/8290 | Cost: 0.0193\n",
|
|
"Epoch: 024/050 | Batch 4000/8290 | Cost: 0.0201\n",
|
|
"Epoch: 024/050 | Batch 4500/8290 | Cost: 0.0205\n",
|
|
"Epoch: 024/050 | Batch 5000/8290 | Cost: 0.0193\n",
|
|
"Epoch: 024/050 | Batch 5500/8290 | Cost: 0.0186\n",
|
|
"Epoch: 024/050 | Batch 6000/8290 | Cost: 0.0187\n",
|
|
"Epoch: 024/050 | Batch 6500/8290 | Cost: 0.0186\n",
|
|
"Epoch: 024/050 | Batch 7000/8290 | Cost: 0.0191\n",
|
|
"Epoch: 024/050 | Batch 7500/8290 | Cost: 0.0202\n",
|
|
"Epoch: 024/050 | Batch 8000/8290 | Cost: 0.0172\n",
|
|
"Time elapsed: 22.58 min\n",
|
|
"Epoch: 025/050 | Batch 0000/8290 | Cost: 0.0195\n",
|
|
"Epoch: 025/050 | Batch 0500/8290 | Cost: 0.0203\n",
|
|
"Epoch: 025/050 | Batch 1000/8290 | Cost: 0.0179\n",
|
|
"Epoch: 025/050 | Batch 1500/8290 | Cost: 0.0182\n",
|
|
"Epoch: 025/050 | Batch 2000/8290 | Cost: 0.0195\n",
|
|
"Epoch: 025/050 | Batch 2500/8290 | Cost: 0.0181\n",
|
|
"Epoch: 025/050 | Batch 3000/8290 | Cost: 0.0191\n",
|
|
"Epoch: 025/050 | Batch 3500/8290 | Cost: 0.0181\n",
|
|
"Epoch: 025/050 | Batch 4000/8290 | Cost: 0.0189\n",
|
|
"Epoch: 025/050 | Batch 4500/8290 | Cost: 0.0199\n",
|
|
"Epoch: 025/050 | Batch 5000/8290 | Cost: 0.0211\n",
|
|
"Epoch: 025/050 | Batch 5500/8290 | Cost: 0.0176\n",
|
|
"Epoch: 025/050 | Batch 6000/8290 | Cost: 0.0179\n",
|
|
"Epoch: 025/050 | Batch 6500/8290 | Cost: 0.0186\n",
|
|
"Epoch: 025/050 | Batch 7000/8290 | Cost: 0.0182\n",
|
|
"Epoch: 025/050 | Batch 7500/8290 | Cost: 0.0174\n",
|
|
"Epoch: 025/050 | Batch 8000/8290 | Cost: 0.0187\n",
|
|
"Time elapsed: 23.52 min\n",
|
|
"Epoch: 026/050 | Batch 0000/8290 | Cost: 0.0196\n",
|
|
"Epoch: 026/050 | Batch 0500/8290 | Cost: 0.0174\n",
|
|
"Epoch: 026/050 | Batch 1000/8290 | Cost: 0.0183\n",
|
|
"Epoch: 026/050 | Batch 1500/8290 | Cost: 0.0193\n",
|
|
"Epoch: 026/050 | Batch 2000/8290 | Cost: 0.0183\n",
|
|
"Epoch: 026/050 | Batch 2500/8290 | Cost: 0.0173\n",
|
|
"Epoch: 026/050 | Batch 3000/8290 | Cost: 0.0190\n",
|
|
"Epoch: 026/050 | Batch 3500/8290 | Cost: 0.0181\n",
|
|
"Epoch: 026/050 | Batch 4000/8290 | Cost: 0.0190\n",
|
|
"Epoch: 026/050 | Batch 4500/8290 | Cost: 0.0193\n",
|
|
"Epoch: 026/050 | Batch 5000/8290 | Cost: 0.0194\n",
|
|
"Epoch: 026/050 | Batch 5500/8290 | Cost: 0.0196\n",
|
|
"Epoch: 026/050 | Batch 6000/8290 | Cost: 0.0193\n",
|
|
"Epoch: 026/050 | Batch 6500/8290 | Cost: 0.0187\n",
|
|
"Epoch: 026/050 | Batch 7000/8290 | Cost: 0.0175\n",
|
|
"Epoch: 026/050 | Batch 7500/8290 | Cost: 0.0176\n",
|
|
"Epoch: 026/050 | Batch 8000/8290 | Cost: 0.0177\n",
|
|
"Time elapsed: 24.45 min\n",
|
|
"Epoch: 027/050 | Batch 0000/8290 | Cost: 0.0190\n",
|
|
"Epoch: 027/050 | Batch 0500/8290 | Cost: 0.0180\n",
|
|
"Epoch: 027/050 | Batch 1000/8290 | Cost: 0.0183\n",
|
|
"Epoch: 027/050 | Batch 1500/8290 | Cost: 0.0182\n",
|
|
"Epoch: 027/050 | Batch 2000/8290 | Cost: 0.0194\n",
|
|
"Epoch: 027/050 | Batch 2500/8290 | Cost: 0.0192\n",
|
|
"Epoch: 027/050 | Batch 3000/8290 | Cost: 0.0196\n",
|
|
"Epoch: 027/050 | Batch 3500/8290 | Cost: 0.0183\n",
|
|
"Epoch: 027/050 | Batch 4000/8290 | Cost: 0.0191\n",
|
|
"Epoch: 027/050 | Batch 4500/8290 | Cost: 0.0177\n",
|
|
"Epoch: 027/050 | Batch 5000/8290 | Cost: 0.0178\n",
|
|
"Epoch: 027/050 | Batch 5500/8290 | Cost: 0.0198\n",
|
|
"Epoch: 027/050 | Batch 6000/8290 | Cost: 0.0205\n",
|
|
"Epoch: 027/050 | Batch 6500/8290 | Cost: 0.0192\n",
|
|
"Epoch: 027/050 | Batch 7000/8290 | Cost: 0.0190\n",
|
|
"Epoch: 027/050 | Batch 7500/8290 | Cost: 0.0194\n",
|
|
"Epoch: 027/050 | Batch 8000/8290 | Cost: 0.0176\n",
|
|
"Time elapsed: 25.39 min\n",
|
|
"Epoch: 028/050 | Batch 0000/8290 | Cost: 0.0173\n",
|
|
"Epoch: 028/050 | Batch 0500/8290 | Cost: 0.0178\n",
|
|
"Epoch: 028/050 | Batch 1000/8290 | Cost: 0.0179\n",
|
|
"Epoch: 028/050 | Batch 1500/8290 | Cost: 0.0169\n",
|
|
"Epoch: 028/050 | Batch 2000/8290 | Cost: 0.0173\n",
|
|
"Epoch: 028/050 | Batch 2500/8290 | Cost: 0.0184\n",
|
|
"Epoch: 028/050 | Batch 3000/8290 | Cost: 0.0185\n",
|
|
"Epoch: 028/050 | Batch 3500/8290 | Cost: 0.0183\n",
|
|
"Epoch: 028/050 | Batch 4000/8290 | Cost: 0.0183\n",
|
|
"Epoch: 028/050 | Batch 4500/8290 | Cost: 0.0174\n",
|
|
"Epoch: 028/050 | Batch 5000/8290 | Cost: 0.0191\n",
|
|
"Epoch: 028/050 | Batch 5500/8290 | Cost: 0.0187\n",
|
|
"Epoch: 028/050 | Batch 6000/8290 | Cost: 0.0171\n",
|
|
"Epoch: 028/050 | Batch 6500/8290 | Cost: 0.0179\n",
|
|
"Epoch: 028/050 | Batch 7000/8290 | Cost: 0.0200\n",
|
|
"Epoch: 028/050 | Batch 7500/8290 | Cost: 0.0189\n",
|
|
"Epoch: 028/050 | Batch 8000/8290 | Cost: 0.0198\n",
|
|
"Time elapsed: 26.33 min\n",
|
|
"Epoch: 029/050 | Batch 0000/8290 | Cost: 0.0195\n",
|
|
"Epoch: 029/050 | Batch 0500/8290 | Cost: 0.0179\n",
|
|
"Epoch: 029/050 | Batch 1000/8290 | Cost: 0.0185\n",
|
|
"Epoch: 029/050 | Batch 1500/8290 | Cost: 0.0184\n",
|
|
"Epoch: 029/050 | Batch 2000/8290 | Cost: 0.0191\n",
|
|
"Epoch: 029/050 | Batch 2500/8290 | Cost: 0.0190\n",
|
|
"Epoch: 029/050 | Batch 3000/8290 | Cost: 0.0183\n",
|
|
"Epoch: 029/050 | Batch 3500/8290 | Cost: 0.0188\n",
|
|
"Epoch: 029/050 | Batch 4000/8290 | Cost: 0.0183\n",
|
|
"Epoch: 029/050 | Batch 4500/8290 | Cost: 0.0181\n",
|
|
"Epoch: 029/050 | Batch 5000/8290 | Cost: 0.0195\n",
|
|
"Epoch: 029/050 | Batch 5500/8290 | Cost: 0.0176\n",
|
|
"Epoch: 029/050 | Batch 6000/8290 | Cost: 0.0183\n",
|
|
"Epoch: 029/050 | Batch 6500/8290 | Cost: 0.0169\n",
|
|
"Epoch: 029/050 | Batch 7000/8290 | Cost: 0.0183\n",
|
|
"Epoch: 029/050 | Batch 7500/8290 | Cost: 0.0181\n",
|
|
"Epoch: 029/050 | Batch 8000/8290 | Cost: 0.0192\n",
|
|
"Time elapsed: 27.27 min\n",
|
|
"Epoch: 030/050 | Batch 0000/8290 | Cost: 0.0165\n",
|
|
"Epoch: 030/050 | Batch 0500/8290 | Cost: 0.0184\n",
|
|
"Epoch: 030/050 | Batch 1000/8290 | Cost: 0.0184\n",
|
|
"Epoch: 030/050 | Batch 1500/8290 | Cost: 0.0187\n",
|
|
"Epoch: 030/050 | Batch 2000/8290 | Cost: 0.0178\n",
|
|
"Epoch: 030/050 | Batch 2500/8290 | Cost: 0.0185\n",
|
|
"Epoch: 030/050 | Batch 3000/8290 | Cost: 0.0199\n",
|
|
"Epoch: 030/050 | Batch 3500/8290 | Cost: 0.0181\n",
|
|
"Epoch: 030/050 | Batch 4000/8290 | Cost: 0.0192\n",
|
|
"Epoch: 030/050 | Batch 4500/8290 | Cost: 0.0199\n",
|
|
"Epoch: 030/050 | Batch 5000/8290 | Cost: 0.0198\n",
|
|
"Epoch: 030/050 | Batch 5500/8290 | Cost: 0.0184\n",
|
|
"Epoch: 030/050 | Batch 6000/8290 | Cost: 0.0170\n",
|
|
"Epoch: 030/050 | Batch 6500/8290 | Cost: 0.0184\n",
|
|
"Epoch: 030/050 | Batch 7000/8290 | Cost: 0.0182\n",
|
|
"Epoch: 030/050 | Batch 7500/8290 | Cost: 0.0184\n",
|
|
"Epoch: 030/050 | Batch 8000/8290 | Cost: 0.0175\n",
|
|
"Time elapsed: 28.21 min\n",
|
|
"Epoch: 031/050 | Batch 0000/8290 | Cost: 0.0193\n",
|
|
"Epoch: 031/050 | Batch 0500/8290 | Cost: 0.0169\n",
|
|
"Epoch: 031/050 | Batch 1000/8290 | Cost: 0.0184\n",
|
|
"Epoch: 031/050 | Batch 1500/8290 | Cost: 0.0187\n",
|
|
"Epoch: 031/050 | Batch 2000/8290 | Cost: 0.0184\n",
|
|
"Epoch: 031/050 | Batch 2500/8290 | Cost: 0.0158\n",
|
|
"Epoch: 031/050 | Batch 3000/8290 | Cost: 0.0178\n",
|
|
"Epoch: 031/050 | Batch 3500/8290 | Cost: 0.0203\n",
|
|
"Epoch: 031/050 | Batch 4000/8290 | Cost: 0.0188\n",
|
|
"Epoch: 031/050 | Batch 4500/8290 | Cost: 0.0180\n",
|
|
"Epoch: 031/050 | Batch 5000/8290 | Cost: 0.0179\n",
|
|
"Epoch: 031/050 | Batch 5500/8290 | Cost: 0.0192\n",
|
|
"Epoch: 031/050 | Batch 6000/8290 | Cost: 0.0191\n",
|
|
"Epoch: 031/050 | Batch 6500/8290 | Cost: 0.0182\n",
|
|
"Epoch: 031/050 | Batch 7000/8290 | Cost: 0.0199\n",
|
|
"Epoch: 031/050 | Batch 7500/8290 | Cost: 0.0181\n",
|
|
"Epoch: 031/050 | Batch 8000/8290 | Cost: 0.0184\n",
|
|
"Time elapsed: 29.14 min\n",
|
|
"Epoch: 032/050 | Batch 0000/8290 | Cost: 0.0186\n",
|
|
"Epoch: 032/050 | Batch 0500/8290 | Cost: 0.0187\n",
|
|
"Epoch: 032/050 | Batch 1000/8290 | Cost: 0.0193\n",
|
|
"Epoch: 032/050 | Batch 1500/8290 | Cost: 0.0189\n",
|
|
"Epoch: 032/050 | Batch 2000/8290 | Cost: 0.0189\n",
|
|
"Epoch: 032/050 | Batch 2500/8290 | Cost: 0.0200\n",
|
|
"Epoch: 032/050 | Batch 3000/8290 | Cost: 0.0188\n",
|
|
"Epoch: 032/050 | Batch 3500/8290 | Cost: 0.0195\n",
|
|
"Epoch: 032/050 | Batch 4000/8290 | Cost: 0.0190\n",
|
|
"Epoch: 032/050 | Batch 4500/8290 | Cost: 0.0182\n",
|
|
"Epoch: 032/050 | Batch 5000/8290 | Cost: 0.0155\n",
|
|
"Epoch: 032/050 | Batch 5500/8290 | Cost: 0.0218\n",
|
|
"Epoch: 032/050 | Batch 6000/8290 | Cost: 0.0190\n",
|
|
"Epoch: 032/050 | Batch 6500/8290 | Cost: 0.0186\n",
|
|
"Epoch: 032/050 | Batch 7000/8290 | Cost: 0.0176\n",
|
|
"Epoch: 032/050 | Batch 7500/8290 | Cost: 0.0197\n",
|
|
"Epoch: 032/050 | Batch 8000/8290 | Cost: 0.0190\n",
|
|
"Time elapsed: 30.07 min\n",
|
|
"Epoch: 033/050 | Batch 0000/8290 | Cost: 0.0188\n",
|
|
"Epoch: 033/050 | Batch 0500/8290 | Cost: 0.0182\n",
|
|
"Epoch: 033/050 | Batch 1000/8290 | Cost: 0.0189\n",
|
|
"Epoch: 033/050 | Batch 1500/8290 | Cost: 0.0172\n",
|
|
"Epoch: 033/050 | Batch 2000/8290 | Cost: 0.0174\n",
|
|
"Epoch: 033/050 | Batch 2500/8290 | Cost: 0.0203\n",
|
|
"Epoch: 033/050 | Batch 3000/8290 | Cost: 0.0205\n",
|
|
"Epoch: 033/050 | Batch 3500/8290 | Cost: 0.0194\n",
|
|
"Epoch: 033/050 | Batch 4000/8290 | Cost: 0.0177\n",
|
|
"Epoch: 033/050 | Batch 4500/8290 | Cost: 0.0192\n",
|
|
"Epoch: 033/050 | Batch 5000/8290 | Cost: 0.0192\n",
|
|
"Epoch: 033/050 | Batch 5500/8290 | Cost: 0.0196\n",
|
|
"Epoch: 033/050 | Batch 6000/8290 | Cost: 0.0180\n",
|
|
"Epoch: 033/050 | Batch 6500/8290 | Cost: 0.0196\n",
|
|
"Epoch: 033/050 | Batch 7000/8290 | Cost: 0.0186\n",
|
|
"Epoch: 033/050 | Batch 7500/8290 | Cost: 0.0180\n",
|
|
"Epoch: 033/050 | Batch 8000/8290 | Cost: 0.0168\n",
|
|
"Time elapsed: 31.01 min\n",
|
|
"Epoch: 034/050 | Batch 0000/8290 | Cost: 0.0188\n",
|
|
"Epoch: 034/050 | Batch 0500/8290 | Cost: 0.0181\n",
|
|
"Epoch: 034/050 | Batch 1000/8290 | Cost: 0.0186\n",
|
|
"Epoch: 034/050 | Batch 1500/8290 | Cost: 0.0177\n",
|
|
"Epoch: 034/050 | Batch 2000/8290 | Cost: 0.0168\n",
|
|
"Epoch: 034/050 | Batch 2500/8290 | Cost: 0.0187\n",
|
|
"Epoch: 034/050 | Batch 3000/8290 | Cost: 0.0193\n",
|
|
"Epoch: 034/050 | Batch 3500/8290 | Cost: 0.0191\n",
|
|
"Epoch: 034/050 | Batch 4000/8290 | Cost: 0.0175\n",
|
|
"Epoch: 034/050 | Batch 4500/8290 | Cost: 0.0169\n",
|
|
"Epoch: 034/050 | Batch 5000/8290 | Cost: 0.0172\n",
|
|
"Epoch: 034/050 | Batch 5500/8290 | Cost: 0.0202\n",
|
|
"Epoch: 034/050 | Batch 6000/8290 | Cost: 0.0190\n",
|
|
"Epoch: 034/050 | Batch 6500/8290 | Cost: 0.0175\n",
|
|
"Epoch: 034/050 | Batch 7000/8290 | Cost: 0.0178\n",
|
|
"Epoch: 034/050 | Batch 7500/8290 | Cost: 0.0184\n",
|
|
"Epoch: 034/050 | Batch 8000/8290 | Cost: 0.0176\n",
|
|
"Time elapsed: 31.94 min\n",
|
|
"Epoch: 035/050 | Batch 0000/8290 | Cost: 0.0174\n",
|
|
"Epoch: 035/050 | Batch 0500/8290 | Cost: 0.0176\n",
|
|
"Epoch: 035/050 | Batch 1000/8290 | Cost: 0.0184\n",
|
|
"Epoch: 035/050 | Batch 1500/8290 | Cost: 0.0180\n",
|
|
"Epoch: 035/050 | Batch 2000/8290 | Cost: 0.0208\n",
|
|
"Epoch: 035/050 | Batch 2500/8290 | Cost: 0.0198\n",
|
|
"Epoch: 035/050 | Batch 3000/8290 | Cost: 0.0177\n",
|
|
"Epoch: 035/050 | Batch 3500/8290 | Cost: 0.0174\n",
|
|
"Epoch: 035/050 | Batch 4000/8290 | Cost: 0.0171\n",
|
|
"Epoch: 035/050 | Batch 4500/8290 | Cost: 0.0187\n",
|
|
"Epoch: 035/050 | Batch 5000/8290 | Cost: 0.0175\n",
|
|
"Epoch: 035/050 | Batch 5500/8290 | Cost: 0.0189\n",
|
|
"Epoch: 035/050 | Batch 6000/8290 | Cost: 0.0191\n",
|
|
"Epoch: 035/050 | Batch 6500/8290 | Cost: 0.0177\n",
|
|
"Epoch: 035/050 | Batch 7000/8290 | Cost: 0.0158\n",
|
|
"Epoch: 035/050 | Batch 7500/8290 | Cost: 0.0175\n",
|
|
"Epoch: 035/050 | Batch 8000/8290 | Cost: 0.0190\n",
|
|
"Time elapsed: 32.87 min\n",
|
|
"Epoch: 036/050 | Batch 0000/8290 | Cost: 0.0170\n",
|
|
"Epoch: 036/050 | Batch 0500/8290 | Cost: 0.0188\n",
|
|
"Epoch: 036/050 | Batch 1000/8290 | Cost: 0.0172\n",
|
|
"Epoch: 036/050 | Batch 1500/8290 | Cost: 0.0193\n",
|
|
"Epoch: 036/050 | Batch 2000/8290 | Cost: 0.0168\n",
|
|
"Epoch: 036/050 | Batch 2500/8290 | Cost: 0.0194\n",
|
|
"Epoch: 036/050 | Batch 3000/8290 | Cost: 0.0190\n",
|
|
"Epoch: 036/050 | Batch 3500/8290 | Cost: 0.0189\n",
|
|
"Epoch: 036/050 | Batch 4000/8290 | Cost: 0.0191\n",
|
|
"Epoch: 036/050 | Batch 4500/8290 | Cost: 0.0180\n",
|
|
"Epoch: 036/050 | Batch 5000/8290 | Cost: 0.0202\n",
|
|
"Epoch: 036/050 | Batch 5500/8290 | Cost: 0.0179\n",
|
|
"Epoch: 036/050 | Batch 6000/8290 | Cost: 0.0200\n",
|
|
"Epoch: 036/050 | Batch 6500/8290 | Cost: 0.0175\n",
|
|
"Epoch: 036/050 | Batch 7000/8290 | Cost: 0.0194\n",
|
|
"Epoch: 036/050 | Batch 7500/8290 | Cost: 0.0190\n",
|
|
"Epoch: 036/050 | Batch 8000/8290 | Cost: 0.0189\n",
|
|
"Time elapsed: 33.80 min\n",
|
|
"Epoch: 037/050 | Batch 0000/8290 | Cost: 0.0180\n",
|
|
"Epoch: 037/050 | Batch 0500/8290 | Cost: 0.0181\n",
|
|
"Epoch: 037/050 | Batch 1000/8290 | Cost: 0.0194\n",
|
|
"Epoch: 037/050 | Batch 1500/8290 | Cost: 0.0185\n",
|
|
"Epoch: 037/050 | Batch 2000/8290 | Cost: 0.0190\n",
|
|
"Epoch: 037/050 | Batch 2500/8290 | Cost: 0.0195\n",
|
|
"Epoch: 037/050 | Batch 3000/8290 | Cost: 0.0197\n",
|
|
"Epoch: 037/050 | Batch 3500/8290 | Cost: 0.0174\n",
|
|
"Epoch: 037/050 | Batch 4000/8290 | Cost: 0.0196\n",
|
|
"Epoch: 037/050 | Batch 4500/8290 | Cost: 0.0182\n",
|
|
"Epoch: 037/050 | Batch 5000/8290 | Cost: 0.0174\n",
|
|
"Epoch: 037/050 | Batch 5500/8290 | Cost: 0.0175\n",
|
|
"Epoch: 037/050 | Batch 6000/8290 | Cost: 0.0167\n",
|
|
"Epoch: 037/050 | Batch 6500/8290 | Cost: 0.0161\n",
|
|
"Epoch: 037/050 | Batch 7000/8290 | Cost: 0.0162\n",
|
|
"Epoch: 037/050 | Batch 7500/8290 | Cost: 0.0179\n",
|
|
"Epoch: 037/050 | Batch 8000/8290 | Cost: 0.0196\n",
|
|
"Time elapsed: 34.73 min\n",
|
|
"Epoch: 038/050 | Batch 0000/8290 | Cost: 0.0174\n",
|
|
"Epoch: 038/050 | Batch 0500/8290 | Cost: 0.0184\n",
|
|
"Epoch: 038/050 | Batch 1000/8290 | Cost: 0.0178\n",
|
|
"Epoch: 038/050 | Batch 1500/8290 | Cost: 0.0190\n",
|
|
"Epoch: 038/050 | Batch 2000/8290 | Cost: 0.0182\n",
|
|
"Epoch: 038/050 | Batch 2500/8290 | Cost: 0.0168\n",
|
|
"Epoch: 038/050 | Batch 3000/8290 | Cost: 0.0185\n",
|
|
"Epoch: 038/050 | Batch 3500/8290 | Cost: 0.0178\n",
|
|
"Epoch: 038/050 | Batch 4000/8290 | Cost: 0.0194\n",
|
|
"Epoch: 038/050 | Batch 4500/8290 | Cost: 0.0178\n",
|
|
"Epoch: 038/050 | Batch 5000/8290 | Cost: 0.0193\n",
|
|
"Epoch: 038/050 | Batch 5500/8290 | Cost: 0.0192\n",
|
|
"Epoch: 038/050 | Batch 6000/8290 | Cost: 0.0182\n",
|
|
"Epoch: 038/050 | Batch 6500/8290 | Cost: 0.0182\n",
|
|
"Epoch: 038/050 | Batch 7000/8290 | Cost: 0.0184\n",
|
|
"Epoch: 038/050 | Batch 7500/8290 | Cost: 0.0189\n",
|
|
"Epoch: 038/050 | Batch 8000/8290 | Cost: 0.0176\n",
|
|
"Time elapsed: 35.66 min\n",
|
|
"Epoch: 039/050 | Batch 0000/8290 | Cost: 0.0190\n",
|
|
"Epoch: 039/050 | Batch 0500/8290 | Cost: 0.0171\n",
|
|
"Epoch: 039/050 | Batch 1000/8290 | Cost: 0.0185\n",
|
|
"Epoch: 039/050 | Batch 1500/8290 | Cost: 0.0180\n",
|
|
"Epoch: 039/050 | Batch 2000/8290 | Cost: 0.0193\n",
|
|
"Epoch: 039/050 | Batch 2500/8290 | Cost: 0.0196\n",
|
|
"Epoch: 039/050 | Batch 3000/8290 | Cost: 0.0176\n",
|
|
"Epoch: 039/050 | Batch 3500/8290 | Cost: 0.0185\n",
|
|
"Epoch: 039/050 | Batch 4000/8290 | Cost: 0.0176\n",
|
|
"Epoch: 039/050 | Batch 4500/8290 | Cost: 0.0181\n",
|
|
"Epoch: 039/050 | Batch 5000/8290 | Cost: 0.0189\n",
|
|
"Epoch: 039/050 | Batch 5500/8290 | Cost: 0.0176\n",
|
|
"Epoch: 039/050 | Batch 6000/8290 | Cost: 0.0183\n",
|
|
"Epoch: 039/050 | Batch 6500/8290 | Cost: 0.0191\n",
|
|
"Epoch: 039/050 | Batch 7000/8290 | Cost: 0.0177\n",
|
|
"Epoch: 039/050 | Batch 7500/8290 | Cost: 0.0179\n",
|
|
"Epoch: 039/050 | Batch 8000/8290 | Cost: 0.0168\n",
|
|
"Time elapsed: 36.60 min\n",
|
|
"Epoch: 040/050 | Batch 0000/8290 | Cost: 0.0181\n",
|
|
"Epoch: 040/050 | Batch 0500/8290 | Cost: 0.0184\n",
|
|
"Epoch: 040/050 | Batch 1000/8290 | Cost: 0.0177\n",
|
|
"Epoch: 040/050 | Batch 1500/8290 | Cost: 0.0194\n",
|
|
"Epoch: 040/050 | Batch 2000/8290 | Cost: 0.0176\n",
|
|
"Epoch: 040/050 | Batch 2500/8290 | Cost: 0.0197\n",
|
|
"Epoch: 040/050 | Batch 3000/8290 | Cost: 0.0184\n",
|
|
"Epoch: 040/050 | Batch 3500/8290 | Cost: 0.0170\n",
|
|
"Epoch: 040/050 | Batch 4000/8290 | Cost: 0.0182\n",
|
|
"Epoch: 040/050 | Batch 4500/8290 | Cost: 0.0175\n",
|
|
"Epoch: 040/050 | Batch 5000/8290 | Cost: 0.0194\n",
|
|
"Epoch: 040/050 | Batch 5500/8290 | Cost: 0.0166\n",
|
|
"Epoch: 040/050 | Batch 6000/8290 | Cost: 0.0177\n",
|
|
"Epoch: 040/050 | Batch 6500/8290 | Cost: 0.0177\n",
|
|
"Epoch: 040/050 | Batch 7000/8290 | Cost: 0.0179\n",
|
|
"Epoch: 040/050 | Batch 7500/8290 | Cost: 0.0208\n",
|
|
"Epoch: 040/050 | Batch 8000/8290 | Cost: 0.0184\n",
|
|
"Time elapsed: 37.53 min\n",
|
|
"Epoch: 041/050 | Batch 0000/8290 | Cost: 0.0196\n",
|
|
"Epoch: 041/050 | Batch 0500/8290 | Cost: 0.0179\n",
|
|
"Epoch: 041/050 | Batch 1000/8290 | Cost: 0.0183\n",
|
|
"Epoch: 041/050 | Batch 1500/8290 | Cost: 0.0169\n",
|
|
"Epoch: 041/050 | Batch 2000/8290 | Cost: 0.0189\n",
|
|
"Epoch: 041/050 | Batch 2500/8290 | Cost: 0.0187\n",
|
|
"Epoch: 041/050 | Batch 3000/8290 | Cost: 0.0177\n",
|
|
"Epoch: 041/050 | Batch 3500/8290 | Cost: 0.0182\n",
|
|
"Epoch: 041/050 | Batch 4000/8290 | Cost: 0.0173\n",
|
|
"Epoch: 041/050 | Batch 4500/8290 | Cost: 0.0185\n",
|
|
"Epoch: 041/050 | Batch 5000/8290 | Cost: 0.0168\n",
|
|
"Epoch: 041/050 | Batch 5500/8290 | Cost: 0.0196\n",
|
|
"Epoch: 041/050 | Batch 6000/8290 | Cost: 0.0179\n",
|
|
"Epoch: 041/050 | Batch 6500/8290 | Cost: 0.0171\n",
|
|
"Epoch: 041/050 | Batch 7000/8290 | Cost: 0.0181\n",
|
|
"Epoch: 041/050 | Batch 7500/8290 | Cost: 0.0184\n",
|
|
"Epoch: 041/050 | Batch 8000/8290 | Cost: 0.0187\n",
|
|
"Time elapsed: 38.47 min\n",
|
|
"Epoch: 042/050 | Batch 0000/8290 | Cost: 0.0185\n",
|
|
"Epoch: 042/050 | Batch 0500/8290 | Cost: 0.0175\n",
|
|
"Epoch: 042/050 | Batch 1000/8290 | Cost: 0.0188\n",
|
|
"Epoch: 042/050 | Batch 1500/8290 | Cost: 0.0196\n",
|
|
"Epoch: 042/050 | Batch 2000/8290 | Cost: 0.0185\n",
|
|
"Epoch: 042/050 | Batch 2500/8290 | Cost: 0.0189\n",
|
|
"Epoch: 042/050 | Batch 3000/8290 | Cost: 0.0193\n",
|
|
"Epoch: 042/050 | Batch 3500/8290 | Cost: 0.0185\n",
|
|
"Epoch: 042/050 | Batch 4000/8290 | Cost: 0.0179\n",
|
|
"Epoch: 042/050 | Batch 4500/8290 | Cost: 0.0185\n",
|
|
"Epoch: 042/050 | Batch 5000/8290 | Cost: 0.0170\n",
|
|
"Epoch: 042/050 | Batch 5500/8290 | Cost: 0.0187\n",
|
|
"Epoch: 042/050 | Batch 6000/8290 | Cost: 0.0184\n",
|
|
"Epoch: 042/050 | Batch 6500/8290 | Cost: 0.0185\n",
|
|
"Epoch: 042/050 | Batch 7000/8290 | Cost: 0.0156\n",
|
|
"Epoch: 042/050 | Batch 7500/8290 | Cost: 0.0183\n",
|
|
"Epoch: 042/050 | Batch 8000/8290 | Cost: 0.0186\n",
|
|
"Time elapsed: 39.41 min\n",
|
|
"Epoch: 043/050 | Batch 0000/8290 | Cost: 0.0179\n",
|
|
"Epoch: 043/050 | Batch 0500/8290 | Cost: 0.0180\n",
|
|
"Epoch: 043/050 | Batch 1000/8290 | Cost: 0.0189\n",
|
|
"Epoch: 043/050 | Batch 1500/8290 | Cost: 0.0182\n",
|
|
"Epoch: 043/050 | Batch 2000/8290 | Cost: 0.0167\n",
|
|
"Epoch: 043/050 | Batch 2500/8290 | Cost: 0.0178\n",
|
|
"Epoch: 043/050 | Batch 3000/8290 | Cost: 0.0180\n",
|
|
"Epoch: 043/050 | Batch 3500/8290 | Cost: 0.0175\n",
|
|
"Epoch: 043/050 | Batch 4000/8290 | Cost: 0.0177\n",
|
|
"Epoch: 043/050 | Batch 4500/8290 | Cost: 0.0180\n",
|
|
"Epoch: 043/050 | Batch 5000/8290 | Cost: 0.0174\n",
|
|
"Epoch: 043/050 | Batch 5500/8290 | Cost: 0.0200\n",
|
|
"Epoch: 043/050 | Batch 6000/8290 | Cost: 0.0168\n",
|
|
"Epoch: 043/050 | Batch 6500/8290 | Cost: 0.0186\n",
|
|
"Epoch: 043/050 | Batch 7000/8290 | Cost: 0.0186\n",
|
|
"Epoch: 043/050 | Batch 7500/8290 | Cost: 0.0202\n",
|
|
"Epoch: 043/050 | Batch 8000/8290 | Cost: 0.0189\n",
|
|
"Time elapsed: 40.35 min\n",
|
|
"Epoch: 044/050 | Batch 0000/8290 | Cost: 0.0166\n",
|
|
"Epoch: 044/050 | Batch 0500/8290 | Cost: 0.0181\n",
|
|
"Epoch: 044/050 | Batch 1000/8290 | Cost: 0.0185\n",
|
|
"Epoch: 044/050 | Batch 1500/8290 | Cost: 0.0186\n",
|
|
"Epoch: 044/050 | Batch 2000/8290 | Cost: 0.0168\n",
|
|
"Epoch: 044/050 | Batch 2500/8290 | Cost: 0.0166\n",
|
|
"Epoch: 044/050 | Batch 3000/8290 | Cost: 0.0178\n",
|
|
"Epoch: 044/050 | Batch 3500/8290 | Cost: 0.0176\n",
|
|
"Epoch: 044/050 | Batch 4000/8290 | Cost: 0.0180\n",
|
|
"Epoch: 044/050 | Batch 4500/8290 | Cost: 0.0175\n",
|
|
"Epoch: 044/050 | Batch 5000/8290 | Cost: 0.0186\n",
|
|
"Epoch: 044/050 | Batch 5500/8290 | Cost: 0.0162\n",
|
|
"Epoch: 044/050 | Batch 6000/8290 | Cost: 0.0181\n",
|
|
"Epoch: 044/050 | Batch 6500/8290 | Cost: 0.0183\n",
|
|
"Epoch: 044/050 | Batch 7000/8290 | Cost: 0.0190\n",
|
|
"Epoch: 044/050 | Batch 7500/8290 | Cost: 0.0175\n",
|
|
"Epoch: 044/050 | Batch 8000/8290 | Cost: 0.0174\n",
|
|
"Time elapsed: 41.30 min\n",
|
|
"Epoch: 045/050 | Batch 0000/8290 | Cost: 0.0166\n",
|
|
"Epoch: 045/050 | Batch 0500/8290 | Cost: 0.0186\n",
|
|
"Epoch: 045/050 | Batch 1000/8290 | Cost: 0.0186\n",
|
|
"Epoch: 045/050 | Batch 1500/8290 | Cost: 0.0176\n",
|
|
"Epoch: 045/050 | Batch 2000/8290 | Cost: 0.0182\n",
|
|
"Epoch: 045/050 | Batch 2500/8290 | Cost: 0.0188\n",
|
|
"Epoch: 045/050 | Batch 3000/8290 | Cost: 0.0171\n",
|
|
"Epoch: 045/050 | Batch 3500/8290 | Cost: 0.0171\n",
|
|
"Epoch: 045/050 | Batch 4000/8290 | Cost: 0.0187\n",
|
|
"Epoch: 045/050 | Batch 4500/8290 | Cost: 0.0172\n",
|
|
"Epoch: 045/050 | Batch 5000/8290 | Cost: 0.0181\n",
|
|
"Epoch: 045/050 | Batch 5500/8290 | Cost: 0.0183\n",
|
|
"Epoch: 045/050 | Batch 6000/8290 | Cost: 0.0183\n",
|
|
"Epoch: 045/050 | Batch 6500/8290 | Cost: 0.0190\n",
|
|
"Epoch: 045/050 | Batch 7000/8290 | Cost: 0.0177\n",
|
|
"Epoch: 045/050 | Batch 7500/8290 | Cost: 0.0164\n",
|
|
"Epoch: 045/050 | Batch 8000/8290 | Cost: 0.0183\n",
|
|
"Time elapsed: 42.24 min\n",
|
|
"Epoch: 046/050 | Batch 0000/8290 | Cost: 0.0171\n",
|
|
"Epoch: 046/050 | Batch 0500/8290 | Cost: 0.0174\n",
|
|
"Epoch: 046/050 | Batch 1000/8290 | Cost: 0.0176\n",
|
|
"Epoch: 046/050 | Batch 1500/8290 | Cost: 0.0176\n",
|
|
"Epoch: 046/050 | Batch 2000/8290 | Cost: 0.0174\n",
|
|
"Epoch: 046/050 | Batch 2500/8290 | Cost: 0.0194\n",
|
|
"Epoch: 046/050 | Batch 3000/8290 | Cost: 0.0175\n",
|
|
"Epoch: 046/050 | Batch 3500/8290 | Cost: 0.0184\n",
|
|
"Epoch: 046/050 | Batch 4000/8290 | Cost: 0.0187\n",
|
|
"Epoch: 046/050 | Batch 4500/8290 | Cost: 0.0177\n",
|
|
"Epoch: 046/050 | Batch 5000/8290 | Cost: 0.0184\n",
|
|
"Epoch: 046/050 | Batch 5500/8290 | Cost: 0.0190\n",
|
|
"Epoch: 046/050 | Batch 6000/8290 | Cost: 0.0171\n",
|
|
"Epoch: 046/050 | Batch 6500/8290 | Cost: 0.0184\n",
|
|
"Epoch: 046/050 | Batch 7000/8290 | Cost: 0.0167\n",
|
|
"Epoch: 046/050 | Batch 7500/8290 | Cost: 0.0185\n",
|
|
"Epoch: 046/050 | Batch 8000/8290 | Cost: 0.0188\n",
|
|
"Time elapsed: 43.18 min\n",
|
|
"Epoch: 047/050 | Batch 0000/8290 | Cost: 0.0184\n",
|
|
"Epoch: 047/050 | Batch 0500/8290 | Cost: 0.0167\n",
|
|
"Epoch: 047/050 | Batch 1000/8290 | Cost: 0.0169\n",
|
|
"Epoch: 047/050 | Batch 1500/8290 | Cost: 0.0182\n",
|
|
"Epoch: 047/050 | Batch 2000/8290 | Cost: 0.0191\n",
|
|
"Epoch: 047/050 | Batch 2500/8290 | Cost: 0.0172\n",
|
|
"Epoch: 047/050 | Batch 3000/8290 | Cost: 0.0169\n",
|
|
"Epoch: 047/050 | Batch 3500/8290 | Cost: 0.0183\n",
|
|
"Epoch: 047/050 | Batch 4000/8290 | Cost: 0.0173\n",
|
|
"Epoch: 047/050 | Batch 4500/8290 | Cost: 0.0163\n",
|
|
"Epoch: 047/050 | Batch 5000/8290 | Cost: 0.0165\n",
|
|
"Epoch: 047/050 | Batch 5500/8290 | Cost: 0.0184\n",
|
|
"Epoch: 047/050 | Batch 6000/8290 | Cost: 0.0163\n",
|
|
"Epoch: 047/050 | Batch 6500/8290 | Cost: 0.0181\n",
|
|
"Epoch: 047/050 | Batch 7000/8290 | Cost: 0.0184\n",
|
|
"Epoch: 047/050 | Batch 7500/8290 | Cost: 0.0181\n",
|
|
"Epoch: 047/050 | Batch 8000/8290 | Cost: 0.0180\n",
|
|
"Time elapsed: 44.11 min\n",
|
|
"Epoch: 048/050 | Batch 0000/8290 | Cost: 0.0173\n",
|
|
"Epoch: 048/050 | Batch 0500/8290 | Cost: 0.0166\n",
|
|
"Epoch: 048/050 | Batch 1000/8290 | Cost: 0.0177\n",
|
|
"Epoch: 048/050 | Batch 1500/8290 | Cost: 0.0174\n",
|
|
"Epoch: 048/050 | Batch 2000/8290 | Cost: 0.0188\n",
|
|
"Epoch: 048/050 | Batch 2500/8290 | Cost: 0.0192\n",
|
|
"Epoch: 048/050 | Batch 3000/8290 | Cost: 0.0170\n",
|
|
"Epoch: 048/050 | Batch 3500/8290 | Cost: 0.0170\n",
|
|
"Epoch: 048/050 | Batch 4000/8290 | Cost: 0.0194\n",
|
|
"Epoch: 048/050 | Batch 4500/8290 | Cost: 0.0185\n",
|
|
"Epoch: 048/050 | Batch 5000/8290 | Cost: 0.0180\n",
|
|
"Epoch: 048/050 | Batch 5500/8290 | Cost: 0.0187\n",
|
|
"Epoch: 048/050 | Batch 6000/8290 | Cost: 0.0168\n",
|
|
"Epoch: 048/050 | Batch 6500/8290 | Cost: 0.0199\n",
|
|
"Epoch: 048/050 | Batch 7000/8290 | Cost: 0.0174\n",
|
|
"Epoch: 048/050 | Batch 7500/8290 | Cost: 0.0189\n",
|
|
"Epoch: 048/050 | Batch 8000/8290 | Cost: 0.0175\n",
|
|
"Time elapsed: 45.03 min\n",
|
|
"Epoch: 049/050 | Batch 0000/8290 | Cost: 0.0185\n",
|
|
"Epoch: 049/050 | Batch 0500/8290 | Cost: 0.0185\n",
|
|
"Epoch: 049/050 | Batch 1000/8290 | Cost: 0.0180\n",
|
|
"Epoch: 049/050 | Batch 1500/8290 | Cost: 0.0173\n",
|
|
"Epoch: 049/050 | Batch 2000/8290 | Cost: 0.0163\n",
|
|
"Epoch: 049/050 | Batch 2500/8290 | Cost: 0.0191\n",
|
|
"Epoch: 049/050 | Batch 3000/8290 | Cost: 0.0177\n",
|
|
"Epoch: 049/050 | Batch 3500/8290 | Cost: 0.0161\n",
|
|
"Epoch: 049/050 | Batch 4000/8290 | Cost: 0.0180\n",
|
|
"Epoch: 049/050 | Batch 4500/8290 | Cost: 0.0179\n",
|
|
"Epoch: 049/050 | Batch 5000/8290 | Cost: 0.0173\n",
|
|
"Epoch: 049/050 | Batch 5500/8290 | Cost: 0.0190\n",
|
|
"Epoch: 049/050 | Batch 6000/8290 | Cost: 0.0165\n",
|
|
"Epoch: 049/050 | Batch 6500/8290 | Cost: 0.0186\n",
|
|
"Epoch: 049/050 | Batch 7000/8290 | Cost: 0.0161\n",
|
|
"Epoch: 049/050 | Batch 7500/8290 | Cost: 0.0173\n",
|
|
"Epoch: 049/050 | Batch 8000/8290 | Cost: 0.0178\n",
|
|
"Time elapsed: 45.96 min\n",
|
|
"Epoch: 050/050 | Batch 0000/8290 | Cost: 0.0171\n",
|
|
"Epoch: 050/050 | Batch 0500/8290 | Cost: 0.0168\n",
|
|
"Epoch: 050/050 | Batch 1000/8290 | Cost: 0.0178\n",
|
|
"Epoch: 050/050 | Batch 1500/8290 | Cost: 0.0169\n",
|
|
"Epoch: 050/050 | Batch 2000/8290 | Cost: 0.0172\n",
|
|
"Epoch: 050/050 | Batch 2500/8290 | Cost: 0.0169\n",
|
|
"Epoch: 050/050 | Batch 3000/8290 | Cost: 0.0168\n",
|
|
"Epoch: 050/050 | Batch 3500/8290 | Cost: 0.0155\n",
|
|
"Epoch: 050/050 | Batch 4000/8290 | Cost: 0.0180\n",
|
|
"Epoch: 050/050 | Batch 4500/8290 | Cost: 0.0187\n",
|
|
"Epoch: 050/050 | Batch 5000/8290 | Cost: 0.0189\n",
|
|
"Epoch: 050/050 | Batch 5500/8290 | Cost: 0.0182\n",
|
|
"Epoch: 050/050 | Batch 6000/8290 | Cost: 0.0193\n",
|
|
"Epoch: 050/050 | Batch 6500/8290 | Cost: 0.0189\n",
|
|
"Epoch: 050/050 | Batch 7000/8290 | Cost: 0.0176\n",
|
|
"Epoch: 050/050 | Batch 7500/8290 | Cost: 0.0174\n",
|
|
"Epoch: 050/050 | Batch 8000/8290 | Cost: 0.0180\n",
|
|
"Time elapsed: 46.89 min\n",
|
|
"Total Training Time: 46.89 min\n"
|
|
]
|
|
}
|
|
],
|
|
"source": [
|
|
"##########################\n",
|
|
"### TRAINING\n",
|
|
"##########################\n",
|
|
"\n",
|
|
"epoch_start = 1\n",
|
|
"\n",
|
|
"\n",
|
|
"torch.manual_seed(random_seed)\n",
|
|
"model = Autoencoder()\n",
|
|
"model = model.to(device)\n",
|
|
"\n",
|
|
"optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)\n",
|
|
"\n",
|
|
"\n",
|
|
"################## Load previous\n",
|
|
"# the code saves the autoencoder\n",
|
|
"# after each epoch so that in case\n",
|
|
"# the training process gets interrupted,\n",
|
|
"# we will not have to start training it\n",
|
|
"# from scratch\n",
|
|
"files = os.listdir()\n",
|
|
"\n",
|
|
"start_time = time.time()\n",
|
|
"for epoch in range(epoch_start, num_epochs+1):\n",
|
|
" \n",
|
|
" \n",
|
|
" for batch_idx, (x, y) in enumerate(train_loader):\n",
|
|
"\n",
|
|
" # don't need labels, only the images (features)\n",
|
|
" features = x.to(device)\n",
|
|
" \n",
|
|
" ### FORWARD AND BACK PROP\n",
|
|
" decoded = model(features)\n",
|
|
" cost = F.mse_loss(decoded, features)\n",
|
|
" optimizer.zero_grad()\n",
|
|
" \n",
|
|
" cost.backward()\n",
|
|
" \n",
|
|
" ### UPDATE MODEL PARAMETERS\n",
|
|
" optimizer.step()\n",
|
|
" \n",
|
|
" ### LOGGING\n",
|
|
" if not batch_idx % 500:\n",
|
|
" print ('Epoch: %03d/%03d | Batch %04d/%04d | Cost: %.4f' \n",
|
|
" %(epoch, num_epochs, batch_idx, \n",
|
|
" len(train_loader), cost))\n",
|
|
"\n",
|
|
" \n",
|
|
" print('Time elapsed: %.2f min' % ((time.time() - start_time)/60))\n",
|
|
"print('Total Training Time: %.2f min' % ((time.time() - start_time)/60))\n",
|
|
" \n",
|
|
"# Save model\n",
|
|
"if os.path.isfile('autoencoder_quickdraw-1_i_%d_%s.pt' % (epoch-1, device)):\n",
|
|
" os.remove('autoencoder_quickdraw-1_i_%d_%s.pt' % (epoch-1, device))\n",
|
|
"torch.save(model.state_dict(), 'autoencoder_quickdraw-1_i_%d_%s.pt' % (epoch, device))"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {
|
|
"colab_type": "text",
|
|
"id": "OBO9L5FnVA2h"
|
|
},
|
|
"source": [
|
|
"## Evaluation"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 10,
|
|
"metadata": {
|
|
"colab": {
|
|
"autoexec": {
|
|
"startup": false,
|
|
"wait_interval": 0
|
|
},
|
|
"base_uri": "https://localhost:8080/",
|
|
"height": 323
|
|
},
|
|
"colab_type": "code",
|
|
"executionInfo": {
|
|
"elapsed": 3782,
|
|
"status": "ok",
|
|
"timestamp": 1525044542253,
|
|
"user": {
|
|
"displayName": "Sebastian Raschka",
|
|
"photoUrl": "//lh6.googleusercontent.com/-cxK6yOSQ6uE/AAAAAAAAAAI/AAAAAAAAIfw/P9ar_CHsKOQ/s50-c-k-no/photo.jpg",
|
|
"userId": "118404394130788869227"
|
|
},
|
|
"user_tz": 240
|
|
},
|
|
"id": "UpJLf9FnVqSw",
|
|
"outputId": "121c6c55-6171-4b1b-c6ea-5a199abb4bc5"
|
|
},
|
|
"outputs": [
|
|
{
|
|
"data": {
|
|
"image/png": "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\n",
|
|
"text/plain": [
|
|
"<Figure size 1296x360 with 10 Axes>"
|
|
]
|
|
},
|
|
"metadata": {
|
|
"needs_background": "light"
|
|
},
|
|
"output_type": "display_data"
|
|
}
|
|
],
|
|
"source": [
|
|
"%matplotlib inline\n",
|
|
"import matplotlib.pyplot as plt\n",
|
|
"\n",
|
|
"\n",
|
|
"model = Autoencoder()\n",
|
|
"model = model.to(device)\n",
|
|
"model.load_state_dict(torch.load('autoencoder_quickdraw-1_i_%d_%s.pt' % (num_epochs, device)))\n",
|
|
"model.eval()\n",
|
|
"torch.manual_seed(random_seed)\n",
|
|
"\n",
|
|
"for batch_idx, (x, y) in enumerate(train_loader):\n",
|
|
" features = x.to(device)\n",
|
|
" decoded = model(features)\n",
|
|
" break\n",
|
|
"\n",
|
|
"\n",
|
|
"\n",
|
|
"##########################\n",
|
|
"### VISUALIZATION\n",
|
|
"##########################\n",
|
|
"\n",
|
|
"n_images = 5\n",
|
|
"\n",
|
|
"fig, axes = plt.subplots(nrows=2, ncols=n_images, \n",
|
|
" sharex=True, sharey=True, figsize=(18, 5))\n",
|
|
"orig_images = features.detach().cpu().numpy()[:n_images]\n",
|
|
"orig_images = np.moveaxis(orig_images, 1, -1)\n",
|
|
"\n",
|
|
"decoded_images = decoded.detach().cpu().numpy()[:n_images]\n",
|
|
"decoded_images = np.moveaxis(decoded_images, 1, -1)\n",
|
|
"\n",
|
|
"\n",
|
|
"for i in range(n_images):\n",
|
|
" for ax, img in zip(axes, [orig_images, decoded_images]):\n",
|
|
" ax[i].axis('off')\n",
|
|
" ax[i].imshow(img[i].reshape(28, 28), cmap='binary')"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 11,
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"name": "stdout",
|
|
"output_type": "stream",
|
|
"text": [
|
|
"numpy 1.15.4\n",
|
|
"pandas 0.23.4\n",
|
|
"PIL.Image 5.3.0\n",
|
|
"torch 1.0.0\n",
|
|
"\n"
|
|
]
|
|
}
|
|
],
|
|
"source": [
|
|
"%watermark -iv"
|
|
]
|
|
}
|
|
],
|
|
"metadata": {
|
|
"accelerator": "GPU",
|
|
"colab": {
|
|
"collapsed_sections": [],
|
|
"default_view": {},
|
|
"name": "autoencoder-conv-2.ipynb",
|
|
"provenance": [],
|
|
"version": "0.3.2",
|
|
"views": {}
|
|
},
|
|
"kernelspec": {
|
|
"display_name": "Python 3",
|
|
"language": "python",
|
|
"name": "python3"
|
|
},
|
|
"language_info": {
|
|
"codemirror_mode": {
|
|
"name": "ipython",
|
|
"version": 3
|
|
},
|
|
"file_extension": ".py",
|
|
"mimetype": "text/x-python",
|
|
"name": "python",
|
|
"nbconvert_exporter": "python",
|
|
"pygments_lexer": "ipython3",
|
|
"version": "3.7.1"
|
|
},
|
|
"toc": {
|
|
"nav_menu": {},
|
|
"number_sections": true,
|
|
"sideBar": true,
|
|
"skip_h1_title": false,
|
|
"title_cell": "Table of Contents",
|
|
"title_sidebar": "Contents",
|
|
"toc_cell": false,
|
|
"toc_position": {},
|
|
"toc_section_display": true,
|
|
"toc_window_display": false
|
|
}
|
|
},
|
|
"nbformat": 4,
|
|
"nbformat_minor": 2
|
|
}
|