wehub-resource-sync
226 lines
6.8 KiB
Python
226 lines
6.8 KiB
Python
# Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
|
|
from functools import reduce
|
|
|
|
import numpy as np
|
|
|
|
import paddle
|
|
from paddle import base
|
|
from paddle.common_ops_import import LayerHelper
|
|
|
|
|
|
def pir_fc(hidden, size, activation, param_attr, bias_attr):
|
|
helper = LayerHelper("fc", **locals())
|
|
if not isinstance(hidden, (list, tuple)):
|
|
hidden = [hidden]
|
|
matmul_results = []
|
|
for i, input in enumerate(hidden):
|
|
input_shape = input.shape
|
|
num_flatten_dims = len(input_shape) - 1
|
|
param_shape = [
|
|
reduce(lambda a, b: a * b, input_shape[num_flatten_dims:], 1),
|
|
size,
|
|
]
|
|
|
|
w = helper.create_parameter(
|
|
attr=param_attr, shape=param_shape, dtype=input.dtype, is_bias=False
|
|
)
|
|
out = paddle.matmul(input, w)
|
|
matmul_results.append(out)
|
|
|
|
if len(matmul_results) == 1:
|
|
pre_bias = matmul_results[0]
|
|
else:
|
|
pre_bias = paddle.add_n(matmul_results)
|
|
bias = helper.create_parameter(
|
|
attr=bias_attr,
|
|
shape=pre_bias.shape[-1:],
|
|
dtype=pre_bias.dtype,
|
|
is_bias=True,
|
|
)
|
|
out = paddle.add(pre_bias, bias)
|
|
act_op = getattr(paddle._C_ops, activation)
|
|
if activation == 'softmax':
|
|
return act_op(out, -1)
|
|
return act_op(out)
|
|
|
|
|
|
def pir_simple_fc_net_with_inputs(img, label, class_num=10):
|
|
hidden = img
|
|
param_attr = base.ParamAttr(initializer=paddle.nn.initializer.Uniform())
|
|
bias_attr = base.ParamAttr(
|
|
initializer=paddle.nn.initializer.Constant(value=1.0)
|
|
)
|
|
for _ in range(2):
|
|
hidden = pir_fc(
|
|
hidden,
|
|
size=100,
|
|
activation='relu',
|
|
param_attr=param_attr,
|
|
bias_attr=bias_attr,
|
|
)
|
|
prediction = pir_fc(
|
|
hidden,
|
|
size=class_num,
|
|
activation='softmax',
|
|
param_attr=param_attr,
|
|
bias_attr=bias_attr,
|
|
)
|
|
loss = paddle.nn.functional.softmax_with_cross_entropy(prediction, label)
|
|
|
|
loss = paddle.mean(loss)
|
|
return loss
|
|
|
|
|
|
def pir_batchnorm_fc_with_inputs(img, label, class_num=10):
|
|
hidden = img
|
|
param_attr = base.ParamAttr(initializer=paddle.nn.initializer.Uniform())
|
|
bias_attr = base.ParamAttr(
|
|
initializer=paddle.nn.initializer.Constant(value=1.0)
|
|
)
|
|
for _ in range(2):
|
|
hidden = pir_fc(
|
|
hidden,
|
|
size=200,
|
|
activation='relu',
|
|
param_attr=param_attr,
|
|
bias_attr=bias_attr,
|
|
)
|
|
batch_norm = paddle.nn.BatchNorm(200)
|
|
hidden = batch_norm(hidden)
|
|
|
|
prediction = pir_fc(
|
|
hidden,
|
|
size=class_num,
|
|
activation='softmax',
|
|
param_attr=param_attr,
|
|
bias_attr=bias_attr,
|
|
)
|
|
loss = paddle.nn.functional.softmax_with_cross_entropy(prediction, label)
|
|
loss = paddle.mean(loss)
|
|
return loss
|
|
|
|
|
|
def simple_fc_net_with_inputs(img, label, class_num=10):
|
|
if paddle.framework.in_pir_mode():
|
|
return pir_simple_fc_net_with_inputs(img, label, class_num)
|
|
|
|
hidden = img
|
|
for _ in range(2):
|
|
hidden = paddle.static.nn.fc(
|
|
hidden,
|
|
size=100,
|
|
activation='relu',
|
|
bias_attr=base.ParamAttr(
|
|
initializer=paddle.nn.initializer.Constant(value=1.0)
|
|
),
|
|
)
|
|
prediction = paddle.static.nn.fc(
|
|
hidden, size=class_num, activation='softmax'
|
|
)
|
|
loss = paddle.nn.functional.cross_entropy(
|
|
input=prediction, label=label, reduction='none', use_softmax=False
|
|
)
|
|
loss = paddle.mean(loss)
|
|
return loss
|
|
|
|
|
|
def simple_fc_net(use_feed=None):
|
|
img = paddle.static.data(name='image', shape=[-1, 784], dtype='float32')
|
|
label = paddle.static.data(name='label', shape=[-1, 1], dtype='int64')
|
|
return simple_fc_net_with_inputs(img, label, class_num=10)
|
|
|
|
|
|
def batchnorm_fc_with_inputs(img, label, class_num=10):
|
|
if paddle.framework.in_pir_mode():
|
|
return pir_batchnorm_fc_with_inputs(img, label, class_num)
|
|
|
|
hidden = img
|
|
for _ in range(2):
|
|
hidden = paddle.static.nn.fc(
|
|
hidden,
|
|
size=200,
|
|
activation='relu',
|
|
bias_attr=base.ParamAttr(
|
|
initializer=paddle.nn.initializer.Constant(value=1.0)
|
|
),
|
|
)
|
|
|
|
hidden = paddle.static.nn.batch_norm(input=hidden)
|
|
|
|
prediction = paddle.static.nn.fc(
|
|
hidden, size=class_num, activation='softmax'
|
|
)
|
|
loss = paddle.nn.functional.cross_entropy(
|
|
input=prediction, label=label, reduction='none', use_softmax=False
|
|
)
|
|
loss = paddle.mean(loss)
|
|
return loss
|
|
|
|
|
|
def fc_with_batchnorm(use_feed=None):
|
|
img = paddle.static.data(name='image', shape=[-1, 784], dtype='float32')
|
|
label = paddle.static.data(name='label', shape=[-1, 1], dtype='int64')
|
|
return batchnorm_fc_with_inputs(img, label, class_num=10)
|
|
|
|
|
|
def bow_net(
|
|
use_feed,
|
|
dict_dim,
|
|
is_sparse=False,
|
|
emb_dim=128,
|
|
hid_dim=128,
|
|
hid_dim2=96,
|
|
class_dim=2,
|
|
):
|
|
"""
|
|
BOW net
|
|
This model is from https://github.com/PaddlePaddle/models:
|
|
base/PaddleNLP/text_classification/nets.py
|
|
"""
|
|
data = paddle.static.data(name="words", shape=[-1, 1], dtype="int64")
|
|
label = paddle.static.data(name="label", shape=[-1, 1], dtype="int64")
|
|
emb = paddle.static.nn.embedding(
|
|
input=data, is_sparse=is_sparse, size=[dict_dim, emb_dim]
|
|
)
|
|
bow = paddle.static.nn.sequence_lod.sequence_pool(
|
|
input=emb, pool_type='sum'
|
|
)
|
|
bow_tanh = paddle.tanh(bow)
|
|
fc_1 = paddle.static.nn.fc(x=bow_tanh, size=hid_dim, activation="tanh")
|
|
fc_2 = paddle.static.nn.fc(x=fc_1, size=hid_dim2, activation="tanh")
|
|
prediction = paddle.static.nn.fc(
|
|
x=[fc_2], size=class_dim, activation="softmax"
|
|
)
|
|
cost = paddle.nn.functional.cross_entropy(
|
|
input=prediction, label=label, reduction='none', use_softmax=False
|
|
)
|
|
avg_cost = paddle.mean(x=cost)
|
|
|
|
return avg_cost
|
|
|
|
|
|
def init_data(batch_size=32, img_shape=[784], label_range=9):
|
|
np.random.seed(5)
|
|
assert isinstance(img_shape, list)
|
|
input_shape = [batch_size, *img_shape]
|
|
img = np.random.random(size=input_shape).astype(np.float32)
|
|
label = (
|
|
np.array([np.random.randint(0, label_range) for _ in range(batch_size)])
|
|
.reshape((-1, 1))
|
|
.astype("int64")
|
|
)
|
|
return img, label
|