test_model.py 29.1 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
# copyright (c) 2020 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 __future__ import division
from __future__ import print_function

import unittest

import os
import numpy as np
import shutil
import tempfile

L
Leo Chen 已提交
25
import paddle
26
from paddle import fluid
27
from paddle import to_tensor
C
cnn 已提交
28
from paddle.nn import Conv2D, Linear, ReLU, Sequential, Softmax
29

30 31
from paddle import Model
from paddle.static import InputSpec
32
from paddle.nn.layer.loss import CrossEntropyLoss
33
from paddle.metric import Accuracy
34 35
from paddle.vision.datasets import MNIST
from paddle.vision.models import LeNet
36
from paddle.io import DistributedBatchSampler, Dataset
37
from paddle.hapi.model import prepare_distributed_context
38 39
from paddle.fluid.dygraph.jit import declarative
from paddle.fluid.dygraph.dygraph_to_static.program_translator import ProgramTranslator
40 41


42
class LeNetDygraph(paddle.nn.Layer):
L
LielinJiang 已提交
43
    def __init__(self, num_classes=10):
44 45 46
        super(LeNetDygraph, self).__init__()
        self.num_classes = num_classes
        self.features = Sequential(
C
cnn 已提交
47
            Conv2D(
48
                1, 6, 3, stride=1, padding=1),
L
LielinJiang 已提交
49
            ReLU(),
50
            paddle.fluid.dygraph.Pool2D(2, 'max', 2),
C
cnn 已提交
51
            Conv2D(
52
                6, 16, 5, stride=1, padding=0),
L
LielinJiang 已提交
53
            ReLU(),
54
            paddle.fluid.dygraph.Pool2D(2, 'max', 2))
55 56 57

        if num_classes > 0:
            self.fc = Sequential(
L
LielinJiang 已提交
58
                Linear(400, 120), Linear(120, 84), Linear(84, 10))
59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100

    def forward(self, inputs):
        x = self.features(inputs)

        if self.num_classes > 0:
            x = fluid.layers.flatten(x, 1)
            x = self.fc(x)
        return x


class MnistDataset(MNIST):
    def __init__(self, mode, return_label=True, sample_num=None):
        super(MnistDataset, self).__init__(mode=mode)
        self.return_label = return_label
        if sample_num:
            self.images = self.images[:sample_num]
            self.labels = self.labels[:sample_num]

    def __getitem__(self, idx):
        img, label = self.images[idx], self.labels[idx]
        img = np.reshape(img, [1, 28, 28])
        if self.return_label:
            return img, np.array(self.labels[idx]).astype('int64')
        return img,

    def __len__(self):
        return len(self.images)


def compute_acc(pred, label):
    pred = np.argmax(pred, -1)
    label = np.array(label)
    correct = pred[:, np.newaxis] == label
    return np.sum(correct) / correct.shape[0]


def dynamic_train(model, dataloader):
    optim = fluid.optimizer.Adam(
        learning_rate=0.001, parameter_list=model.parameters())
    model.train()
    for inputs, labels in dataloader:
        outputs = model(inputs)
101
        loss = CrossEntropyLoss(reduction="sum")(outputs, labels)
102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127
        avg_loss = fluid.layers.reduce_sum(loss)
        avg_loss.backward()
        optim.minimize(avg_loss)
        model.clear_gradients()


def dynamic_evaluate(model, dataloader):
    with fluid.dygraph.no_grad():
        model.eval()
        cnt = 0
        for inputs, labels in dataloader:
            outputs = model(inputs)

            cnt += (np.argmax(outputs.numpy(), -1)[:, np.newaxis] ==
                    labels.numpy()).astype('int').sum()

    return cnt / len(dataloader.dataset)


@unittest.skipIf(not fluid.is_compiled_with_cuda(),
                 'CPU testing is not supported')
class TestModel(unittest.TestCase):
    @classmethod
    def setUpClass(cls):
        if not fluid.is_compiled_with_cuda():
            self.skipTest('module not tested when ONLY_CPU compling')
128
        cls.device = paddle.set_device('gpu')
129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144
        fluid.enable_dygraph(cls.device)

        sp_num = 1280
        cls.train_dataset = MnistDataset(mode='train', sample_num=sp_num)
        cls.val_dataset = MnistDataset(mode='test', sample_num=sp_num)
        cls.test_dataset = MnistDataset(
            mode='test', return_label=False, sample_num=sp_num)

        cls.train_loader = fluid.io.DataLoader(
            cls.train_dataset, places=cls.device, batch_size=64)
        cls.val_loader = fluid.io.DataLoader(
            cls.val_dataset, places=cls.device, batch_size=64)
        cls.test_loader = fluid.io.DataLoader(
            cls.test_dataset, places=cls.device, batch_size=64)

        seed = 333
C
cnn 已提交
145
        paddle.seed(seed)
L
Leo Chen 已提交
146
        paddle.framework.random._manual_program_seed(seed)
147 148 149 150 151 152 153

        dy_lenet = LeNetDygraph()
        cls.init_param = dy_lenet.state_dict()
        dynamic_train(dy_lenet, cls.train_loader)

        cls.acc1 = dynamic_evaluate(dy_lenet, cls.val_loader)

154 155
        cls.inputs = [InputSpec([-1, 1, 28, 28], 'float32', 'image')]
        cls.labels = [InputSpec([None, 1], 'int64', 'label')]
156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172

        cls.save_dir = tempfile.mkdtemp()
        cls.weight_path = os.path.join(cls.save_dir, 'lenet')
        fluid.dygraph.save_dygraph(dy_lenet.state_dict(), cls.weight_path)

        fluid.disable_dygraph()

    @classmethod
    def tearDownClass(cls):
        shutil.rmtree(cls.save_dir)

    def test_fit_dygraph(self):
        self.fit(True)

    def test_fit_static(self):
        self.fit(False)

173 174 175 176 177 178
    def test_fit_dynamic_with_rank(self):
        self.fit(True, 2, 0)

    def test_fit_static_with_rank(self):
        self.fit(False, 2, 0)

179 180 181 182 183 184 185 186 187 188 189 190 191 192 193
    def test_evaluate_dygraph(self):
        self.evaluate(True)

    def test_evaluate_static(self):
        self.evaluate(False)

    def test_predict_dygraph(self):
        self.predict(True)

    def test_predict_static(self):
        self.predict(False)

    def test_prepare_context(self):
        prepare_distributed_context()

194
    def fit(self, dynamic, num_replicas=None, rank=None):
195 196
        fluid.enable_dygraph(self.device) if dynamic else None
        seed = 333
C
cnn 已提交
197
        paddle.seed(seed)
L
Leo Chen 已提交
198
        paddle.framework.random._manual_program_seed(seed)
199

L
LielinJiang 已提交
200
        net = LeNet()
201
        optim_new = fluid.optimizer.Adam(
202 203
            learning_rate=0.001, parameter_list=net.parameters())
        model = Model(net, inputs=self.inputs, labels=self.labels)
204 205
        model.prepare(
            optim_new,
206
            loss=CrossEntropyLoss(reduction="sum"),
207
            metrics=Accuracy())
208 209 210 211 212 213
        model.fit(self.train_dataset, batch_size=64, shuffle=False)

        result = model.evaluate(self.val_dataset, batch_size=64)
        np.testing.assert_allclose(result['acc'], self.acc1)

        train_sampler = DistributedBatchSampler(
214 215 216 217 218
            self.train_dataset,
            batch_size=64,
            shuffle=False,
            num_replicas=num_replicas,
            rank=rank)
219
        val_sampler = DistributedBatchSampler(
220 221 222 223 224
            self.val_dataset,
            batch_size=64,
            shuffle=False,
            num_replicas=num_replicas,
            rank=rank)
225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242

        train_loader = fluid.io.DataLoader(
            self.train_dataset,
            batch_sampler=train_sampler,
            places=self.device,
            return_list=True)

        val_loader = fluid.io.DataLoader(
            self.val_dataset,
            batch_sampler=val_sampler,
            places=self.device,
            return_list=True)

        model.fit(train_loader, val_loader)
        fluid.disable_dygraph() if dynamic else None

    def evaluate(self, dynamic):
        fluid.enable_dygraph(self.device) if dynamic else None
243 244
        model = Model(LeNet(), self.inputs, self.labels)
        model.prepare(metrics=Accuracy())
245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263
        model.load(self.weight_path)
        result = model.evaluate(self.val_dataset, batch_size=64)
        np.testing.assert_allclose(result['acc'], self.acc1)

        sampler = DistributedBatchSampler(
            self.val_dataset, batch_size=64, shuffle=False)

        val_loader = fluid.io.DataLoader(
            self.val_dataset,
            batch_sampler=sampler,
            places=self.device,
            return_list=True)

        model.evaluate(val_loader)

        fluid.disable_dygraph() if dynamic else None

    def predict(self, dynamic):
        fluid.enable_dygraph(self.device) if dynamic else None
264 265
        model = Model(LeNet(), self.inputs)
        model.prepare()
266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286
        model.load(self.weight_path)
        output = model.predict(
            self.test_dataset, batch_size=64, stack_outputs=True)
        np.testing.assert_equal(output[0].shape[0], len(self.test_dataset))

        acc = compute_acc(output[0], self.val_dataset.labels)
        np.testing.assert_allclose(acc, self.acc1)

        sampler = DistributedBatchSampler(
            self.test_dataset, batch_size=64, shuffle=False)

        test_loader = fluid.io.DataLoader(
            self.test_dataset,
            batch_sampler=sampler,
            places=self.device,
            return_list=True)

        model.evaluate(test_loader)

        fluid.disable_dygraph() if dynamic else None

287 288 289 290 291 292 293 294 295 296 297
    def test_predict_without_inputs(self):
        fluid.enable_dygraph(self.device)
        model = Model(LeNet())
        model.prepare()
        model.load(self.weight_path)
        model._inputs = None
        output = model.predict(
            self.test_dataset, batch_size=64, stack_outputs=True)
        np.testing.assert_equal(output[0].shape[0], len(self.test_dataset))
        fluid.disable_dygraph()

298 299 300 301 302 303
    def test_summary_gpu(self):
        paddle.disable_static(self.device)
        rnn = paddle.nn.LSTM(16, 32, 2)
        params_info = paddle.summary(
            rnn, [(-1, 23, 16), ((2, None, 32), (2, -1, 32))])

304

305
class MyModel(paddle.nn.Layer):
L
LielinJiang 已提交
306
    def __init__(self):
307
        super(MyModel, self).__init__()
308
        self._fc = Linear(20, 10)
309 310 311 312 313 314

    def forward(self, x):
        y = self._fc(x)
        return y


315 316 317 318 319 320 321 322 323
class MyDataset(Dataset):
    def __getitem__(self, idx):
        return np.random.random(size=(20,)).astype(np.float32), \
               np.random.randint(0, 10, size=(1,)).astype(np.int64)

    def __len__(self):
        return 40


324 325
class TestModelFunction(unittest.TestCase):
    def set_seed(self, seed=1024):
C
cnn 已提交
326
        paddle.seed(seed)
L
Leo Chen 已提交
327
        paddle.framework.random._manual_program_seed(seed)
328 329 330 331 332 333 334 335 336

    def test_train_batch(self, dynamic=True):
        dim = 20
        data = np.random.random(size=(4, dim)).astype(np.float32)
        label = np.random.randint(0, 10, size=(4, 1)).astype(np.int64)

        def get_expect():
            fluid.enable_dygraph(fluid.CPUPlace())
            self.set_seed()
L
LielinJiang 已提交
337
            m = MyModel()
338 339 340
            optim = fluid.optimizer.SGD(learning_rate=0.001,
                                        parameter_list=m.parameters())
            m.train()
341 342
            output = m(to_tensor(data))
            loss = CrossEntropyLoss(reduction='sum')(output, to_tensor(label))
343 344 345 346 347 348 349 350 351
            avg_loss = fluid.layers.reduce_sum(loss)
            avg_loss.backward()
            optim.minimize(avg_loss)
            m.clear_gradients()
            fluid.disable_dygraph()
            return avg_loss.numpy()

        ref = get_expect()
        for dynamic in [True, False]:
352
            device = paddle.set_device('cpu')
353 354 355
            fluid.enable_dygraph(device) if dynamic else None
            self.set_seed()

L
LielinJiang 已提交
356
            net = MyModel()
357
            optim2 = fluid.optimizer.SGD(learning_rate=0.001,
358
                                         parameter_list=net.parameters())
359

360 361
            inputs = [InputSpec([None, dim], 'float32', 'x')]
            labels = [InputSpec([None, 1], 'int64', 'label')]
362
            model = Model(net, inputs, labels)
363
            model.prepare(optim2, loss=CrossEntropyLoss(reduction="sum"))
364 365 366 367
            loss, = model.train_batch([data], [label])
            np.testing.assert_allclose(loss.flatten(), ref.flatten())
            fluid.disable_dygraph() if dynamic else None

368
    def test_test_batch(self):
369 370 371 372 373 374 375 376
        dim = 20
        data = np.random.random(size=(4, dim)).astype(np.float32)

        def get_expect():
            fluid.enable_dygraph(fluid.CPUPlace())
            self.set_seed()
            m = MyModel()
            m.eval()
377
            output = m(to_tensor(data))
378 379 380 381 382
            fluid.disable_dygraph()
            return output.numpy()

        ref = get_expect()
        for dynamic in [True, False]:
383
            device = paddle.set_device('cpu')
384 385
            fluid.enable_dygraph(device) if dynamic else None
            self.set_seed()
386
            net = MyModel()
387
            inputs = [InputSpec([None, dim], 'float32', 'x')]
388 389
            model = Model(net, inputs)
            model.prepare()
390
            out, = model.predict_batch([data])
391

392
            np.testing.assert_allclose(out, ref, rtol=1e-6)
393 394 395 396 397
            fluid.disable_dygraph() if dynamic else None

    def test_save_load(self):
        path = tempfile.mkdtemp()
        for dynamic in [True, False]:
398
            device = paddle.set_device('cpu')
399
            fluid.enable_dygraph(device) if dynamic else None
L
LielinJiang 已提交
400
            net = MyModel()
401 402
            inputs = [InputSpec([None, 20], 'float32', 'x')]
            labels = [InputSpec([None, 1], 'int64', 'label')]
403
            optim = fluid.optimizer.SGD(learning_rate=0.001,
404 405
                                        parameter_list=net.parameters())
            model = Model(net, inputs, labels)
406
            model.prepare(
407
                optimizer=optim, loss=CrossEntropyLoss(reduction="sum"))
408 409 410 411 412
            model.save(path + '/test')
            model.load(path + '/test')
            shutil.rmtree(path)
            fluid.disable_dygraph() if dynamic else None

413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435
    def test_dynamic_load(self):
        mnist_data = MnistDataset(mode='train')
        for new_optimizer in [True, False]:
            path = tempfile.mkdtemp()
            paddle.disable_static()
            net = LeNet()
            inputs = [InputSpec([None, 1, 28, 28], 'float32', 'x')]
            labels = [InputSpec([None, 1], 'int64', 'label')]
            if new_optimizer:
                optim = paddle.optimizer.Adam(
                    learning_rate=0.001, parameters=net.parameters())
            else:
                optim = fluid.optimizer.Adam(
                    learning_rate=0.001, parameter_list=net.parameters())
            model = Model(net, inputs, labels)
            model.prepare(
                optimizer=optim, loss=CrossEntropyLoss(reduction="sum"))
            model.fit(mnist_data, batch_size=64, verbose=0)
            model.save(path + '/test')
            model.load(path + '/test')
            shutil.rmtree(path)
            paddle.enable_static()

436 437
    def test_dynamic_save_static_load(self):
        path = tempfile.mkdtemp()
438
        # dynamic saving
439
        device = paddle.set_device('cpu')
440
        fluid.enable_dygraph(device)
441
        model = Model(MyModel())
442 443
        optim = fluid.optimizer.SGD(learning_rate=0.001,
                                    parameter_list=model.parameters())
444
        model.prepare(optimizer=optim, loss=CrossEntropyLoss(reduction="sum"))
445 446
        model.save(path + '/test')
        fluid.disable_dygraph()
447

448 449
        inputs = [InputSpec([None, 20], 'float32', 'x')]
        labels = [InputSpec([None, 1], 'int64', 'label')]
L
LielinJiang 已提交
450
        model = Model(MyModel(), inputs, labels)
451 452
        optim = fluid.optimizer.SGD(learning_rate=0.001,
                                    parameter_list=model.parameters())
453
        model.prepare(optimizer=optim, loss=CrossEntropyLoss(reduction="sum"))
454 455 456 457 458 459
        model.load(path + '/test')
        shutil.rmtree(path)

    def test_static_save_dynamic_load(self):
        path = tempfile.mkdtemp()

L
LielinJiang 已提交
460
        net = MyModel()
461 462
        inputs = [InputSpec([None, 20], 'float32', 'x')]
        labels = [InputSpec([None, 1], 'int64', 'label')]
463
        optim = fluid.optimizer.SGD(learning_rate=0.001,
464 465
                                    parameter_list=net.parameters())
        model = Model(net, inputs, labels)
466
        model.prepare(optimizer=optim, loss=CrossEntropyLoss(reduction="sum"))
467 468
        model.save(path + '/test')

469
        device = paddle.set_device('cpu')
470 471
        fluid.enable_dygraph(device)  #if dynamic else None

L
LielinJiang 已提交
472
        net = MyModel()
473 474
        inputs = [InputSpec([None, 20], 'float32', 'x')]
        labels = [InputSpec([None, 1], 'int64', 'label')]
475
        optim = fluid.optimizer.SGD(learning_rate=0.001,
476 477
                                    parameter_list=net.parameters())
        model = Model(net, inputs, labels)
478
        model.prepare(optimizer=optim, loss=CrossEntropyLoss(reduction="sum"))
479 480 481 482 483 484
        model.load(path + '/test')
        shutil.rmtree(path)
        fluid.disable_dygraph()

    def test_parameters(self):
        for dynamic in [True, False]:
485
            device = paddle.set_device('cpu')
486
            fluid.enable_dygraph(device) if dynamic else None
487
            net = MyModel()
488
            inputs = [InputSpec([None, 20], 'float32', 'x')]
489 490
            model = Model(net, inputs)
            model.prepare()
491 492 493 494 495
            params = model.parameters()
            self.assertTrue(params[0].shape[0] == 20)
            self.assertTrue(params[0].shape[1] == 10)
            fluid.disable_dygraph() if dynamic else None

L
LielinJiang 已提交
496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515
    def test_summary(self):
        def _get_param_from_state_dict(state_dict):
            params = 0
            for k, v in state_dict.items():
                params += np.prod(v.numpy().shape)
            return params

        for dynamic in [True, False]:
            device = paddle.set_device('cpu')
            fluid.enable_dygraph(device) if dynamic else None
            net = MyModel()
            inputs = [InputSpec([None, 20], 'float32', 'x')]
            model = Model(net, inputs)
            model.prepare()
            params_info = model.summary()
            gt_params = _get_param_from_state_dict(net.state_dict())

            np.testing.assert_allclose(params_info['total_params'], gt_params)
            print(params_info)

516 517
            model.summary(input_size=(20))
            model.summary(input_size=[(20)])
L
LielinJiang 已提交
518
            model.summary(input_size=(20), dtype='float32')
519

L
LielinJiang 已提交
520
    def test_summary_nlp(self):
521 522 523 524 525 526
        def _get_param_from_state_dict(state_dict):
            params = 0
            for k, v in state_dict.items():
                params += np.prod(v.numpy().shape)
            return params

L
LielinJiang 已提交
527 528 529 530 531
        nlp_net = paddle.nn.GRU(input_size=2,
                                hidden_size=3,
                                num_layers=3,
                                direction="bidirectional")
        paddle.summary(nlp_net, (1, 1, 2))
532

L
LielinJiang 已提交
533
        rnn = paddle.nn.LSTM(16, 32, 2)
534 535 536 537 538 539 540 541 542 543 544 545 546 547
        params_info = paddle.summary(
            rnn, [(-1, 23, 16), ((2, None, 32), (2, -1, 32))])
        gt_params = _get_param_from_state_dict(rnn.state_dict())
        np.testing.assert_allclose(params_info['total_params'], gt_params / 2.0)

        rnn = paddle.nn.GRU(16, 32, 2, direction='bidirectional')
        params_info = paddle.summary(rnn, (4, 23, 16))
        gt_params = _get_param_from_state_dict(rnn.state_dict())
        np.testing.assert_allclose(params_info['total_params'], gt_params / 2.0)

        rnn = paddle.nn.SimpleRNN(16, 32, 2, direction='bidirectional')
        params_info = paddle.summary(rnn, (4, 23, 16))
        gt_params = _get_param_from_state_dict(rnn.state_dict())
        np.testing.assert_allclose(params_info['total_params'], gt_params / 2.0)
L
LielinJiang 已提交
548

L
LielinJiang 已提交
549 550 551 552 553
    def test_summary_dtype(self):
        input_shape = (3, 1)
        net = paddle.nn.Embedding(10, 3, sparse=True)
        paddle.summary(net, input_shape, dtypes='int64')

L
LielinJiang 已提交
554 555 556
    def test_summary_error(self):
        with self.assertRaises(TypeError):
            nlp_net = paddle.nn.GRU(input_size=2, hidden_size=3, num_layers=3)
L
LielinJiang 已提交
557
            paddle.summary(nlp_net, (1, 1, '2'))
L
LielinJiang 已提交
558 559 560 561 562 563 564

        with self.assertRaises(ValueError):
            nlp_net = paddle.nn.GRU(input_size=2, hidden_size=3, num_layers=3)
            paddle.summary(nlp_net, (-1, -1))

        paddle.disable_static()
        nlp_net = paddle.nn.GRU(input_size=2, hidden_size=3, num_layers=3)
L
LielinJiang 已提交
565
        paddle.summary(nlp_net, (1, 1, 2))
L
LielinJiang 已提交
566

567
    def test_export_deploy_model(self):
568
        self.set_seed()
569
        np.random.seed(201)
570
        for dynamic in [True, False]:
571
            paddle.disable_static() if dynamic else None
572 573
            prog_translator = ProgramTranslator()
            prog_translator.enable(False) if not dynamic else None
574
            net = LeNet()
575
            inputs = [InputSpec([None, 1, 28, 28], 'float32', 'x')]
576 577 578 579 580 581 582
            model = Model(net, inputs)
            model.prepare()
            save_dir = tempfile.mkdtemp()
            if not os.path.exists(save_dir):
                os.makedirs(save_dir)
            tensor_img = np.array(
                np.random.random((1, 1, 28, 28)), dtype=np.float32)
583

584
            model.save(save_dir, training=False)
585
            ori_results = model.predict_batch(tensor_img)
586
            fluid.disable_dygraph() if dynamic else None
587

588 589 590 591 592 593 594 595 596 597 598 599 600 601
            place = fluid.CPUPlace() if not fluid.is_compiled_with_cuda(
            ) else fluid.CUDAPlace(0)
            new_scope = fluid.Scope()
            with fluid.scope_guard(new_scope):
                exe = fluid.Executor(place)
                [inference_program, feed_target_names, fetch_targets] = (
                    fluid.io.load_inference_model(
                        dirname=save_dir, executor=exe))
                results = exe.run(inference_program,
                                  feed={feed_target_names[0]: tensor_img},
                                  fetch_list=fetch_targets)
                np.testing.assert_allclose(
                    results, ori_results, rtol=1e-5, atol=1e-7)
                shutil.rmtree(save_dir)
602
            paddle.enable_static()
603

L
LiuChiachi 已提交
604
    def test_dygraph_export_deploy_model_about_inputs(self):
605 606
        mnist_data = MnistDataset(mode='train')
        paddle.disable_static()
L
LiuChiachi 已提交
607
        # without inputs
608
        for initial in ["fit", "train_batch", "eval_batch", "predict_batch"]:
609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628
            save_dir = tempfile.mkdtemp()
            if not os.path.exists(save_dir):
                os.makedirs(save_dir)
            net = LeNet()
            model = Model(net)
            optim = fluid.optimizer.Adam(
                learning_rate=0.001, parameter_list=model.parameters())
            model.prepare(
                optimizer=optim, loss=CrossEntropyLoss(reduction="sum"))
            if initial == "fit":
                model.fit(mnist_data, batch_size=64, verbose=0)
            else:
                img = np.array(
                    np.random.random((1, 1, 28, 28)), dtype=np.float32)
                label = np.array(np.random.rand(1, 1), dtype=np.int64)
                if initial == "train_batch":
                    model.train_batch([img], [label])
                elif initial == "eval_batch":
                    model.eval_batch([img], [label])
                else:
629
                    model.predict_batch([img])
630 631 632

            model.save(save_dir, training=False)
            shutil.rmtree(save_dir)
L
LiuChiachi 已提交
633 634 635 636 637 638 639 640 641 642 643 644
        # with inputs, and the type of inputs is InputSpec
        save_dir = tempfile.mkdtemp()
        if not os.path.exists(save_dir):
            os.makedirs(save_dir)
        net = LeNet()
        inputs = InputSpec([None, 1, 28, 28], 'float32', 'x')
        model = Model(net, inputs)
        optim = fluid.optimizer.Adam(
            learning_rate=0.001, parameter_list=model.parameters())
        model.prepare(optimizer=optim, loss=CrossEntropyLoss(reduction="sum"))
        model.save(save_dir, training=False)
        shutil.rmtree(save_dir)
645

646

647
class TestModelWithLRScheduler(unittest.TestCase):
648 649 650 651
    def test_fit_by_step(self):
        base_lr = 1e-3
        boundaries = [5, 8]

652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670
        def make_optimizer(parameters=None):
            momentum = 0.9
            weight_decay = 5e-4
            values = [base_lr * (0.1**i) for i in range(len(boundaries) + 1)]
            learning_rate = paddle.optimizer.lr.PiecewiseDecay(
                boundaries=boundaries, values=values)
            learning_rate = paddle.optimizer.lr.LinearWarmup(
                learning_rate=learning_rate,
                warmup_steps=4,
                start_lr=base_lr / 5.,
                end_lr=base_lr,
                verbose=True)
            optimizer = paddle.optimizer.Momentum(
                learning_rate=learning_rate,
                weight_decay=weight_decay,
                momentum=momentum,
                parameters=parameters)
            return optimizer

671
        # dynamic test
672 673 674 675 676 677 678 679 680 681 682 683
        device = paddle.set_device('cpu')
        fluid.enable_dygraph(device)
        net = MyModel()
        inputs = [InputSpec([None, 20], 'float32', 'x')]
        labels = [InputSpec([None, 1], 'int64', 'label')]
        optim = make_optimizer(net.parameters())
        model = Model(net, inputs, labels)
        model.prepare(optimizer=optim, loss=CrossEntropyLoss(reduction="sum"))

        dataset = MyDataset()
        model.fit(dataset, dataset, batch_size=4, epochs=10, num_workers=0)

684 685
        np.testing.assert_allclose(model._optimizer._learning_rate.last_lr,
                                   base_lr * (0.1**len(boundaries)))
686
        # static test
687 688
        paddle.enable_static()

689 690 691 692 693 694 695 696 697 698
        net = MyModel()
        inputs = [InputSpec([None, 20], 'float32', 'x')]
        labels = [InputSpec([None, 1], 'int64', 'label')]
        optim = make_optimizer(net.parameters())
        model = Model(net, inputs, labels)
        model.prepare(optimizer=optim, loss=CrossEntropyLoss(reduction="sum"))

        dataset = MyDataset()
        model.fit(dataset, dataset, batch_size=4, epochs=10, num_workers=0)

699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785
        np.testing.assert_allclose(model._optimizer._learning_rate.last_lr,
                                   base_lr * (0.1**len(boundaries)))

    def test_fit_by_epoch(self):
        base_lr = 1e-3
        boundaries = [5, 8]
        epochs = 10
        wamup_epochs = 4

        def make_optimizer(parameters=None):
            momentum = 0.9
            weight_decay = 5e-4
            values = [base_lr * (0.1**i) for i in range(len(boundaries) + 1)]
            learning_rate = paddle.optimizer.lr.PiecewiseDecay(
                boundaries=boundaries, values=values)
            learning_rate = paddle.optimizer.lr.LinearWarmup(
                learning_rate=learning_rate,
                warmup_steps=wamup_epochs,
                start_lr=base_lr / 5.,
                end_lr=base_lr,
                verbose=True)
            optimizer = paddle.optimizer.Momentum(
                learning_rate=learning_rate,
                weight_decay=weight_decay,
                momentum=momentum,
                parameters=parameters)
            return optimizer

        # dynamic test
        device = paddle.set_device('cpu')
        fluid.enable_dygraph(device)
        net = MyModel()
        inputs = [InputSpec([None, 20], 'float32', 'x')]
        labels = [InputSpec([None, 1], 'int64', 'label')]
        optim = make_optimizer(net.parameters())
        model = Model(net, inputs, labels)
        model.prepare(optimizer=optim, loss=CrossEntropyLoss(reduction="sum"))

        dataset = MyDataset()

        lr_scheduler_callback = paddle.callbacks.LRScheduler(
            by_step=False, by_epoch=True)

        model.fit(dataset,
                  dataset,
                  batch_size=4,
                  epochs=epochs,
                  num_workers=0,
                  callbacks=lr_scheduler_callback)

        cnt = 0
        for b in boundaries:
            if b + wamup_epochs <= epochs:
                cnt += 1

        np.testing.assert_allclose(model._optimizer._learning_rate.last_lr,
                                   base_lr * (0.1**cnt))
        # static test
        paddle.enable_static()

        net = MyModel()
        inputs = [InputSpec([None, 20], 'float32', 'x')]
        labels = [InputSpec([None, 1], 'int64', 'label')]
        optim = make_optimizer(net.parameters())
        model = Model(net, inputs, labels)
        model.prepare(optimizer=optim, loss=CrossEntropyLoss(reduction="sum"))

        dataset = MyDataset()

        lr_scheduler_callback = paddle.callbacks.LRScheduler(
            by_step=False, by_epoch=True)

        model.fit(dataset,
                  dataset,
                  batch_size=4,
                  epochs=epochs,
                  num_workers=0,
                  callbacks=lr_scheduler_callback)

        cnt = 0
        for b in boundaries:
            if b + wamup_epochs <= epochs:
                cnt += 1

        np.testing.assert_allclose(model._optimizer._learning_rate.last_lr,
                                   base_lr * (0.1**cnt))

786

787 788
class TestRaiseError(unittest.TestCase):
    def test_input_without_name(self):
L
LielinJiang 已提交
789
        net = MyModel()
790

791 792
        inputs = [InputSpec([None, 10], 'float32')]
        labels = [InputSpec([None, 1], 'int64', 'label')]
793 794 795
        with self.assertRaises(ValueError):
            model = Model(net, inputs, labels)

796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811
    def test_static_without_inputs(self):
        paddle.enable_static()
        net = MyModel()
        with self.assertRaises(TypeError):
            model = Model(net)

    def test_save_infer_model_without_inputs_and_run_in_dygraph(self):
        paddle.disable_static()
        net = MyModel()
        save_dir = tempfile.mkdtemp()
        if not os.path.exists(save_dir):
            os.makedirs(save_dir)
        with self.assertRaises(RuntimeError):
            model = Model(net)
            model.save(save_dir, training=False)
        paddle.enable_static()
812

813

814 815
if __name__ == '__main__':
    unittest.main()