test_quant_post_hpo.py

# Copyright (c) 2021  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.
import sys
import os
sys.path.append(".")
sys.path[0] = os.path.join(os.path.dirname("__file__"), os.path.pardir)

import paddle
import unittest
from paddleslim.quant import quant_post_hpo
from static_case import StaticCase
sys.path.append("../demo")
from models import MobileNet
from layers import conv_bn_layer
import numpy as np


class TestQuantPostHpoCase1(StaticCase):
    def test_accuracy(self):
        image = paddle.static.data(
            name='image', shape=[None, 1, 28, 28], dtype='float32')
        label = paddle.static.data(name='label', shape=[None, 1], dtype='int64')
        model = MobileNet()
        out = model.net(input=image, class_dim=10)
        cost = paddle.nn.functional.loss.cross_entropy(input=out, label=label)
        avg_cost = paddle.mean(x=cost)
        acc_top1 = paddle.metric.accuracy(input=out, label=label, k=1)
        acc_top5 = paddle.metric.accuracy(input=out, label=label, k=5)
        optimizer = paddle.optimizer.Momentum(
            momentum=0.9,
            learning_rate=0.01,
            weight_decay=paddle.regularizer.L2Decay(4e-5))
        optimizer.minimize(avg_cost)
        main_prog = paddle.static.default_main_program()
        val_prog = main_prog.clone(for_test=True)

        place = paddle.CUDAPlace(
            0) if paddle.is_compiled_with_cuda() else paddle.CPUPlace()
        exe = paddle.static.Executor(place)
        exe.run(paddle.static.default_startup_program())

        def transform(x):
            return np.reshape(x, [1, 28, 28])

        train_dataset = paddle.vision.datasets.MNIST(
            mode='train', backend='cv2', transform=transform)
        test_dataset = paddle.vision.datasets.MNIST(
            mode='test', backend='cv2', transform=transform)

        train_loader = paddle.io.DataLoader(
            train_dataset,
            places=place,
            feed_list=[image, label],
            drop_last=True,
            batch_size=64,
            return_list=False)

        valid_loader = paddle.io.DataLoader(
            test_dataset,
            places=place,
            feed_list=[image, label],
            batch_size=64,
            return_list=False)

        def sample_generator_creator():
            def __reader__():
                for data in test_dataset:
                    image, label = data
                    image = np.expand_dims(image, axis=0)
                    label = np.expand_dims(label, axis=0)
                    yield image, label

            return __reader__

        def train(program):
            iter = 0
            for data in train_loader():
                cost, top1, top5 = exe.run(
                    program,
                    feed=data,
                    fetch_list=[avg_cost, acc_top1, acc_top5])
                iter += 1
                if iter % 100 == 0:
                    print(
                        'train iter={}, avg loss {}, acc_top1 {}, acc_top5 {}'.
                        format(iter, cost, top1, top5))

        def test(program, outputs=[avg_cost, acc_top1, acc_top5]):
            iter = 0
            result = [[], [], []]
            for data in valid_loader():
                cost, top1, top5 = exe.run(
                    program, feed=data, fetch_list=outputs)
                iter += 1
                if iter % 100 == 0:
                    print('eval iter={}, avg loss {}, acc_top1 {}, acc_top5 {}'.
                          format(iter, cost, top1, top5))
                result[0].append(cost)
                result[1].append(top1)
                result[2].append(top5)
            print(' avg loss {}, acc_top1 {}, acc_top5 {}'.format(
                np.mean(result[0]), np.mean(result[1]), np.mean(result[2])))
            return np.mean(result[1]), np.mean(result[2])

        train(main_prog)
        top1_1, top5_1 = test(val_prog)
        paddle.static.save_inference_model(
            path_prefix='./test_quant_post_hpo/model',
            feed_vars=[image, label],
            fetch_vars=[avg_cost, acc_top1, acc_top5],
            executor=exe,
            program=val_prog)

        quant_post_hpo(
            exe,
            place,
            "./test_quant_post_hpo",
            "./test_quant_post_hpo_inference",
            train_sample_generator=sample_generator_creator(),
            eval_sample_generator=sample_generator_creator(),
            train_dataloader=valid_loader,
            eval_dataloader=valid_loader,
            model_filename="model.pdmodel",
            params_filename="model.pdiparams",
            save_model_filename='model.pdmodel',
            save_params_filename='model.pdiparams',
            runcount_limit=2)
        quant_post_prog, feed_target_names, fetch_targets = paddle.static.load_inference_model(
            path_prefix='./test_quant_post_hpo_inference/model', executor=exe)
        top1_2, top5_2 = test(quant_post_prog, fetch_targets)
        print("before quantization: top1: {}, top5: {}".format(top1_1, top5_1))
        print("after quantization: top1: {}, top5: {}".format(top1_2, top5_2))


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