Created by: sguuaa

写了一个函数，想测试一下这个函数，程序如下。

import numpy as np
import paddle
import paddle.fluid as fluid
def fun(prediction, gt):
    valid_mask = (gt >=0).astype('float32')
    gt_mask = []
    for i in range(0, 3):
        temp1 = np.array((gt == i).astype('float32'))
        # print(valid_mask.shape)
        temp2 = np.array(valid_mask)
        # print(temp2.shape)
        temp2 = temp2[:, i, :, :]
        temp = fluid.layers.elementwise_mul(temp1, temp2)
        gt_mask.append(temp)

    prediction = fluid.layers.elementwise_mul(prediction, valid_mask)
    loss = fluid.layers.elementwise_mul(prediction.clamp(min=1e-7).log(), gt_mask).sum()
    return loss


if __name__ == '__main__':
    prediction = fluid.layers.data(name='prediction', shape=[3, 10, 10], dtype='float32')
    gt = fluid.layers.data(name = 'gt', shape = [1, 10, 10], dtype='float32')

    cost = fun(prediction, gt)

    def create_reader():
       def reader():
           prediction = np.random.rand(3, 10, 10)
           gt = np.random.rand(1, 10, 10) * 3
           yield prediction, gt
       return reader

    db = paddle.batch(create_reader(), batch_size = 2)
    place = fluid.CPUPlace()
    feeder = fluid.DataFeeder(place = place, feed_list = [prediction, gt])
    fetch_list = [cost]
    exe = fluid.Executor(place)

    for idx, data in enumerate(db()):
       print('data:',data)
       loss = exe.run(feed = feeder.feed(data), fetch_list = fetch_list)
       print(loss)
       break