按照https://keras.io/api/layers/convolution_layers/convolution1d/、https://keras.io/api/layers/pooling_layers/max_pooling1d/和https://keras.io/api/layers/recurrent_layers/lstm/ ，原代码中`MaxPooling1D`设置的`data_format='channels_first'`，输出维度为`(batch_size, features, downsampled_steps)`。LSTM layer的输入为`(batch, timesteps, feature)`。原代码没有经过维度转化，LSTM的通道和时间步反了
按照https://keras.io/api/layers/convolution_layers/convolution1d/、https://keras.io/api/layers/pooling_layers/max_pooling1d/和https://keras.io/api/layers/recurrent_layers/lstm/ ，原代码中
MaxPooling1D设置的data_format='channels_first'，输出维度为(batch_size, features, downsampled_steps)。LSTM layer的输入为(batch, timesteps, feature)。原代码没有经过维度转化，LSTM的通道和时间步反了