I have three issues about A3C-lstm

[vincenschan]

1. In the first layer of ACLSTM-Net, 'W_conv1' shape is [8,8,4,16] and what is the mean of the two 8-dim?

2. Input data format and what is the mean of the every dimension?
   I want to realize a price-predict with A3C, but the input-data is different with this project.

3. In the training process, how is running in multi-threading and the order of counting gradients may be affect the updating of w&b ?

Thanks!