|
# Reproducing SRNN's sequence subsequence selection as done in |
|
# https://github.com/asheshjain399/RNNexp/blob/master/structural_rnn/CRFProblems/H3.6m/processdata.py#L343 |
|
for i in xrange( batch_size ): |
|
|
|
_, subsequence, idx = seeds[i] |
|
idx = idx + 50 |
|
|
|
data_sel = data[ (subject, action, subsequence, 'even') ] |
|
|
|
data_sel = data_sel[(idx-source_seq_len):(idx+target_seq_len) ,:] |
|
|
|
encoder_inputs[i, :, :] = data_sel[0:source_seq_len-1, :] |
|
decoder_inputs[i, :, :] = data_sel[source_seq_len-1:(source_seq_len+target_seq_len-1), :] |
|
decoder_outputs[i, :, :] = data_sel[source_seq_len:, :] |
In the above, at line 550, I believe the 50 is actually a source_seq_len?