CUBLAS error about unsupervised-videos HOT 3 CLOSED

It looks like the GPU is out of memory and thats why it can't create cmat matrix. Trying reducing the number of hidden units in LSTM.

Which GPU are you using ?

from unsupervised-videos.

Thanks for reply.
Yes, you were right, I solved the memory problem and now after running the following script:

python lstm_combo.py models/lstm_combo_1layer_mnist.pbtxt datasets/bouncing_mnist.pbtxt datasets/bouncing_mnist_valid.pbtxt 0

Using gpu: 0
Using board 0
2048
2048
2048
Model saved at /home/n9033297/unsupervised-videos/models/lstm_combo_20160602160022.pbtxt
Step 1Traceback (most recent call last):
File "/home/n9033297/unsupervised-videos/lstm_combo.py", line 407, in
main()
File "/home/n9033297/unsupervised-videos/lstm_combo.py", line 397, in main
lstm_autoencoder.Train(train_data, valid_data)
File "/home/n9033297/unsupervised-videos/lstm_combo.py", line 349, in Train
self.Fprop(train=True)
File "/home/n9033297/unsupervised-videos/lstm_combo.py", line 60, in Fprop
self.lstm_stack_enc_.Fprop(input_frame=self.v_.col_slice(t * self.num_dims_, (t+1) * self.num_dims_))
File "/home/n9033297/unsupervised-videos/lstm.py", line 327, in Fprop
train=train, copy_init_state=copy_init_state)
File "/home/n9033297/unsupervised-videos/lstm.py", line 120, in Fprop
use_relu=self.use_relu_, init=init)
File "/home/n9033297/unsupervised-videos/cudamat/cudamat.py", line 2226, in lstm_fprop
err_code = _cudamat.lstm_fprop(s_in.p_mat, s_out.p_mat, w_dense.p_mat, w_diag.p_mat, b.p_mat, ct.c_bool(init), ct.c_bool(use_relu))
AttributeError: 'module' object has no attribute 'c_bool'

Is that normal to get this error and how can I fix it?
Thanks.

from unsupervised-videos.

It is something to do with ctypes library build in into python.

We were using python 2.7 (i believe) for experiments so, newer versions of python might have made some changes to ctypes library that breaks the code.

from unsupervised-videos.