Can't test it

[Shanicky-RenzhiWang]

I setup the environment with anaconda2.7 under ubuntu16.04, packages keras1.2.2, z3-solver and theano0.9.0 have been installed. When I run Dlv.py, there's an error like "Negative dimension size caused by subtracting 2 from 1...", no matter what dataset I choose, it occurs. I wonder that what the problem.
PS. if I replace 1,img_rows, img_cols with img_cols, img_rows,1 in mnist_network.py, the mnist would run its method of training

[xiaoweih]

Please change the backend of Keras from the default tensorflow into theano.

If it still does not work, please do the following
(1) uninstall the keras
(2) remove the ~/.keras/keras.json file
(3) install keras 1.2.2
(4) change the backend from tensorflow into theano

[xiaoweih]

The changing of backend can be done by editing the ~/.keras/keras.json file from

"backend": "tensorflow",
"image_dim_ordering": "tf",

into

"backend": "theano",
"image_dim_ordering": "th",

[Shanicky-RenzhiWang]

It works! Thanks!
I'm reading your paper, what a good idea

[xiaoweih]

Thanks!