emtiyaz / vmp-for-svae Goto Github PK

could this approach be applied to cases when we have non-parametric PGMs such as the DP Mixture Model?

Please can you also add a sample command to run wheel dataset ?

2019-06-12 08:37:57.382950: W tensorflow/core/kernels/queue_base.cc:277] _0_data_prep/shuffle_batch/random_shuffle_queue: Skipping cancelled enqueue attempt with queue not closed 2019-06-12 08:37:57.383017: W tensorflow/core/kernels/queue_base.cc:277] _0_data_prep/shuffle_batch/random_shuffle_queue: Skipping cancelled enqueue attempt with queue not closed Crashed for config: {'dataset': 'auto', 'method': 'svae-cvi', 'lr': 0.0003, 'lrcvi': 0.2, 'decay_rate': 0.95, 'K': 10, 'L': 6, 'U': 50, 'seed': 0} 'c' argument has 20 elements, which is not acceptable for use with 'x' with size 10, 'y' with size 10.

Nothing change to codes. env: tf 1.11.0 GPU

hi,

Is the code for LDS experiment mentioned in Appendix C of the paper missing in the code repo?

thanks

I use the same setting as auto dataset and displayed problem:

Crashed for config:
 {'dataset': 'mnist', 'method': 'svae-cvi', 'lr': 0.0003, 'lrcvi': 0.2, 'decay_rate': 0.
95, 'K': 20, 'L': 6, 'U': 50, 'seed': 0}
list index out of range

if there is a example of how to config MNIST dataset would be nice !

Hi,

you mention that your model is limited to small scale data. Didn't Johnson et. al apply their SLDS on mouse videos that contained millions of frames of video data?

Does there model suffer from the same limitation as you mentioned in your paper? If there model is scalable, could you comment on why it is so?

Thanks

I am running the file 'experiments_multi_gpu.py' and am getting the following error:

module 'tensorflow.contrib.linalg' has no attribute 'LinearOperatorTriL'

Hello, I am trying to run the experiments.py for the first time. and got the following error:

module 'tensorflow.contrib' has no attribute 'linalg'

which prompted me to think which version of tensorflow does the code use? I have tf.version == '1.12.0'

Thanks,
Machuck