tensorflow / probability

Probabilistic reasoning and statistical analysis in TensorFlow
https://www.tensorflow.org/probability/
Apache License 2.0
4.23k stars 1.09k forks source link

some issues abort example latent_dirichlet_allocation.py #110

Closed haowuliao closed 6 years ago

haowuliao commented 6 years ago

when I use win10 ,python 3.6, tensorflow 1.9, run example latent_dirichlet_allocation.py without any change about default params, I found that I could not get the result as this example said, the alpha in print allways were 0.7 until end ,and final perplexity was more than 1000. when I let learning_rate 0.1, I saw the alpha in print changed. so , I cannot think out why happened.

mfigurnov commented 6 years ago

Could you try running the example with tf-nightly or the release candidate of 1.10? TensorFlow 1.9 does not have gradients for tf.random_gamma (see https://github.com/tensorflow/probability/issues/51), which are essential for this example to work.

haowuliao commented 6 years ago

Thanks, this problem was dealed. What you said was reason.