search

e2crawfo / auto_yolo

Original implementation of Spatially Invariant Attend, Infer, Repeat (SPAIR) in TensorFlow.
MIT License
51 stars 17 forks source link

Add tensorflow_probability as requirement #3

Closed alcinos closed 5 years ago

tijsmaas commented 5 years ago

Hi I am trying to reproduce the results and i'm running into tensorflow_probability as well. The authors claim to have used tensorflow 1.8.0 and in order to run the experiments tensorflow_probability is required, thanks for the update @alcinos! All (recent) tensorflow_probability versions require tensorflow >=1.9.0. Installing tensorflow_probability=0.1.0 proves that the old version is insufficient: AttributeError: module 'tensorflow_probability' has no attribute 'distributions'. What combination of tensorflow_probability and tensorflow did you use? Do you remember at what version/commit of tensorflow_probability you used in the original setup @e2crawfo?

e2crawfo commented 5 years ago

@tijsmaas For the original setup (i.e. branch aaai_2019) I used tf 1.8 without tfp. For recent development I've been using tf 1.13.1 and tfp 0.6.0. See new README.md.

@alcinos Thanks for the PR. After some thought I decided to add instructions to the README.md to install tfp instead of putting it in requirements.txt since the tfp version is so dependent on tf version.