Minor bug in weight sparsification learner

Tencent / PocketFlow

An Automatic Model Compression (AutoMC) framework for developing smaller and faster AI applications.

https://pocketflow.github.io

Other

2.78k stars 490 forks source link

Minor bug in weight sparsification learner #130

Closed haolibai closed 5 years ago

haolibai commented 5 years ago

Describe the bug According to To prune, or not to prune: exploring the efficacy of pruning for model compression, the mask should be updated based on the un-pruned weights. However in the code, the mask is updated based on the pruned weights.

Screenshots

jiaxiang-wu commented 5 years ago

Thanks for pointing it out. Both mask_thres and mask should be updated based on var_bkup, rather than var in the current implementation. We will fix it ASAP.

jiaxiang-wu commented 5 years ago

Resolved in PR #134.