Skip to content

Catalyst 19.09
Compare
Choose a tag to compare
@Scitator Scitator released this 02 Sep 18:47
· 819 commits to master since this release
v19.09
b37b527

Catalyst 19.08 -> 19.09

Ecosystem

We are happy to announce MLComp release – a distributed DAG (Directed acyclic graph) framework for machine learning with UI. Powered by Catalyst.Team.

We also release a detailed classification tutorial Open In Colab and comprehensive classification pipeline.

Slowly-slowly, more and more challenges are powered by catalyst #302.

We also update the licence to Apache 2.0, start the Patreon and even run catalyst-info repo!

And finally, we have integrate wandb to the catalyst, both DL & RL!

Overall

  • config dump to tensorboard #284 – even more reproducibility for both DL and RL
  • bash scripts for parallel-gpu-run and catalyst-rl-run #288
  • improver CUDNN deterministic and CUDNN benchmark support #299 #309
  • available gpus check feature #300
  • SequentialNet update with "soft" residual mode #301
  • docs update #303 #336
  • layer-wise learning rate support #283
  • torchnet dependency removal #304
  • environment variables and packages versions dump to tensorboard #313 #317
  • tracing script update #319
  • additional general case CV scripts #321
  • RAdam, LookAhead, Ranger optimizers support #332
  • wandb integration #337 #339 #341

DL

  • Exception handling support for callbacks #281 #307
  • tests update #286
  • traced models support for dataset prediction #287
  • tensorboard logger fix #294
  • state_dict param support for all contrib encoders #292
  • bias weight decay autoremove #293
  • checkpoint save on exception #295
  • classification tutorial #296 #297 #326
  • better console logging #298
  • segmentation models support for dl registry #329
  • one-hot feature #331
  • accuracy metric fix #340
  • path dataset support #335
  • callback ordering support #343
  • encoders requires_grad logic update #346

RL

  • DB dump/load refactoring #318
  • mongo fix #345
Assets 2
Loading