Update to latest classifier inference

Remove unneeded scripts. Change eval to use table 4 datasets. Clean

Update CSV logger

More cleanup, edit configs

Add new scripts/configs for eval and easy usage. Rename configs.

Merge branch 'main' into classifier-inference

Re-sample effects if STFT too low

Update datagen silence threshold to 1e-4

adding a dynamic dataset for on the fly generation of training data

adding support for parallel processing in dataset generation

Add 5-1 experiment. Update test batch_size to 2

Fix custom inferencing issues

Add custom-inference dataset

Enhance dataset unique string2

Enhance dataset unique string

Remove debug statement from dataset

Prepare apply <=n remove n cfgs

Update dataset to use range for num effects

Remove previous DPTNet/DCUNet implementations

Remove all effects being printed

Add normalization between effects

Remove IDMT-SMT-GUITAR for now

Improve speed of selecting random chunks2

Improve speed of selecting random chunks

Pick new chunk if too short

Fix issue where chunk length is 0

Add random chunk choice

Add random sampling of datasets to prevent class imbalance

Add dsd100 dataset

Merge branch 'main' into cjs--classifier-v2

support for 5 datasets

Fix bug in removing single effects

Update callbacks, debug new models

Init dcunet and dptnet

Fix new dataset to work for remfx training

rework dataset generation for better logging of effects added

Add orderedset

Allow for specifying max_effects as -1

Add max_removed_effects

Refactor commit, see README for details

Add ability to remove all effects

Refactor dataset to apply multiple effects at a time

Fix effect_type dirname

Fix render_path location so it doesn't overlap on same deivce

Add gradient clipping and lr scheduler

Randomize train and val effects the same, drop audio that is too short

Fix render_root naming2

Re-add change to config output processed files in separate dir

Clean project. Add 'all effects' to experiments

Fix output_root -> self.output_root3