Enchanter v0.9.0

Release 0.9.0

Updates

• #170 Support Python 3.9
• Change comet.ml to an option.
• Change netron to an option.

If you want to use enchanter with Comet.ml, you should type

$ pip install enchanter[comet]

If you want to use enchanter with netron,

$ pip install enchanter[netron]

Enchanter v0.9.0b0

v0.9.0 beta 0

• #167 Change netron & comet to an option.
• #170 Support Python 3.9
• close #186

Enchanter v0.9.0a0

Enchanter 0.9.0 alpha 0

• #170 Support Python 3.9
• #167 Change netron & comet to an option.

Enchanter v0.8.1

Release 0.8.1

Bug Fix

• #157 Fixed an issue which torch.jit.script could not be done with CausalConv1d and TemporalConvBlock.
• #160 Change to the method using Union instead of Protocol
• #165 GitHub Actions add-path

Documentation

• #159 fix missing enchanter.utils.datasets in documentation.

Enchanter v0.8.0

Release 0.8.0

New Features

• #117 Unsupervised Representation Learning for Time Series.(Only Fixed Length)
• #108 Temporal Convolution Block
• #139 Callback Manager
• #150 TypeHint Module

Enhancement

• update Early Stopping
• #154 New Logo

Breaking Changes

• Remove wrappers package

Enchanter v0.8.0 alpha

v0.8.0a

Enchanter v0.7.1

Release 0.7.1

Improve

• #136 Improve TensorBoardLogger
• #134 Improve RunnerIO.save()

Bug fix

• fix #132
• fix #129

Documentation update

• #113 JA -> EN relate(#141 )
• #144 Typo fix

Enchanter v0.7.0

Release 0.7.0

New Features

• #103 Support AMP

from torch.cuda.amp import GradScaler
from enchanter.tasks import ClassificationRunner


runner = ClassificationRunner(...)
runner.scaler = GradScaler()    # overwrite
...
runner.run()

• #104 Migrating from setup.py to poetry
• #110 Add enchanter.addons.FReLU1d, enchanter.addons.FReLU2d.
• #112 [Experimental] Add ResidualSequential.
• #119 [Experimental] Support TensorFlow Dataset
• #121 Context API

with enchanter.tasks.ClassificationRunner( ... ) as runner:
  for epoch in range(10):
    with runner.experiment.train():
      for train_batch in train_loader:
        runner.optimizer.zero_grad()
        train_out = runner.train_step(batch)
        runner.backward(train_out["loss"])
        runner.update_optimizer()
		
        with runner.experiment.validate(), torch.no_grad():
           for val_batch in val_loader:
             val_out = runner.val_step(batch)
			
    with runner.experiment.test(), torch.no_grad():
      for test_batch in test_loader:
        test_out = runner.test_step(test_batch)

• #126 Graph Visualizer

import torch
from enchanter.utils import visualize

model = ...
visualize.with_netron(model)
# and open web browser 

Bug Fix

• #106 Fix SELayer1d, SELayer2d

API Changes

• #107 change enchanter.wrappers -> enchanter.tasks.
  wrappers package has been renamed to tasks since v0.7.0 and will be remove in v0.8.0.

Documentation

• #105 closing GitHub Pages
• #116 Fix typo

Enchanter v0.7.0 rc0

v0.7.0 RC0 Release

Enchanter v0.7.0beta1

v0.7.0 beta release 1