Skip to content

Releases: ArtificialIntelligenceToolkit/aitk.networks

v0.3.8

29 Apr 00:53
Compare
Choose a tag to compare

Major changes:

  • Allow SimpleNetwork to take Layers
  • initialize() checks inputs for color; make_image() respects it
  • Raise an error for malformed SimpleNetwork layer representation
  • Use a input_dataset in predict_to
  • fix display_colormaps()
  • Added network.propagate(show=False)
  • Initialize limits with first generation of image
  • Syntax of setup requires
  • Use "Flatten" rather than 0
  • allow tolerance as kwarg to fit()

0.3.2

02 Apr 00:40
Compare
Choose a tag to compare

Major changes:

  • Network.propagate() returns values
  • Network.fit() updates watched items with random dataset pattern
  • Fixed a bug in Network.fit(); finish update - > propagate changes

0.3.1

01 Apr 10:35
Compare
Choose a tag to compare

Major changes:

  • Changed network.update() to network.propagate(an_input)
  • Changed network.update(dataset=true) to network.propagate_each(inputs)

0.3.0

31 Mar 14:27
Compare
Choose a tag to compare

Major changes:

  • Added network.watch(), network.watch_layer(), and network.watch_weights()
  • Added network.update(pattern) to update the watchers
  • network.fit() calls network.update() on the first pattern every report_rate epochs
  • Added network.enumerate_dataset(inputs)
  • Draw blue border around dynamic/watched views
  • Added Network(layers=[]) and network.connect(name, name)
  • Added get_image(), display(), and get_widget() to Network

0.2.5

06 Mar 20:56
Compare
Choose a tag to compare

Major changes:

  • requires aitk.utils
  • Added predict_histogram()

0.2.4

05 Mar 15:02
Compare
Choose a tag to compare

Major changes:

  • Added Network.predict_histogram()
  • Updated titles in network display

0.2.3

05 Mar 00:41
Compare
Choose a tag to compare

Major changes:

  • Network.predict_from() working on all tested layer types
  • Renamed take_picture to display
  • Removed set_weights(flat) flag
  • PCA now in title of network when Network.predict_pca()

0.2.2

03 Mar 10:06
Compare
Choose a tag to compare

Major changes:

  • work on representations for single and multi-input bank networks

0.2.1

28 Feb 05:18
Compare
Choose a tag to compare

Major changes:

  • keep aspect ratio of PCA charts
  • added sizes for PCA scatter points
  • allow PCA on 1-unit layers
  • PlotCallback is now UpdateCallback

0.2.0

27 Feb 18:57
Compare
Choose a tag to compare

Major changes:

  • Added Network.predict_pca(inputs, colors=colors, set_pca_spaces=True)
  • Added Network.fit(save=N) - saves weights every N epochs
  • Network.take_picture() now displays as side-effect rather than return value
  • Removed Network.display_picture()
  • All display methods have a clear=True argument