Conx

Released Fri May 4, 2018

* propagate_to_features() scales layer[0]
* added cx.array
* fixed (somewhat) array_to_image(colormap=)
* added VGG16 and ImageNet notebook
* New Network.info()
* Updated Network.propagate_to_features(), util.image()
* Network.info() describes predefined networks
* new utility image(filename)
* rewrote Network.propagate_to_features() to be faster
* added VGG preprocessing and postprocessing
* Picture autoscales inputs by default
* Add net.picture(minmax=)
* Rebuild models on import_keras
* Added VGG19
* Added vgg16 and idea of applications as Network.networks
* Bug in building intermediate hidden -> output models

Released Tue April 17, 2018

* Fixed bug in building hidden -> output intermediate models

Released Fri April 13, 2018

* Added cx.view_image_list(pivot) - rotates list and layout
* Added colors dataset
* Added Dataset.delete_bank(), Dataset.append_bank()
* Added Dataset.ITEM[V] = value

Released Fri April 6, 2018

* Removed examples; use notebooks or help instead
* cx.view_image_list() can have layout=None, (int, None), or (None, int)
* Added cx.scale(vector, range, dtype, truncate)
* Added cx.scatter_images(images, xys) - creates scatter plot of images
* Fixed pca.translate(scale=SCALE) bug
* downgrade tensorflow on readthedocs because memory hog kills build

Released Thur April 5, 2018

* changed "not allowed" warning on multi-dim outputs to
"are you sure?"
* fix colormap on array_to_image; added tests
* fix cx.view(array)
* Allow dataset to load generators, zips, etc.

Released Tue April 3, 2018

* Two fixes for array_to_image: div by float; move cmap conversion to end
* Protection for list/array for range and shape
* from kmader: Adding jyro to binder requirements