Each example has 3 actions:
-
train
trains a new example
-
sample
samples from a trained example
-
search
randomly searches for a configuration and outputs JSON / metrics
Trains a generator to output 2d points (pixels) matching a known distribution.
Search: 2d-distance measure from generator batch to known distribution
Colorizer feeds a black and white version of the input into the generator.
State: working
Search: Distance from black and white image to black and white version of generated output
Align images and black and white versions of those images.
State: working
Search: Distance from Gab(Xa),black_and_white(Xa)
Reconstruct input images using AutoencoderGAN
State: working
Search: Reconstruction cost
Memorize X and Z values then test against them.
State: working
Search: Reconstruction