Add RepeatVector layer #123
Comments
|
I could do this one :) |
|
Hey @zaleslaw I am a bit lost on how best to write tests for RepeatVector class. Should I dig into the unit tests of Keras and convert them to Kotlin, or come up with tests myself? |
|
If the logic is simple, like in max pooling 2d layer, you could make it as you wish, if it includes complex calculations you could use some numerical data from keras, but you don't need make full copy of its tests, 1 or 2 for forward and main exceptions or cornerstones. Or you could use keras to pass input and get output of the tested layer. |
|
@zaleslaw alright. I added a simple test in dosier@bf61c5d. This is probably not enough though. What do u mean by cornerstones? |
|
hmm, maybe test for the negative number n=-10, for example, to get an IllegalArgumentException (you could use |
* Added missing saving functions for ReLU and ELU activation layers (JetBrains#78) * Reverted changes to the imports * Added RepeatVector layer #123 * Added serialisation support for RepeatVector layer #123 * Wrote test for RepeatVector #123 * Made changed requested by avan (see desc.) - added missing require check in init block of RepeatVector - updated docs - reformatted code - housekeeping * Removed redundant Obs.repeat ext fun * Made changed requested by avan (see desc.) - change require message in computeOutputShape - used inputShape.size(...) for creating shape - removed author tag * Used `=` instead of `return` block, added TODO * Implemented changes requested by zaleslaw - save trainability status - renamed tests * Added test for negative `n` #123 * Added missing newline
Currently, the support for
RepeatVectoris missed, and it would be great to add support for these layers. The desired PR addressing this issue should include:RepeatVector(you can take inspiration from the implementation ofReshapeas reference)apimoduleAlso, if needed, you can take a look at Keras documentation for
RepeatVectorNOTE: for the moment, there is no need to add support for "data format" (i.e., channels last vs. channels first) in your PR; you can assume the channels are always in the last dimension.
P.S. If you want to take this ticket, please leave the comment below
P.P.S Read the Contributing Guidelines.
The text was updated successfully, but these errors were encountered: