Karlson Pfannschmidt

This is a follow-up to #44.

Support for variably sized inputs

3

cs-ranking is currently requiring the user to provide a fixed input size. There exist different tricks to be able to handle variably sized inputs (e.g. padding to maximum length), but...

Device placement is logged by default

4

We have a utility function `configure_numpy_keras` which is used in some of the experiment scripts: https://github.com/kiudee/cs-ranking/blob/a635d59a254e7d4cad06c3e04f7593392e0b9cec/csrank/tensorflow_util.py#L40-L58 It does the following: - Set random seeds - Sets the `KERAS_BACKEND` to Tensorflow...

Dataset generators like https://github.com/kiudee/cs-ranking/blob/229d5dd02a10d1efa8f3d44538b83e84f2b3d927/csrank/dataset_reader/choicefunctions/choice_data_generator.py#L8-L17 inherit the docstring of the parent class, which is not very informative.

Rationale --------- When training the models on different datasets, it would be advantageous to be able to save the model as is to a file. That way it is easy...

Add 1/2 pseudocounts to WDL in SPRT.

14

As mentioned in #360, the SPRT class is currently not able to handle the case where one or more counts (WDL) is 0. This simple fix implements a Dirichlet([0.5, 0.5,...

Implement pseudo random sampling of positions

1

Motivation ---------- Currently, we sample positions independently with a certain probability. This can cause clusters of nearby positions which are very similar and could affect generalization of the network. By...

[WIP] Add Dockerfile for compiling lc0 on linux

1

This is a first draft for a Docker file which we can use to allow users to more easily compile lc0 (which turns out to be non-trival as we can...