ramsey-devs / ramsey Goto Github PK

Models to be implemented:

• clustering processes
• centered partition processes
• recent NP developments

Implement stabler solutions than the currently implemented predict and marginal functions of GP.

Add info on precommit, code-coverage , tests, etc.

The current data iterators are slow. We could use the tf.data iterators.

Does that work with dynamic input/output dimensions?

Not sure which is better

__call__(self, method="predict", **kwargs):

or dispatch based on number of args:

__call__(self, x **kwargs):
  if "y" in kwargs:
      return self._this_method(x, kwargs["y"]):
  return return self._that_method(x)   

Check what Haiku/Flax/.. recommend.

Code coverage does not update properly (wrong link in README) and does not show coverage reports on PRs.

Add a high-level API, that contains methods to train/predict every model easily, following GPFlow and the like....

This is not good. Rewrite like this and document the fn:

 shuffle_key, rng_key = random.split(rng_key)
    shuffle_idxs = random.choice(shuffle_key, jnp.arange(n), shape=(n,), replace=False)
    if shuffle:
        data = ctor(*[el[shuffle_idxs] for _, el in enumerate(data)])

    y_train = ctor(*[el[:n_train] for el in data])
    y_val = ctor(*[el[n_train:] for el in data])
    train_rng_key, val_rng_key = random.split(rng_key)

• fix codecov
• fix documentation not showing
• replace all old links (one in index.rst to dirmeier/...)
• beautify code
• gitlint ini, pylintrc, jupytext file, mypy init, pytest ini, setup cfg ,manifest, makefile, pydocstyle, bandit.yaml
• code of conduct, contribuing,
• codecov/coveragerc