Defaults in python interface

[tvandera]

• No default for noise model. It seems I need to specify e.g. precision = 5.0
• LambdaBeta is different in Python IF compared to C++

[ipasechnikov]

There are no default values for all noise models because you set noise type based on whether its value is set or not. I suppose we can just add new argument like noise_type to set noise type. Then we can have noise values set by default.

https://github.com/ExaScience/smurff/blob/5f4891b94f52ca5741e2e0691864f5ba296fd700/python/smurff/smurff.pyx#L309-L322

[ipasechnikov]

Add two parameters to Python smurff function: Ynoise - noise tuple for Y matrix, prior_noises - list of noise tuple for each prior in priors list.

Commit 8c2c75ddb35c428b40f7cc5cf3b7d12922d104f4 has all these changes.

Here is an example how to use these parameters: https://github.com/ExaScience/smurff/blob/5ba7b683035542b89536af979ad3a61d81bed722/python/smurff/test/test_smurff.py#L94-L103

You can omit Ynoise parameter. Default C++ noise would be used. As for prior_noises, you can leave them empty, then default noise would be used for every prior. Also you can set any element of prior_noises list to None, then corresponding prior would use default noise.

[ipasechnikov]

Heavily related to issue #105. Noise parameters are updated and should work fine. Tests, examples and jupyter notebook are fixed and use new noise parameters.

[ipasechnikov]

Replaced prior_noises parameter with side_info_noises and aux_data_noises.

side_info_noises is a list of noise tuples. You can omit noise model for any side info by setting it as None, then a default C++ noise model would be used.

# side_info_noises example
[('fixed', 1.0, None, None, None), None, ('adaptive', None, 0.1, 1.0, None)]

So for the first side info fixed noise with precision set as 10 would be used. For the second side info it will use default noise model and for the last one it would be an adaptive noise model with initial signal-to-noise ratio 0.1 and max signal-to-noise ratio 1.0.

aux_data_noises is a list of lists of noise models.

# aux_data_noises example
[ [('fixed', 1.0, None, None, None), None], 
  [None, ('probit', None, None, None, 0.5)], 
  None, 
  [],
  [('adaptive', None, 0.5, 1.0, None)] ]

• 1st dimension has 2 aux-data, first one will use fixed noise model with precision of 10, second one would be a default noise model.
• 2nd dimension also has 2 aux-data. First one is default and second one is probit noise with 0.5 threshold.
• 3rd dimension will use default noise model for any aux-data
• 4th dimension will also use default noise model for any aux-data
• 5th dimension has only 1 aux-data and it uses adaptive noise model with initial signal-to-noise ratio 0.5 and max signal-to-noise ratio 1.0