Features/340 gaussian nb

[ClaudiaComito]

Description

Initiating submodule naive_bayes and class GaussianNB().

Implementation of Gaussian Naive Bayes classification along the scikit-learn lines. Known issues listed below.

Issue/s resolved: #340

Changes proposed:

• New submodule naive_bayes (on the same level as cluster and regression)
• New classifier naive_bayes.GaussianNB()
• Implemented functions GaussianNB().fit(), GaussianNB().predict(), GaussianNB().predict_proba() plus helper functions
• Implemented test_gaussiannb() for local and distributed mode. In distributed mode, only the case where split=0 is tested for the time being (labels are 1D). Different split configurations require resplit to work from 1 to 0 (cf. Resplit from 1 -> 0 gives wrong order #425).

Known issues:

• ht.average() should allow weighted average across multiple axes  #351 , Implement isin() (and in1d()) #450, logical._sanitate_close_input should check shapes, and broadcast before redistributing #457, get logical.isclose to work with unbalanced tensors #458, Implement sanitation module #468, test_gaussiannb fails when distributed on 7 processes #490
• Address case where X.split != 0. Redistribute? (needs Resplit from 1 -> 0 gives wrong order #425)
• Adapt documentation, conventions from sklearn to HeAT (I know about those missing dunderscores!) DONE
• Create demo

Type of change

• New feature (non-breaking change which adds functionality)
• Documentation update

Due Diligence

• All split configurations tested- NOT YET: Different split configurations require resplit to work from 1 to 0 (cf. Resplit from 1 -> 0 gives wrong order #425).
• Multiple dtypes tested in relavent functions
• Documentation updated (if needed)
• Updated changelog.md under the title "Pending Additions"

Does this change modify the behaviour of other functions? If so, which?

no

[coquelin77]

i made a few comments here. nothing crazy though. other than that it looks good to go

[coquelin77]

too many tabs and 1 line below this it looks like there is dead code

[ClaudiaComito]

too many tabs (black's doing) or too many nested conditional statements?

if former:
🤷‍♀️
else:
I'll think about it

[coquelin77]

i was more confused about the structure here. but after sleeping on it i see what you meant. i read it now with the meaning of after issue #425 is solved then the if statement above these lines and the raise within will both be removed. i misunderstood it before i think

[coquelin77]

possible dtype problem here.

[ClaudiaComito]

is var() supposed to return a float32? Even if x is float64?

[coquelin77]

it should not. i guess i missed a couple dtype calls here. can you add them for me? it should be just in this spot

[ClaudiaComito]

That's what I thought, so is this the possible dtype problem you were talking about? I guess I just misunderstood your first comment

[coquelin77]

i just also realized that the dtype of n_tot doesnt matter because it is only used internally

[coquelin77]

good work!