SSML Models Implementation

[stompsjo]

This PR introduces much of the code used for my research comparing several different semi-supervised machine learning models (SSML). I wrote and used these models quite separate from how I have used RadClass in the past (i.e. I ran experimental data through RadClass and H0 to get observations I could use for training and testing ML models separately in a downstream analysis).

This raises a design question regarding how these models and scripts should be integrated into RadClass. Should they be made into a post_analysis object? Should they remain separate, thereby making the user have a script that both runs RadClass for data processing and SSML models for training and testing?

• Another note that popped into my head: should I include a .load method for each model class? This may remove one manual step a user would have to take to use joblib for loading a written model and storing as a model class.

[stompsjo]

One problem to note: creating tests for neural networks has been tricky. Ideally, these tests should not take several minutes to train and test a method/model. But if we want meaningful test-cases, we will want to test meaningful neural network architectures (which take longer to train).

I will be thinking of a compromise...

[stompsjo]

The most recent commit changes each fresh_start method for each ML model to use the parent class train method rather than copying its logic outright. This should help to make each class concise and more comprehensible as a whole.

[gonuke]

Split into more PRs?

[stompsjo]

Suggested change

	fig, axes = plt.subplots(n, n, figsize=(15, 15))
	fig, axes = plt.subplots(n, n, figsize=(3*n, 3*n))

[coveralls]

Pull Request Test Coverage Report for Build 3363647923

• 533 of 534 (99.81%) changed or added relevant lines in 6 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage decreased (-0.1%) to 99.868%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
models/SSML/ShadowCNN.py	142	143	99.3%

Totals
Change from base Build 3348444582:	-0.1%
Covered Lines:	754
Relevant Lines:	755

---

💛 - Coveralls

[stompsjo]

This PR has been split into #41, #42, #44, and #45