Feature/dl cluster splitting release

[imawby]

Hello hello,

In this PR I add the cluster splitting model training scripts to the LArMachineLearningData repository. There are two models used in the cluster splitting algorithm: 1) ContaminationModel, used to determine whether a cluster 'window' is shower-like, contaminated or not contaminated, and 2) SplitPointModel, used to classify each position in a window as a signal/background split point. The scripts should be ran in this order:

1. Windows.ipynb: Used to slim/filter the DLThreeDClusterSplittingAlgorithm training tree.
2. MergeFiles.ipynb: Used to combine the output files from Windows.ipynb.
3. EncoderTraining.ipynb: Used to train the ContaminationModel.
4. EncoderPerformance.ipynb: Used to obtain classification score/confusion matrices for a a specified ContaminationModel.
5. EncoderDecoderTraining.ipynb: Used to train the SplitPointModel.
6. EncoderDecoderPerformance.ipynb: Used to obtain classification score/confusion matrices for a a specified SplitPointModel.

(The training of the encoder and encoder-decoder is orthogonal, so steps 3/4 and 5/6 can be swapped).

The other files are:

• Datasets.py: Contains the datasets used in the Encoder and EncoderDecoder trainings.
• Models.py: Contains the ContaminationModel and SplitPointModel definitions.
• TrainingMetrics.py: Contains functions used to draw training metric plots i.e. the confusion matrix and classification score plots.
• Utilities.py: Contains the functions used to create the time-series windows.

Let me know if you have any questions!

[AndyChappell]

Hi Isobel, thanks for the PR. Mostly formatting and stylistic requests, with a little refactoring and one more substantive point re garbage collection. Otherwise I think this is looking good.

[AndyChappell]

Updates look good, thanks Isobel.