Hyperparameter settings

[anirudh1666]

Hi,

This branch has a bunch of changes that are ready for merging. Here is a quick overview:

• Added a new clusterer called adaptive max pooler which adaptively expands out the cluster size as necessary. It still has room for improvement (switching from bins to k-dist matrix to locate neighbours) but that can be completed after analysis of results and writing of paper. Tests are also included.
• Normalisation is added. Normalisation does not improve performance in most cases and leads to more clusters being discovered, so cluttered graphs at the expense of more discovered channels. Need to do more experiments during the analysis to see the full impact. Tests are included. Note: normalisation means mapping dataframes to [0, 1] and sums to 1 rather than Normal(0, 1)
• The optimal hyperparameters are found and are now the default values for each denoiser. Tests updated.
• Clustering on the latency dimension solely can actually improve performance as the latency dimension is far more useful in clustering than the magnitudes. It is now an option. Tests added.
• Posterior pooling is global max pooling after denoising to improve transparency of graphs. Added as a new parameter along with tests.
• data_tools contains useful functions for analysis, e.g., copy_hemisphere to copy noisy hemisphere to denoised so you can compare them; k-dist graphs for selecting optimal hyperparameters for DBSCAN; etc. Tests are added.

[anirudh1666]

Note: When merging with main branch, please do squash and merge, so we don't pollute the commit history.