Implemented dbscan for RGDR

[BSchilperoort]

Implements dbscan clustering functionality. The code included in here is sufficient to reproduce the prototype_RGDR notebook.
See the wip notebook

---

From @geek-yang:

• Implement DBSCAN clustering function
• Check if the longitude is between -180 : 180 and convert to this range if not
  not required, haversine uses radians.
• Handle input with lags dimension to vectorize DBSCAN clustering and keep labels
  --> currently supports n_targets == 1, see Supporting multiple target intervals in map_correlation. #56
• Handle clustering for +/- correlation separately
• Mask data ( [lat, lon] pairs) using significance (p_value)

When these methods are implemented, we also need to validate the results:

• Check if it is necessary to have weight by area for DBSCAN calculation (the impact of area is already reflected by the haversine distance)
  weighing by area is required for reducing clusters to the mean of each cluster. This is implemented
• Add relevant checks for the results and raise warnings (e.g. check whether the results are noise or not).
• Verify DBSCAN clustering results

Note: For RGDR we will want to support grouping the DBSCAN operations over splits, due to the computational intensity when working with full resolution data, see #58

This PR fixes #55

---

Still to be implemented, but out of scope for this PR:

• Investigate currently hard-coded latitude & longitude dimensions. Do we want to make this more dynamic?
• RGDR wrapper that uses the functionality implemented in this PR
  • We need to discuss what structure this will have, e.g. how to show the cluster maps to the user.
• Link between RGDR and train/test splits.
• Supporting multiple target intervals in map_correlation. #56
• Support grouping over splits in rgdr #58
• Tracking clusters over different splits, as the clusters can differ between folds.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[BSchilperoort]

Note that the Python 3.7 Windows build fails. This is due to an issue with the netcdf4 package, which has become a dependency to be able to use the example data from the proto.

To silence this issue we can specify the version for netcdf4, which might resolve the error. However, for how long do we want to keep supporting Python 3.7? It will reach its end of life in a couple of months.

[Peter9192]

Nice work! Perhaps combine all in one module (rgdr.py) and consider adding a transform method.

[sonarqubecloud]

Kudos, SonarCloud Quality Gate passed!   

0 Bugs
0 Vulnerabilities
0 Security Hotspots
0 Code Smells

96.6% Coverage
0.0% Duplication

[Peter9192]

Nice, let's follow it up in the next PR