To Do List (by Method)

Madpack:

• Add info output about target db and schema during installation/updates etc.

K-means clustering:

• DONE: Allow dense arrays as input (not SVEC only)
• Allow the method to start processing a source table w/o a "PointID" column (now it needs both PID and POSITION columns)
• Implement other distance measures for k-means.
• Fix the "goodness of fit" test to be scale, rotation and transition invariant
• Rewrite the following plpgsql stored procedure into C:
  • udf: __kmeans_bestCentroid
  • uda: __kmeans_meanPosition
• Overload main kmeans_run pl/python function with a 2nd version that takes additional dictionary argument to overwrite the algorithm constants (like: sampling_size, max_iterations, etc)

SVD Matrix Factorisation:

• Add multi-user / multi-session support (output tables based on RUN_ID parameter).
• Modify API to take a table/view name with predefined columns, instead of table and column names as parameters.
• Add validation for all parameters.
• Overload main svdmf_run pl/python function with a 2nd version that takes additional dictionary argument to overwrite the algorithm constants (like: original_step, num_iterations, etc)
• Convert all support tables to temporary tables.

SVEC

• Support for other base types.
• Indexing on svecs. This requires indexing capability on arrays, which is currently unsupported in GP.

Kernel-machines:

• Adjust documentation to our standards.
• Add support for sparse vectors (now it's only array of float8s).
• Prefix support functions with "__"
• Review API, add support for multi-user/session environment.

Quantile:

• Convert to Python
• Add parameter validation

Logistic Regression:

• Add measures for the goodness of fit