opening the black box

[jeremymanning]

consistency of API

• [x] hyp.tools.normalize: add support for reduce and align

new data object

• [x] all functions: add support for a new "HypO" object that has the following fields:

  • .data: the data (a list, HypO object, pandas dataframe, or numpy array)
  • .reduce: dictionary of reduce model/args
  • .align: dictionary of align model/args
  • .normalize: string with normalization method (or None)
  • .xform: a function that can take new data (in any form) and applies the sequence of transformations implied by the reduce, align, and normalize args
  • .args: save out all non-xform arguments
  • .plot: a function that takes in new data (with no other arguments), transforms it with the .xform function, plots it using any specified args (.args), and returns a new HypO objects. And if no data is specified, use .data as the data.
  • .figure: handle to the figure
  • .axes: handle to the axes
  • .lineani: handle to the animation (None if animate=False)
  • .version: hypertools version number

• [x] in normalize, align, and reduce, add support for model/method = None (just return the original data). This will help with the .xform function.

caching

• [x] in normalize, align, and reduce, add a "save" function that saves the results of time-consuming operations (and pre-loads the answer if they exist). This requires computing a hash for the data and normalize/align/reduce arguments. (see notes in slack for more details)

selecting number of components

• [x] hyp.tools.describe: extend describe_pca to support other models

[andrewheusser]

still need to:

• [x] update docstrings
• [x] check readthedocs
• [x] check readme for API changes
• [x] write the save function
• [x] write tests for DataGeometry
• [x] make sure deprecated functions/args are still working
• [x] check that deprecated stuff throws warnings
• [x] update version number
• [x] write up all the changes made and make a release
• [x] push to pip