How-to Guide Contents

[BalzaniEdoardo]

The old notes in the how to guide needs to be revisited with the diataxis classification in mind:

1. Short and to the point
2. Solve a specific problem
3. Assume already background knowledge (liking to some additional resources for background or a real data application)

Proposal for edits (to be discussed in NeMoS meetings):

• plot_02_glm_demo.md:

  • Consider if we need a basic glm.fit demo (the quickstart should be sufficient)
  • Split in multiple how-tos:
    1. basic glm? (probably not needed)
    2. K-fold cross-validation (basic, select a regularizer)
    3. simulate neurons : show how generate coupling filters and simulate neural pop; highlight the potential instability of the dynamical system and what to tweak to have a sable simulation regime.

• plot_03_population_glm.md:

  • Basic model with one predictor, show the coefficient shape
  • Using masks to select features
  • Add interactions (also, point to how-tos and head dir tutorial as reference)

• plot_04_batch_glm.md:

  • for now no ok, but after the fit_stochastic we should discuss to compare different algorithms (SVRG vs GD)

• plot_06_sklearn_pipeline_cv_demo.md:

  • Split content content by Moving the sklearn pipeline discussion and the note on K-fold as two sessions in the background
  • Restructure residual content:
    • Section 1: cross-validate an atomic basis (1 predictor type only) setting basis attributes.
    • Section 2: cross-validate the basis type itself.
    • Section 3: cross-validate a composite basis. Mention the exponentially growing grid of parameters when many basis components are used. Mention alternative cv methods (random searches or similar).