MicrogradEx.PlotData (MicrogradEx v0.1.0)

Converts MicrogradEx datasets and training runs into plain plotting rows.

The returned values are ordinary lists of maps. They are designed to be consumed by Livebook, Vega-Lite, CSV exporters, or tests without making the core library depend on any plotting package.

Summary

Functions

accuracy_history(other)

Converts training accuracy into percentage rows for charts.

dataset_points(other)

Converts dataset points into chart-friendly rows.

decision_boundary(model, dataset, opts \\ [])

Evaluates a model over a padded two-dimensional grid.

loss_history(other)

Expands training history into loss metric rows.

training_history(other)

Returns full training-history rows with percentage accuracy added.

Functions

accuracy_history(other)

@spec accuracy_history(MicrogradEx.Trainer.Run.t()) :: [map()]

Converts training accuracy into percentage rows for charts.

dataset_points(other)

@spec dataset_points(MicrogradEx.Datasets.Dataset.t()) :: [map()]

Converts dataset points into chart-friendly rows.

Numeric labels are preserved as :label_value, while :label is a friendly legend string.

decision_boundary(model, dataset, opts \\ [])

@spec decision_boundary(
  MicrogradEx.NN.model(),
  MicrogradEx.Datasets.Dataset.t(),
  Keyword.t()
) :: [
  map()
]

Evaluates a model over a padded two-dimensional grid.

Options:

:h - positive grid spacing, default 0.25
:padding - non-negative padding around dataset bounds, default 1.0

loss_history(other)

@spec loss_history(MicrogradEx.Trainer.Run.t()) :: [map()]

Expands training history into loss metric rows.

training_history(other)

@spec training_history(MicrogradEx.Trainer.Run.t()) :: [map()]

Returns full training-history rows with percentage accuracy added.