Multi-Plot Grids

Multi-plot creates a grid of subplots, each showing the same plot type for a different value of one or two fields. This is useful for comparing across dimensions that are hard to overlay on a single axes (e.g., different datasets, models, or sparsity levels).

Basic usage

Use -multi_plot_fields to split the plot into subplots:

# One row of subplots, one per sparsity level
malet-plot -exp_folder ./experiments/my_exp \
  -mode curve-epoch-val_accuracy \
  -multi_plot_fields 'sp' \
  -best_at_max

Each subplot shares the same x and y axes for easy comparison.

Two-dimensional grid

Specify two fields to create a row-column grid:

# Columns for sparsity, rows for noise level
malet-plot -exp_folder ./experiments/my_exp \
  -mode curve-epoch-val_accuracy \
  -multi_plot_fields 'sp noise' \
  -best_at_max

The first field controls columns, the second controls rows. The figure size scales automatically: [width * num_cols, height * num_rows].

Combining with multi-line

Multi-plot and multi-line can be used together. For example, show a grid of subplots (one per sparsity level) where each subplot contains multiple lines (one per optimizer):

malet-plot -exp_folder ./experiments/my_exp \
  -mode curve-epoch-val_accuracy \
  -multi_plot_fields 'sp' \
  -multi_line_fields 'optimizer' \
  -best_at_max

Supported plot types

All plot types support multi-plot with up to 2 fields, including heatmaps. This makes it possible to create a grid of heatmaps — one per model or dataset:

malet-plot -exp_folder ./experiments/my_exp \
  -mode 'heatmap-lr weight_decay-val_accuracy' \
  -multi_plot_fields 'model' \
  -filter 'step last' \
  -best_at_max