fit_per_origin

Back to L4.5 | Browse all axes | Browse all options

Axis fit_per_origin on sub-layer L4_5_A_in_sample_fit (layer l4_5).

Sub-layer

L4_5_A_in_sample_fit

Axis metadata

  • Default: 'last_origin_only'

  • Sweepable: False

  • Status: operational

Operational status summary

  • Operational: 3 option(s)

  • Future: 0 option(s)

Options

all_origins – operational

Compute fit views for every OOS origin.

L4.5.A fit-per-origin cadence all_origins.

This option configures the fit_per_origin axis on the L4_5_A_in_sample_fit sub-layer of L4.5; output is emitted under manifest.diagnostics/l4_5/L4_5_A_in_sample_fit/ alongside the other selected views.

When to use

Detailed walk-forward audit; expensive but thorough.

References

  • macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’

Related options: every_n_origins, last_origin_only

Last reviewed 2026-05-05 by macroforecast author.

every_n_origins – operational

Compute fit views every n origins.

L4.5.A fit-per-origin cadence every_n_origins.

This option configures the fit_per_origin axis on the L4_5_A_in_sample_fit sub-layer of L4.5; output is emitted under manifest.diagnostics/l4_5/L4_5_A_in_sample_fit/ alongside the other selected views.

When to use

Compromise between coverage and runtime; params.every_n controls cadence.

References

  • macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’

Related options: all_origins, last_origin_only

Last reviewed 2026-05-05 by macroforecast author.

last_origin_only – operational

Compute fit views only for the last training window.

L4.5.A fit-per-origin cadence last_origin_only.

This option configures the fit_per_origin axis on the L4_5_A_in_sample_fit sub-layer of L4.5; output is emitted under manifest.diagnostics/l4_5/L4_5_A_in_sample_fit/ alongside the other selected views.

When to use

Default; cheapest summary while still capturing the most-informative model state.

References

  • macroforecast design Part 4: ‘diagnostic layers default-off; non-blocking; produce JSON + matplotlib views attached to manifest.diagnostics/.’

Related options: all_origins, every_n_origins

Last reviewed 2026-05-05 by macroforecast author.