Skip to content

docs(forecasting): replace wrong numeric sequence-length/horizon defaults with the real sources - #67

Merged
LukasWodka merged 2 commits into
mainfrom
docs/forecasting-window-defaults
Jul 23, 2026
Merged

docs(forecasting): replace wrong numeric sequence-length/horizon defaults with the real sources#67
LukasWodka merged 2 commits into
mainfrom
docs/forecasting-window-defaults

Conversation

@LukasWodka

@LukasWodkaLukasWodka commented Jul 23, 2026

Copy link
Copy Markdown
Contributor

Summary

The forecasting accordion in join-use-case/how-training-works.mdx claimed the sliding-window sequence length "you set in the notebook" defaults to 60, and forecast horizon to 1. Neither number is a platform default — 60 is only the engine's internal parameters.get("sequence_length", 60) fallback, which platform runs can never hit (the backend always sends the stored value in experiment_parameters).

What a no-touch forecasting experiment actually runs with (verified end-to-end on develop, 2026-07-23):

  1. Model-zoo template → the template's module-level declaration wins: lstm/gru/rnn declare 60/1, but patch_tst declares 96/24 and timesfm512/128. The SDK's upload rewriter extracts and retains these for time-series categories, and they flow through the experiment payload to the engine.
  2. Custom model with no declaration → the SDK sends nothing (None is dropped from the form-encoded POST) and the backend's model defaults apply: 25/1.
  3. reset_training_plan() before a successful start() (which happens automatically after a failed submission) → the SDK re-seeds 24/12 and submits those.

So "default 60"/"default 1" is accidentally right only for path 1 with the basic templates. This PR replaces the numeric claims with the actual sources (template declarations / notebook setters) and points readers at the experiment view — the same approach the new TSC section takes in #66.

Canonicalizing one cross-layer default (and aligning SDK/backend/engine) is tracked in tracebloc/backend#1196; once the team ratifies a number, the docs can state it.

Type

  • Docs

Test plan

  • Wording verified against develop-state code in tracebloc-py-package (validation/rewriter.py, training/submitter.py, link_model_dataset.py), backend (metaApi/models/Experiment.py, common/utils/bus_client_util.py) and tracebloc-engine (core/domains/time_series/base.py)
  • Touches a different accordion than docs: add the missing time-series classification section to how-training-works #66 (same file, disjoint hunks — whichever merges second rebases cleanly)

🤖 Generated with Claude Code


Note

Low Risk
Documentation-only wording change with no runtime or API impact.

Overview
Time series forecasting docs in how-training-works.mdx no longer claim lookback sequence length defaults to 60 from the notebook or that forecast horizon defaults to 1 in a misleading way.

The sliding-window bullet now states that both values come from model template declarations or notebook settings, with platform fallbacks of 25 and 1 when nothing is set, and that templates often override those defaults—readers are directed to the experiment view for what actually ran.

Reviewed by Cursor Bugbot for commit 56edc87. Bugbot is set up for automated code reviews on this repo. Configure here.

…eal sources
The forecasting accordion claimed sequence length defaults to 60 and
forecast horizon to 1. Neither is a platform default: the effective
values come from the model template's module-level declarations
(lstm/gru/rnn declare 60/1, patch_tst 96/24, timesfm 512/128), from
the notebook setters, or — only when neither sets a value — from the
backend's stored defaults (25/1). The 60 in the engine is an
unreachable fallback on the platform path.
Mirror the TSC section's approach (#66): name where the values come
from and point readers at the experiment view instead of asserting a
number that is wrong for most paths.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
@mintlify

mintlifyBot commented Jul 23, 2026

Copy link
Copy Markdown
Contributor

Preview deployment for your docs. Learn more about Mintlify Previews.

ProjectStatusPreviewUpdated (UTC)
tracebloc🟢 ReadyView PreviewJul 23, 2026, 9:31 AM

💡 Tip: Enable Workflows to automatically generate PRs for you.

@cursorcursorBot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

Reviewed by Cursor Bugbot for commit 124133a. Configure here.

@LukasWodka

Copy link
Copy Markdown
ContributorAuthor

👋 Heads-up — Code review queue is at 32 / 30

Above the WIP limit. The team convention is to review existing PRs before opening new work.

Open PRs currently in Code review (oldest first):

  • .github#64 — fix(closure-router): add GH_TOKEN to closing-PR lookup so issues mirror their closing PR's Status · author: @LukasWodka · no reviewer assigned
  • backend#1153 — WS-C: delegate CheckModelMixin rewrite path to SDK ModelFileRewriter + pin tracebloc==0.14.0 (#1101) · author: @divyasinghds · reviewer: @shujaatTracebloc
  • backend#1166 — feat(global_meta): metadata-only backfill endpoint for pre-cutover datasets (di#378) · author: @divyasinghds · no reviewer assigned
  • backend#1188 — test(checker): vendor SDK rewrite golden corpus as WS-C contract gate (#1101) · author: @shujaatTracebloc · no reviewer assigned
  • backend#1190 — perf(celery): move task results off MSSQL to Redis, 24h TTL (#981) · author: @aptracebloc · reviewer: @saqlainsyed007
  • backend#1193 — Reject non-federatable sklearn estimators (KNN / GaussianProcess) at upload (#923) · author: @shujaatTracebloc · no reviewer assigned
  • cli#389 — fix(delete): verify the host-data wipe before printing ✔ (RFC-0003) · author: @saadqbal · reviewer: @LukasWodka
  • cli#390 — feat(cli): auto-update — nudge + tracebloc upgrade (F1) · author: @LukasWodka · no reviewer assigned
  • cli#391 — fix(push): stop progress bar colliding with staging log lines (D3) · author: @LukasWodka · no reviewer assigned
  • cli#392 — feat(install): prefer ~/bin when already on PATH (B2, RFC 0001) · author: @LukasWodka · reviewer: @divyasinghds

Pull from review before opening new work. (This is a nudge from the kanban WIP check, not a block.)

…emplate-override rule
Follow-up to the interim correction: now that backend#1196 has ratified
Option A, positively state the platform-canonical defaults (sequence
length 25, forecast horizon 1) and that model templates commonly declare
their own window and override them — rather than only removing the old
wrong numbers. Completes the docs item of the #1196 alignment checklist.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@divyasinghds

Copy link
Copy Markdown
Contributor

Pushed 56edc87: now that backend#1196 ratified Option A, this positively states the platform-canonical defaults (sequence length 25, forecast horizon 1) and the template-override rule, rather than only removing the wrong numbers. Completes the docs item of the #1196 checklist.

@mintlify

mintlifyBot commented Jul 23, 2026

Copy link
Copy Markdown
Contributor

Preview deployment for your docs. Learn more about Mintlify Previews.

ProjectStatusPreviewUpdated (UTC)
tracebloc🟡 BuildingJul 23, 2026, 9:30 AM

💡 Tip: Enable Workflows to automatically generate PRs for you.

@LukasWodka
LukasWodka merged commit 849982d into mainJul 23, 2026
4 checks passed
@LukasWodka
LukasWodka deleted the docs/forecasting-window-defaults branch August 14, 2026 13:54
Sign up for freeto join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants

@LukasWodka@divyasinghds