Skip to content

JOURNAL.md

Latest commit

 

History

History
34 lines (25 loc) · 1.45 KB

JOURNAL.md

File metadata and controls

34 lines (25 loc) · 1.45 KB
Raw

JOURNAL

Status

  • Project / dataset: IBM breast cancer workshop — data/public.csv (450 labeled samples, 30 numeric features + target)

  • Goal: maximize ROC-AUC on data/public.csv, then evaluate on data/test/ and push to Skore Hub

  • Last experiment: 01_baseline — done (pipeline reset to dummy baseline)

  • Last result: re-run after pipeline fix

  • Workspace decisions (immutable unless the user pivots):

    • tabular library: pandas — recorded: 2026-06-24
    • env manager: uv — recorded: 2026-06-24
    • package name (src/<pkg>/): ibm_workshop — recorded: 2026-06-24
    • skore mode: hub — recorded: 2026-06-24
    • skore hub workspace: ibm-workshop/competition — recorded: 2026-06-24
    • CV splitter family: hold-out 0.2 — recorded: 2026-06-24

Data understanding (EDA)

  • Status: done — 2026-06-23
  • Summary: 450×32 numeric tabular set (data/public.csv); no missing values; binary target imbalanced (136×0, 314×1); no datetime/group structure; strongest target links on radius/concavity features; tight collinearity among size-related columns. Drop sample_id from features.
  • Report: data/eda/eda.md

History

Stem Intent (one line) Status Headline result Design note
01_baseline Uniform random classifier, ROC-AUC hold-out done re-run after fix design note

Backlog

# Item Source