Spaces:
Running
Running
Commit History
-
Ensemble evaluation: complementary failure, zero missed scams, three-state design
9470bcc -
v2 fine tune: length bias fixed, Nigerian accuracy regressed; do not deploy
7e72e1d -
Build train/val/test splits with template grouping; document dataset and leakage controls
b624ee7 -
Add ExAIS loader and fraud/marketing triage; ExAIS spam is telco bulk, not fraud
c4f6a67
-
Add SMS collection schema, labelling guide and validator
d69d436 -
Add length experiment: model classifies document shape, not fraud
23c24d1 -
Add Nigerian SMS evaluation set, eval harness and failure analysis: 80% accuracy vs 99.24% on public corpora
b0e32f1
-
Day 2: add dataset download script and document data provenance
- Add scripts/download_data.py to fetch UCI SMS Spam, Enron-Spam, SpamAssassin, and Kaggle Fraudulent Email Corpus - Pin requests and kaggle in requirements.txt - Document dataset sources, sizes, and licenses in data/raw/README.md - Raw data files remain gitignored; pipeline is reproducible via the script81f6cf7 -
Initialize Wilma project structure and dependencies
bbd7f6a