User-provided data and source files

In a sandbox get started with

uv run --with ansel-sh python forecast.py

User-provided data and source files

ansel-sh auto-discovers CSV files, first from ./data then from cwd. You may optionally specify a directory to discover files.

from ansel_sh import forecasts

results = forecasts(data_dir="/path/to/data")

Claude Cowork

User uploads land in ./uploads. That's a good place to start.

Reuse classified files

You may optionally reuse classified files across multiple forecasting runs.

from ansel_sh import files, forecasts

classified = files()

one = forecasts(classified)
two = forecasts(classified)

Running background tasks

bash -c 'your-command' > /tmp/task.log 2>&1 &
echo $! > /tmp/task.pid

# Check status, progress, logs
kill -0 $(cat /tmp/task.pid) 2>/dev/null && echo "Running" || echo "Done"
cat /tmp/task.log

Auditable output artifacts

All output artifacts are written to .ansel/<run_id>/ in the cwd. Each run gets an auto-incrementing ID. Artifacts you may find include:

• files_manifest.csv — classified input files
• series_manifest.csv — extracted time series
• training_snapshot.csv / testing_snapshot.csv — model training and testing data
• model_instructions.md — model configuration and instructions
• target_snapshot.csv — forecast target data
• forecast_result.csv — the final forecast output