README.md

metadata

title: Solar_Culient_Predictor
app_file: enhanced_app.py
sdk: gradio
sdk_version: 4.26.0

SOLAI Scoring Dashboard (Gradio)

A lightweight UI to train a baseline logistic regression on your solar leads dataset and generate probability_to_buy predictions. Uses the same feature candidates and preprocessing approach as scripts/batch_scoring.py.

• Default dataset: examples/synthetic_v2
• Outputs are always written to /Users/git/solai/scores and are also downloadable from the UI.

Features

• Choose data source:
  • Use preset example data: examples/synthetic_v2/leads_features.csv and examples/synthetic_v2/outcomes.csv
  • Upload your own CSVs (features and outcomes)
• Train + score with a single click
• Evaluation metrics (test split):
  • ROC AUC, PR AUC, Brier score (gracefully handles degenerate label cases)
• Preview:
  • predictions.csv (lead_id, probability_to_buy)
  • leads_features_scored.csv (features merged with probability_to_buy)
• Download both files from the UI in addition to saving to disk (/Users/git/solai/scores)

Requirements

• Python 3.9+ recommended
• macOS (as per environment), should also work on Linux/Windows

Install dependencies (ideally in a virtual environment):

python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate
pip install -r dashboard_gradio/requirements.txt

Run the App

python dashboard_gradio/app.py

Gradio will launch on a local URL (typically http://127.0.0.1:7860). Open it in your browser.

Usage

1. Start the app.
2. Select a data source:
   • Default: “Use example synthetic_v2”
   • Or switch to “Upload CSVs” and provide:
     • Features CSV (must include lead_id and a subset of feature columns listed below)
     • Outcomes CSV (must include lead_id and sold columns)
3. Click “Train + Score”.
4. Review metrics and preview tables.
5. Download the generated files or find them on disk under /Users/git/solai/scores.

Expected Columns

• Features CSV must contain lead_id and some subset of these candidate features:

  • living_area_sqft
  • average_monthly_kwh
  • average_monthly_bill_usd
  • shading_factor
  • roof_suitability_score
  • seasonality_index
  • electric_panel_amperage
  • has_pool
  • is_remote_worker_household
  • tdsp
  • rate_structure
  • credit_score_range
  • household_income_bracket
  • preferred_financing_type
  • neighborhood_type

• Outcomes CSV must contain:

  • lead_id
  • sold (0/1)

Outputs

Saved to /Users/git/solai/scores with a timestamp suffix:

• predictions_YYYYMMDD_HHMMSS.csv
  • Columns: lead_id, probability_to_buy
• leads_features_scored_YYYYMMDD_HHMMSS.csv
  • Original features merged with probability_to_buy

Both files are also offered as downloads directly in the UI.

Notes and Troubleshooting

• If the outcomes data has only a single class (all sold=0 or all sold=1), ROC AUC and PR AUC are undefined; the app shows “N/A” for those metrics but still computes Brier score and produces predictions.
• If you see “No candidate features found”, ensure your features CSV contains at least one of the listed feature names.
• If port 7860 is in use, Gradio will choose another port automatically, displayed in the terminal.
• For larger datasets, training time may increase but should remain quick for typical CSV sizes.

Development

• Core logic is in dashboard_gradio/app.py.
• The pipeline mirrors scripts/batch_scoring.py: ColumnTransformer with passthrough numeric features and OneHotEncoder for categoricals, then LogisticRegression.
• Extend easily with additional visualizations (e.g., calibration plots), feature importance, or a data dictionary viewer.