Home / Libraries / ML4T Diagnostic / Docs
ML4T Diagnostic
ML4T Diagnostic Documentation
Feature validation, strategy diagnostics, and Deflated Sharpe Ratio
Source on GitHub
Skip to content

Trade-SHAP Dashboard

The Streamlit dashboard presents trade-level SHAP results, statistical checks, worst trades, feature effects, and recurring error patterns.

Install

pip install "ml4t-diagnostic[dashboard]"

Run the checked example

Clone the repository, then run the public example:

streamlit run examples/trade_shap_dashboard_demo.py

The release test suite starts this file with Streamlit's application test runner and fails if the app raises an exception.

Load saved results

The sidebar accepts JSON only. Pickle input is not supported because loading a pickle can execute code. Treat uploaded JSON as untrusted data and validate its contents before sharing or archiving reports.

The normalized result can contain:

  • trade identifiers, timestamps, symbols, PnL, and percentage returns
  • entry and exit prices, duration, direction, and quantity
  • feature values and aligned SHAP values
  • aggregate statistical validation results
  • clustered error patterns and their supporting trades

The dashboard disables sections whose required data is absent.

Export

The application can export normalized trades and patterns as CSV and the full dashboard as HTML. CSV exports may begin with characters that spreadsheet applications interpret as formulas; sanitize them before opening untrusted exports in a spreadsheet.

Supported import surfaces

The application implementation lives under ml4t.diagnostic.evaluation.trade_dashboard. The compatibility module ml4t.diagnostic.evaluation.trade_shap_dashboard remains available for code written against beta releases. New applications should start from the checked example script so data normalization and Streamlit state follow the supported path.