Improve the reporting for EDA and model performance

[vaibs-d]

Is your feature request related to a problem? Please describe.
A suggestion for doing this would be to use https://github.com/StructuredLabs/preswald for creating better reports for model performance and data analysis.

[ptdatta]

Hey @vaibs-d , Can I work on it?

[ptdatta]

Hi @vaibs-d ,

Since Plexe follows a config-driven, modular workflow using YAML + Python callbacks (e.g., MLFlowCallback, ProviderConfig), we could extend the system with an optional PreswaldReportCallback.

Why Preswald? What It Offers Beyond Plexe Console

1. Interactive, In-Browser Dashboards
   Preswald turns Python into interactive UIs (tables, sliders, filters) without needing JS or HTML.
   Unlike the static plots in the Plexe console, Preswald dashboards allow exploration within the report.

2. Reactive State & DAG Execution
   User interactions trigger only necessary updates via DAG-based tracking.
   Plexe's static pipeline doesn’t support such interactivity or live recomputation.

3. Instant Python-to-UI Conversion

text("EDA Summary")
slider(...)
view(df)

This becomes a polished report in seconds. No front-end setup required.

4. Shareable, Self-Contained Outputs
   Reports are standalone .html files—perfect for sharing with stakeholders without needing backend hosting or login.

High Level Implementation Plan

• Callback Integration: Add a PreswaldReportCallback to the plexe.callbacks module.

• Trigger Points: Use on_train_end and/or on_evaluate_end to gather:

  • EDA insights (missing values, distributions, correlations)

  • Model metrics (accuracy, confusion matrix, SHAP, etc.)

• Export: Generate a rich, interactive .html report using preswald.export(...).

• YAML Config: Allow users to enable this in their pipeline config under the callbacks: section, like so:

callbacks:
  - _target_: plexe.callbacks.PreswaldReportCallback
    export_path: "reports/eda_model_report.html"

[vaibs-d]

Hi @vaibs-d ,

Since Plexe follows a config-driven, modular workflow using YAML + Python callbacks (e.g., MLFlowCallback, ProviderConfig), we could extend the system with an optional PreswaldReportCallback.

Why Preswald? What It Offers Beyond Plexe Console

1. Interactive, In-Browser Dashboards
   Preswald turns Python into interactive UIs (tables, sliders, filters) without needing JS or HTML.
   Unlike the static plots in the Plexe console, Preswald dashboards allow exploration within the report.
2. Reactive State & DAG Execution
   User interactions trigger only necessary updates via DAG-based tracking.
   Plexe's static pipeline doesn’t support such interactivity or live recomputation.
3. Instant Python-to-UI Conversion

text("EDA Summary")
slider(...)
view(df)
This becomes a polished report in seconds. No front-end setup required.

4. Shareable, Self-Contained Outputs
   Reports are standalone .html files—perfect for sharing with stakeholders without needing backend hosting or login.

High Level Implementation Plan

• Callback Integration: Add a PreswaldReportCallback to the plexe.callbacks module.

• Trigger Points: Use on_train_end and/or on_evaluate_end to gather:

  • EDA insights (missing values, distributions, correlations)
  • Model metrics (accuracy, confusion matrix, SHAP, etc.)

• Export: Generate a rich, interactive .html report using preswald.export(...).

• YAML Config: Allow users to enable this in their pipeline config under the callbacks: section, like so:

callbacks:

• target: plexe.callbacks.PreswaldReportCallback
  export_path: "reports/eda_model_report.html"

Looks good to me! Exited to see this working!