Data Analysis Toolkit

Problem

When working with CSV or Excel data, the hard part is not charting but stitching a repeatable workflow: unstable dtypes, missing values, schema checks, test selection, time-series frequency, and scattered deliverables. Data Analysis Toolkit provides a composable Python layer for the analysis lifecycle: load → profile → validate → clean → transform → analyze → visualize → report → export, with a cli.py front end.

How it works

Modules are organized by responsibility. profiler.py inspects shape, dtypes, missingness, and column statistics; validator.py checks a DataFrame against a JSON schema for dtypes, ranges, uniqueness, allowed values, and regex; cleaner.py standardizes names, coerces types, drops empties and duplicates, imputes missing values, and records actions; statistics.py covers descriptive and inferential tests, with compare_groups choosing parametric or non-parametric tests from a normality check; ml.py provides baseline RandomForest classification/regression and KMeans clustering; timeseries.py supports resampling, rolling statistics, seasonal decomposition, ADF stationarity, and Holt-Winters forecasting; report.py and export.py produce Markdown/HTML reports plus CSV, JSON, Parquet, Excel, or multi-sheet workbooks. Charts use a headless Matplotlib backend, so it runs in servers, CI, or displayless environments.

Boundaries

It is useful for rapid EDA, data-quality checks, baseline modeling, and report generation, but it is not a production feature store, MLOps platform, or rigorous statistical modeling framework. When composing steps manually, keep the order profile → clean → model and surface the cleaner.clean action report. Time-series decomposition and forecasting require a regular frequency via freq=; irregular indexes can affect resampling and interpolation. Treat ML metrics as a signal check, not a final verdict.