f2a — File to Analysis

One line of code → Full statistical analysis + interactive HTML report. 24+ file formats, HuggingFace datasets, 6 languages, 20+ analysis modules, 50+ visualizations.

Generated from f2a.analyze("lerobot/roboturk") — a single line of code.

Live Sample Report

📊 View Sample Report (lerobot/roboturk) ← GitHub Pages (recommended)

A fully self-contained interactive HTML report generated from the lerobot/roboturk dataset.

Alternative: Download raw HTML and open in your browser.

Installation

pip install f2a

For advanced analyses (UMAP, ADF tests):

pip install f2a[advanced]

Quick Start

import f2a

# ── Local files ──────────────────────────────────────
report = f2a.analyze("data/sales.csv")
report.show()                    # Print summary to console
report.to_html("output/")       # Save interactive HTML report

# ── HuggingFace datasets ────────────────────────────
report = f2a.analyze("https://huggingface.co/datasets/imdb")
report = f2a.analyze("hf://imdb")
report = f2a.analyze("imdb")    # org/dataset pattern auto-detected

# ── Access results ───────────────────────────────────
report.stats.summary             # Descriptive statistics (DataFrame)
report.stats.correlation_matrix  # Correlation matrix
report.stats.advanced_stats      # Advanced analysis results
report.schema.columns            # Column type information
report.to_dict()                 # Everything as a dictionary

Example: Analyzing a HuggingFace Dataset

import f2a

report = f2a.analyze("https://huggingface.co/datasets/lerobot/roboturk")

shape: (187507, 11) | subsets: 1
  default/train: (187507, 11)

report.show()

╔══════════════════════════════════════════════════════════╗
║  f2a Analysis Report — lerobot/roboturk                 ║
╠══════════════════════════════════════════════════════════╣
║  Rows: 187,507  ·  Columns: 11                         ║
║  Numeric: 9  ·  Categorical: 0  ·  Text: 0             ║
║  Datetime: 0  ·  Boolean: 0                             ║
╚══════════════════════════════════════════════════════════╝

# Save interactive HTML report (2.5 MB self-contained file)
path = report.to_html("output/")
print(path)
# → output/lerobot_roboturk_20260317_090024_report.html

📊 View this report live

# Access statistics programmatically
report.stats.summary
#          timestamp   episode_index  frame_index  ...
# count   187507.00      187507.00    187507.00   ...
# mean         ...           ...          ...     ...
# std          ...           ...          ...     ...

report.stats.correlation_matrix
#                   timestamp  episode_index  frame_index  ...
# timestamp          1.000000      0.978193     0.054412  ...
# episode_index      0.978193      1.000000    -0.003887  ...

# Advanced analysis results
report.stats.advanced_stats.keys()
# dict_keys(['advanced_distribution', 'advanced_correlation', 'clustering',
#            'dimreduction', 'feature_insights', 'advanced_anomaly', ...])

Multi-Subset HuggingFace Datasets

Datasets with multiple configs and splits are automatically discovered and analyzed.

report = f2a.analyze("FINAL-Bench/ALL-Bench-Leaderboard")
print(f"Total: {report.shape[0]} rows across {len(report.subsets)} subsets")

for s in report.subsets:
    print(f"  {s.subset}/{s.split}: {s.shape}")

# Load specific subset
report = f2a.analyze("FINAL-Bench/ALL-Bench-Leaderboard", config="agent", split="train")

The HTML report generates tabbed navigation — each subset/split gets its own analysis page.

HTML Report Features

report.to_html() generates a single self-contained HTML file (no external dependencies) with:

📑 Two-Depth Tab Navigation

[Subset/Split Tabs]
  └── [Basic] | [Advanced]
        ├── Basic: 13 analysis sections
        └── Advanced: 10 advanced analysis sections

🎯 Interactive Elements

📖 Beginner-Friendly Descriptions

Every section and every metric includes:

• Detailed modal descriptions with HTML formatting, examples, and analogies
• Beginner tips (초심자 팁 / Anfänger-Tipp / Conseil débutant / 初心者向けヒント / 初学者提示)
• Interpretation guidance — what does this number actually mean?
• All descriptions are fully translated into 6 languages (not machine-translated placeholders)

Analysis Modules

Basic Analysis (13 sections)

Advanced Analysis (10 sections)

Visualizations (50+)

All charts are inline base64 PNG — no external image files needed.

Supported Formats (24+)

Configuration

from f2a import AnalysisConfig

# ── Preset configs ───────────────────────────────────
config = AnalysisConfig.fast()        # Skip PCA, feature importance, advanced
config = AnalysisConfig.minimal()     # Descriptive + missing only
config = AnalysisConfig.basic_only()  # All basic on, all advanced off

# ── Custom config ────────────────────────────────────
config = AnalysisConfig(
    advanced=True,
    clustering=True,
    advanced_anomaly=True,
    statistical_tests=True,
    insight_engine=True,
    cross_analysis=True,
    ml_readiness=True,
    outlier_method="zscore",          # "iqr" (default) or "zscore"
    outlier_threshold=3.0,            # Z-score cutoff
    correlation_threshold=0.9,        # High-correlation warning threshold
    pca_max_components=10,
    max_cluster_k=10,                 # Max K for K-Means elbow search
    tsne_perplexity=30.0,
    bootstrap_iterations=1000,
    max_sample_for_advanced=5000,     # Subsample for expensive analyses
)

report = f2a.analyze("data.csv", config=config)

Config Options

API Reference

f2a.analyze(source, **kwargs) → AnalysisReport

AnalysisReport

Project Structure

f2a/
├── __init__.py              # Public API: analyze(), AnalysisConfig
├── _version.py
├── core/
│   ├── analyzer.py          # Main analysis orchestrator
│   ├── config.py            # AnalysisConfig dataclass
│   ├── loader.py            # 24+ format data loader
│   ├── preprocessor.py      # Data preprocessing pipeline
│   └── schema.py            # Schema inference
├── stats/                   # 20 analysis modules
│   ├── descriptive.py       # Mean, median, std, quartiles, etc.
│   ├── distribution.py      # Normality tests, skew/kurtosis
│   ├── correlation.py       # Pearson, Spearman, Kendall, VIF
│   ├── missing.py           # Missing data analysis
│   ├── outlier.py           # IQR / Z-score outlier detection
│   ├── categorical.py       # Frequency, entropy, chi-square
│   ├── feature_importance.py
│   ├── pca_analysis.py
│   ├── duplicates.py
│   ├── quality.py           # 4-dimension quality scoring
│   ├── advanced_distribution.py
│   ├── advanced_correlation.py
│   ├── advanced_anomaly.py  # Isolation Forest, LOF, Mahalanobis
│   ├── advanced_dimreduction.py  # t-SNE, UMAP, Factor Analysis
│   ├── clustering.py        # K-Means, DBSCAN, hierarchical
│   ├── feature_insights.py  # Interaction, leakage detection
│   ├── statistical_tests.py # Levene, KW, Mann-Whitney, ADF
│   ├── insight_engine.py    # Auto insight generation
│   ├── cross_analysis.py    # Cross-dimensional analysis
│   ├── column_role.py       # Column role inference
│   └── ml_readiness.py      # ML readiness scoring
├── viz/                     # 15 visualization modules
│   ├── plots.py             # Base plot utilities
│   ├── theme.py             # Consistent theming
│   ├── dist_plots.py
│   ├── corr_plots.py
│   ├── missing_plots.py
│   ├── outlier_plots.py
│   ├── categorical_plots.py
│   ├── pca_plots.py
│   ├── quality_plots.py
│   ├── cluster_plots.py
│   ├── advanced_dist_plots.py
│   ├── advanced_corr_plots.py
│   ├── advanced_anomaly_plots.py
│   ├── dimreduction_plots.py
│   ├── insight_plots.py
│   └── cross_plots.py
├── report/
│   ├── generator.py         # HTML report generator
│   └── i18n.py              # 6-language translations
└── utils/
    ├── exceptions.py
    ├── logging.py
    ├── type_inference.py
    └── validators.py

Internationalization (i18n)

The HTML report supports 6 languages with a language selector in the header:

Each language includes:

• ~120 metric tooltip translations — hover any table header
• ~50 section modal descriptions — click the ⓘ button on each section
• All UI labels, buttons, and messages

Requirements

• Python ≥ 3.10
• Core: pandas, numpy, matplotlib, seaborn, scipy, scikit-learn
• Formats: datasets (HuggingFace), openpyxl, pyarrow, pyreadstat, tables, odfpy, lxml, duckdb
• UI: rich, jinja2
• Optional: networkx, umap-learn, statsmodels (install with pip install f2a[advanced])

Development

# Clone and install
git clone https://github.com/CocoRoF/f2a.git
cd f2a
pip install -e ".[dev]"

# Run tests (88 tests)
pytest git_action/tests/ -q

# Lint
ruff check f2a/

License

MIT License — See LICENSE for details.