Getting Started

Overview

Beginner-first machine learning. Train a model in 3 lines.

fling wraps scikit-learn behind a clean, safe API designed for people learning ML. It handles data splitting, preprocessing, missing values, and encoding automatically - and explains every decision in plain English.

● python

from fling import Classifier

model = Classifier(data="titanic.csv", target="Survived")
model.train()
model.evaluate()

● output

[ok] Loaded 891 rows, 12 columns
[ok] Split: 712 training rows, 179 test rows (split before preprocessing)
[ok] Auto-handled: 5 text column(s) encoded (Pclass, Name, Sex, Ticket, Embarked)
[ok] Auto-handled: 2 column(s) had missing values - filled with median/mode (Age, Embarked)
[ok] Trained: LogisticRegression (small dataset - logistic regression is fast and interpretable)

Results:
  Accuracy:  81.6%
  In plain English: Your model correctly predicted 'Survived' for 146 out of 179 rows in the test set.

Install

bash

pip install git+https://github.com/vedanshalways/Fling.git

Quick reference

Task	Code
Classify (predict a category)	`Classifier(data=..., target=...)`
Regress (predict a number)	`Regressor(data=..., target=...)`
Train	`.train()`
Evaluate on test set	`.evaluate()`
Predict new rows	`.predict(new_df)`
Visualize results	`.visualize()`
Explain feature importance	`.explain()`
Compare 5 models	`.compare()`
Get algorithm recommendations	`.recommend()`
See the raw sklearn code	`.show_sklearn_code()`
Escape to sklearn	`.to_sklearn()` / `.get_splits()`

What fling handles automatically

Problem	What fling does
Text / categorical columns	OneHotEncodes them
Missing numeric values	Fills with column median
Missing categorical values	Fills with most frequent value
Columns with > 50% missing	Drops them, logs in report
ID-like columns (all unique)	Drops them automatically
Choosing a model	Picks based on dataset size
Train/test split ordering	Always splits before preprocessing

How fling prevents data leakage

Data leakage is the #1 silent bug in beginner ML. It happens when test data influences the training process - making accuracy look better than it really is.

The wrong way (what most tutorials show)

python

# WRONG: scaler sees all data before the split
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)            # computes mean/std on ALL data
X_train, X_test = train_test_split(X_scaled)  # test is already contaminated

⚠The scaler's mean and std were computed using test rows. Your accuracy score is optimistic - it won't hold up on truly new data.

The right way (what fling always does)

python

# RIGHT: split first, then fit preprocessing on train only
X_train, X_test, y_train, y_test = train_test_split(X, y)

pipeline = Pipeline([
    ("preprocessor", column_transformer),  # fitted on X_train only
    ("model", estimator),
])
pipeline.fit(X_train, y_train)   # scaler learns from train rows only
pipeline.predict(X_test)         # transforms test using train stats

You can verify this yourself:

python

model.train()
X_train, X_test, y_train, y_test = model.get_splits()
print(X_train.shape[0])  # e.g. 712 - scaler computed on these rows only

Requirements

Python 3.8+
scikit-learn ≥ 1.0
pandas ≥ 1.3
numpy ≥ 1.21
matplotlib ≥ 3.4 (optional - only needed for .visualize())

Examples

The examples/ folder has four runnable scripts:

File	What it shows
`01_your_first_classifier.py`	Basic 3-line classifier
`02_your_first_regressor.py`	Basic 3-line regressor
`03_compare_models.py`	recommend → train → compare → explain → visualize
`04_graduate_to_sklearn.py`	show_sklearn_code → to_sklearn → get_splits

bash

python 01_your_first_classifier.py

Next steps

Classifier reference - full parameter docs, all algorithm strings, auto-selection logic
.train() / .evaluate() - output format, R² interpretation, classification report
.visualize() - 4-panel figures for classifiers and regressors
Data leakage - why fling splits before preprocessing and how to verify it
Extended API - clustering, text, neural networks, ensembles, and more
GitHub ↗ - source code, issues, contributing