MEX — Machine Learning EXpression Language

A beginner-friendly programming language that teaches machine learning concepts through interactive code. Write clean, readable code that converts to real TensorFlow.js or Python.

No dependencies required — includes a minimal neural network built from scratch.

Quick Start (2 minutes)

Install

npm install -g mex-ml

Create Your First Project

mex init my-first-ml
cd my-first-ml
mex run hello.mx

That's It!

You now have a working ML project with:

• hello.mx — your first program
• examples/ — example files
• datasets/ — your data files
• README.md — documentation

Features

• Simple Syntax — reads like English
• Zero Dependencies — runs anywhere Node.js runs
• 100+ Built-in Functions — math, statistics, arrays, strings, data
• DataFrame API — Pandas-like data manipulation with method chaining
• Visualization — ASCII charts (bar, line, scatter, histogram, pie)
• Dual Export — generate TensorFlow.js or Python code
• Smart Debug — catches NaN, exploding loss, overfitting
• Interactive Playground — web-based editor with syntax highlighting
• REPL Mode — interactive command line for experimenting
• Type System — optional type annotations with runtime checking
• Package Manager — install and share MEX packages
• Learning System — 6 lessons, challenges, achievements

Commands

Getting Started

mex init <project>     # Create new project
mex new <name>         # Create new .mx file
mex run <file.mx>      # Run a .mx file
mex repl               # Start interactive REPL
mex --help             # Show all commands
mex --version          # Show version

Learning

mex lesson 1           # Start lesson 1
mex challenge 1        # Try challenge 1
mex lessons            # List all lessons
mex status             # Check your progress
mex achievements       # View achievements

Code Generation

mex --generate file.mx # Generate TensorFlow.js
mex --python file.mx   # Generate Python
mex --show-tf file.mx  # Show TF.js alongside
mex --stats file.mx    # Show compression stats

Example: Linear Regression

Write MEX Code

data points
  (1, 2)
  (2, 4)
  (3, 6)
  (4, 8)

model simple
train 100 epochs
predict 5

Run It

mex run hello.mx

Output

Data loaded: 4 points
Model created: simple (linear)
Training for 100 epochs...
Training complete! Error: 0.001234
Prediction for 5: 9.9876

Convert to Python

mex --python hello.mx

# Generated by MEX → Python/TensorFlow
import tensorflow as tf
import numpy as np

xs = np.array([1, 2, 3, 4]).reshape(-1, 1)
ys = np.array([2, 4, 6, 8]).reshape(-1, 1)

model = tf.keras.Sequential([
    tf.keras.layers.Dense(1, input_shape=[1])
])

model.compile(optimizer='sgd', loss='mse')
model.fit(xs, ys, epochs=100, verbose=0)

pred = model.predict(np.array([[5]]))
print(f"Prediction: {pred.numpy()[0][0]:.4f}")

Language Syntax

Data

data points (1, 2) (3, 4)       # Inline data
data csv "houses.csv"            # CSV file

Models

model simple                     # Linear regression
model sequential                 # Neural network
  dense 8 relu                   # Hidden layer
  dropout 0.2                    # Regularization
  dense 1 sigmoid                # Output layer

Training

train 100 epochs                 # Basic training
train 200 epochs learning_rate 0.01  # Custom learning rate

Prediction

predict 5                        # Single value
predict [5, 10, 15]              # Multiple values

Control Flow

if (x > 5) { ... } else { ... }
for i in range(10) { ... }
while (x > 0) { ... }

Functions

fn add(a, b) {
  return a + b
}

let doubled = map([1, 2, 3], fn(x) { return x * 2 })

Type Annotations (Optional)

let x: number = 42
fn add(a: number, b: number) -> number {
  return a + b
}

DataFrame Chaining

data csv "houses.csv"

let result = data
  .filter("size", ">", 2000)
  .sort("price", false)
  .head(3)

print(result.to_json())

Built-in Functions (100+)

Math

sqrt, pow, abs, sin, cos, tan, log, exp, floor, ceil, round, random

Statistics

mean, sum, std, variance, median, mode, min_arr, max_arr

String

len, upper, lower, trim, split, join, replace, includes, type

Array

sort, sort_by, reverse, unique, flatten, slice, head, tail, zip

Higher-Order

map, filter, reduce, each, find, find_index, every, some, group_by

Data

column, columns, select, where, limit, offset, sample, shuffle, distinct

Visualization

bar_chart, line_chart, scatter_plot, histogram, pie_chart

Learning Path

| Lesson | Topic | Duration | |--------|-------|----------| | 1 | What is Data? | 20 min | | 2 | Finding Patterns | 25 min | | 3 | Classification | 25 min | | 4 | Neural Networks | 30 min | | 5 | Real Data | 30 min | | 6 | Advanced Topics | 35 min |

Interactive Playground

Open playground/index.html in your browser:

• Syntax highlighting
• Autocomplete (50+ functions)
• 12 built-in examples
• File tabs with auto-save
• CSV import/export

Documentation

| Document | Description | |----------|-------------| | DOCS.md | Complete language reference | | CHANGES.md | Changelog | | CONTRIBUTING.md | How to contribute |

System Requirements

• Node.js: v14 or higher
• OS: Windows, macOS, Linux

License

MIT License — see LICENSE for details.

Support

• Issues: GitHub Issues
• npm: npmjs.com/package/mex-ml