The micromlgen library in Python is designed to convert scikit-learn machine learning models into C code that can be used on microcontrollers like ESP32, ESP8266, and Arduino. It helps bridge the gap between machine learning development on powerful computers and deployment on embedded systems with constrained resources.

Installation:

pip install micromlgen

Supported Models

micromlgen works with scikit-learn models, including:

• Decision Trees (DecisionTreeClassifier, DecisionTreeRegressor)
• Random Forests (RandomForestClassifier, RandomForestRegressor)
• Naïve Bayes (GaussianNB)
• K-Nearest Neighbors (KNN) (KNeighborsClassifier)
• Support Vector Machines (SVM) (SVC, SVR)
• Gradient Boosting
• Extra Trees Classifier
• MLP (Neural Networks) – Small models only

It is best suited for lightweight models since embedded systems have limited memory and processing power.

Converting a Decision Tree to C Code:

Train a Model in Python

code:

from sklearn.tree import DecisionTreeClassifier

from micromlgen import port

import numpy as np

# Sample dataset (X: Features, y: Labels)

X = np.array([[0, 0], [1, 0], [0, 1], [1, 1]]) # Input features

y = np.array([0, 1, 1, 0]) # Output labels

# Train a simple Decision Tree model

model = DecisionTreeClassifier()

model.fit(X, y)

# Convert the trained model to C code

c_code = port(model)

# Print the generated C code

print(c_code)

 Generated C Code:

code:

#include <stdint.h>

// Decision tree prediction function

int predict(const float *features) {

  if (features[0] <= 0.5) {

    if (features[1] <= 0.5) {

      return 0;

    } else {

      return 1;

    }

  } else {

    if (features[1] <= 0.5) {

      return 1;

    } else {

      return 0;

    }

  }

}

Explanation of the Generated C Code

• The predict function takes a feature array (float *features).
• It follows a decision tree structure based on the trained model.
• The function returns a predicted class (0 or 1).

Deploying on ESP32 or Arduino

Step 1: Include the C Code in Your Microcontroller Sketch

• Copy the generated C code into an Arduino or ESP32 program.
• Call predict() with real-time sensor data.

code:

void setup() {

  Serial.begin(115200);

  float sampleData[2] = {0.5, 0.5}; // Example input features

  int prediction = predict(sampleData);

  Serial.print(“Predicted Class: “);

  Serial.println(prediction);

}

void loop() {

}

Why Use micromlgen?

• Converts Python ML models into C code for microcontrollers
• No external ML library required on the embedded device
• Works with lightweight models (decision trees, random forests, etc.)
• Optimized for microcontrollers like ESP32, ESP8266, Arduino

Limitations

• Only works with scikit-learn models (No TensorFlow or PyTorch models)
• Best for simple models (Large models will not fit in microcontroller memory)
• Does not support deep learning (TinyML tools like TensorFlow Lite are better for this)

Alternatives to micromlgen

• TensorFlow Lite for Microcontrollers (TFLM) → For Deep Learning on ESP32/Arduino
• Edge Impulse → Cloud-based TinyML model deployment
• Arduino Nano 33 BLE Sense ML Kit → Built-in ML capabilities