cl-ml

Machine Learning library for Common Lisp

Martin Kersner, [email protected]

cl-ml currently supports following machine learning algorithms:

• k-Nearest Neighbors
• Linear Regression
• Logistic Regression
• Decision Trees (ID3)
• Support Vector Machines
• Multinomial Naive Bayes Classifier
• Artificial Neural Networks

Requirements

cl-ml requires asdf to be installed. cl-math and cl-plot are included as submodules to cl-ml project.

Get the latest version

git clone --recursive https://github.com/martinkersner/cl-ml.git

Start using cl-ml

(load "init")

Algorithms

Naive Bayes Classifier

Currently, only Multinomial Naive Bayes Classifier is supported.

1. Create instance of Naive Bayes Classifier

(defparameter *nbc*
  (make-instance 'naive-bayes-classifier))

2. Pass training data. Training data will be transformed to feature vectors within training procedure.

; X-train represent training data in list of lists format
; y-train represent training labels in cl-math matrix format
(fit *nbc* X-train y-train)

3. Predict on new data

; X-test represent training data in list of lists format
(predict *nbc* X-test)

Artificial Neural Networks

Currently, only fully connected layers with sigmoid activation function are supported. Optimization is performed using Stochastic Gradient Descent (SGD).

1. Create network with 2 neurons in input layer, 3 neurons in hidden layer and 1 neuron the output layer.

(defparameter *nn*
  (make-instance 'neural-network :nn-dims '(2 3 1)))

2. Define training parameters for SGD.

(defparameter *params*
  (generate-params '((num-epoch       50)
                     (lr              0.3)
                     (mini-batch-size 20))))

3. Start training.

; X-train represent training data in cl-math matrix format
; y-train represent training labels in cl-math matrix format
(fit *nn* X-train y-train *params*)

4. Predict on new data.

; X-test represent testing data in cl-math matrix format
(predict *nn* X-test)

Project Roadmap

1. Add new algorithms
   • k-Means
   • Adaboost
   • Autoencoders
   • CNN
2. Add new default data sets
3. Methods for data manipulation
   • Generating new data
4. Methods enabling easier training
   • Cross validation
   • Grid search