nickleomartin / ml_model_lib

Assortment of TensorFlow and Numpy models

ml_model_lib

Assortment of TensorFlow and Numpy models. Loosely follows Scikit-Learn API. For personal learning, not production. Will refactor as I go. Starting with more traditional machine learning models.

Factorization Machines:

TensorFlow implementation of the original paper by Steffen Rendel.

from factorization_machines_tf import FactorizationMachines
from utils import generate_rendle_style_dataset

## Obtain data
x_data, y_data = generate_rendle_style_dataset()

## Fit and predict
fm = FactorizationMachines(l_factors=10)
fm.fit(x_data, y_data)
fm.predict(x_data)

Logistic Regression:

A multi-class, Numpy implementation with l2-norm regularization:

from logistic_regression_np import LogisticRegression
from utils import generate_classification_style_dataset

## Obtain data
X, Y = generate_classification_style_dataset()

## Fit and predict
lr = LogisticRegression()
lr.fit(X, Y)
lr.predict(X)

Support Vector Machines:

A binary class, Tensorflow implementation with l2-norm regularization:

from support_vector_machines_tf import SupportVectorMachines
from utils import generate_classification_style_dataset

## Obtain data
X, Y = generate_classification_style_dataset('binary')

## Fit and predict
lr = SupportVectorMachines()
lr.fit(X, Y)
lr.predict(X)

To-Do:

Models no particular order:

• Word2Vec/Item2Vec
• Word Mover's Distance
• Sequence2Sequence models
• Singular Value Decomposition & Latent Semantic Indexing
• Bayesian Linear Regression
• K-means
• T-SNE
• Conditional Random Fields
• StarSpace
• Metric Learning models
• Decision Tree
• Random Forests

Extras:

• Standard datasets: classification, regression, ranking etc.
• Visualisations of the learned decision functions