Skip to content
This repository has been archived by the owner on May 8, 2023. It is now read-only.
/ xlearn-ruby Public archive

High performance factorization machines for Ruby

License

Notifications You must be signed in to change notification settings

ankane/xlearn-ruby

Repository files navigation

xLearn Ruby

xLearn - the high performance machine learning library - for Ruby

Supports:

  • Linear models
  • Factorization machines
  • Field-aware factorization machines

Build Status

Installation

Add this line to your application’s Gemfile:

gem "xlearn"

Getting Started

Prep your data

x = [[1, 2], [3, 4], [5, 6], [7, 8]]
y = [1, 2, 3, 4]

Train a model

model = XLearn::Linear.new(task: "reg")
model.fit(x, y)

Use XLearn::FM for factorization machines and XLearn::FFM for field-aware factorization machines

Make predictions

model.predict(x)

Save the model to a file

model.save_model("model.bin")

Load the model from a file

model.load_model("model.bin")

Save a text version of the model

model.save_txt("model.txt")

Pass a validation set

model.fit(x_train, y_train, eval_set: [x_val, y_val])

Train online

model.partial_fit(x_train, y_train)

Get the bias term, linear term, and latent factors

model.bias_term
model.linear_term
model.latent_factors # fm and ffm only

Parameters

Pass parameters - default values below

XLearn::FM.new(
  task: "binary",      # binary (classification), reg (regression)
  metric: nil,         # acc, prec, recall, f1, auc, mae, mape, rmse, rmsd
  lr: 0.2,             # learning rate
  lambda: 0.00002,     # lambda for l2 regularization
  k: 4,                # latent factors for fm and ffm
  alpha: 0.3,          # hyper parameter for ftrl
  beta: 1.0,           # hyper parameter for ftrl
  lambda_1: 0.00001,   # hyper parameter for ftrl
  lambda_2: 0.00002,   # hyper parameter for ftrl
  epoch: 10,           # number of epochs
  fold: 3,             # number of folds
  opt: "adagrad",      # sgd, adagrad, ftrl
  block_size: 500,     # block size for on-disk training in MB
  early_stop: true,    # use early stopping
  stop_window: 2,      # size of stop window for early stopping
  sign: false,         # convert predition output to 0 and 1
  sigmoid: false,      # convert predition output using sigmoid
  seed: 1              # random seed to shuffle data set
)

Cross-Validation

Cross-validation

model.cv(x, y)

Specify the number of folds

model.cv(x, y, folds: 5)

Data

Data can be an array of arrays

[[1, 2, 3], [4, 5, 6]]

Or a Numo array

Numo::NArray.cast([[1, 2, 3], [4, 5, 6]])

Or a Rover data frame

Rover.read_csv("houses.csv")

Or a Daru data frame

Daru::DataFrame.from_csv("houses.csv")

Performance

For large datasets, read data directly from files

model.fit("train.txt", eval_set: "validate.txt")
model.predict("test.txt")
model.cv("train.txt")

For linear models and factorization machines, use CSV:

label,value_1,value_2,...,value_n

Or the libsvm format (better for sparse data):

label index_1:value_1 index_2:value_2 ... index_n:value_n

You can also use commas instead of spaces for separators

For field-aware factorization machines, use the libffm format:

label field_1:index_1:value_1 field_2:index_2:value_2 ...

You can also use commas instead of spaces for separators

You can also write predictions directly to a file

model.predict("test.txt", out_path: "predictions.txt")

Credits

This library is modeled after xLearn’s Scikit-learn API.

History

View the changelog

Contributing

Everyone is encouraged to help improve this project. Here are a few ways you can help:

To get started with development:

git clone https://github.com/ankane/xlearn-ruby.git
cd xlearn-ruby
bundle install
bundle exec rake vendor:all
bundle exec rake test

About

High performance factorization machines for Ruby

Resources

License

Stars

Watchers

Forks

Packages

No packages published