Eps

Machine learning for Ruby

• Build predictive models quickly and easily
• Serve models built in Ruby, Python, R, and more

Check out this post for more info on machine learning with Rails

Add this line to your application’s Gemfile:

gem "eps"

On Mac, also install OpenMP:

brew install libomp

Create a model

data = [
  {bedrooms: 1, bathrooms: 1, price: 100000},
  {bedrooms: 2, bathrooms: 1, price: 125000},
  {bedrooms: 2, bathrooms: 2, price: 135000},
  {bedrooms: 3, bathrooms: 2, price: 162000}
]
model = Eps::Model.new(data, target: :price)
puts model.summary

Make a prediction

model.predict(bedrooms: 2, bathrooms: 1)

Store the model

File.write("model.pmml", model.to_pmml)

Load the model

pmml = File.read("model.pmml")
model = Eps::Model.load_pmml(pmml)

A few notes:

• The target can be numeric (regression) or categorical (classification)
• Pass an array of hashes to predict to make multiple predictions at once
• Models are stored in PMML, a standard for model storage

Often, the goal of building a model is to make good predictions on future data. To help achieve this, Eps splits the data into training and validation sets if you have 30+ data points. It uses the training set to build the model and the validation set to evaluate the performance.

If your data has a time associated with it, it’s highly recommended to use that field for the split.

Eps::Model.new(data, target: :price, split: :listed_at)

Otherwise, the split is random. There are a number of other options as well.

Performance is reported in the summary.

• For regression, it reports validation RMSE (root mean squared error) - lower is better
• For classification, it reports validation accuracy - higher is better

Typically, the best way to improve performance is feature engineering.

Features are extremely important for model performance. Features can be:

1. numeric
2. categorical
3. text

For numeric features, use any numeric type.

{bedrooms: 4, bathrooms: 2.5}

For categorical features, use strings or booleans.

{state: "CA", basement: true}

Convert any ids to strings so they’re treated as categorical features.

{city_id: city_id.to_s}

For dates, create features like day of week and month.

{weekday: sold_on.strftime("%a"), month: sold_on.strftime("%b")}

For times, create features like day of week and hour of day.

{weekday: listed_at.strftime("%a"), hour: listed_at.hour.to_s}

For text features, use strings with multiple words.

{description: "a beautiful house on top of a hill"}

This creates features based on word count.

You can specify text features explicitly with:

Eps::Model.new(data, target: :price, text_features: [:description])

You can set advanced options with:

text_features: {
  description: {
    min_occurences: 5,          # min times a word must appear to be included in the model
    max_features: 1000,         # max number of words to include in the model
    min_length: 1,              # min length of words to be included
    case_sensitive: true,       # how to treat words with different case
    tokenizer: /\s+/,           # how to tokenize the text, defaults to whitespace
    stop_words: ["and", "the"]  # words to exclude from the model
  }
}

We recommend putting all the model code in a single file. This makes it easy to rebuild the model as needed.

In Rails, we recommend creating a app/ml_models directory. Be sure to restart Spring after creating the directory so files are autoloaded.

bin/spring stop

Here’s what a complete model in app/ml_models/price_model.rb may look like:

class PriceModel < Eps::Base
  def build
    houses = House.all

    # train
    data = houses.map { |v| features(v) }
    model = Eps::Model.new(data, target: :price, split: :listed_at)
    puts model.summary

    # save to file
    File.write(model_file, model.to_pmml)

    # ensure reloads from file
    @model = nil
  end

  def predict(house)
    model.predict(features(house))
  end

  private

  def features(house)
    {
      bedrooms: house.bedrooms,
      city_id: house.city_id.to_s,
      month: house.listed_at.strftime("%b"),
      listed_at: house.listed_at,
      price: house.price
    }
  end

  def model
    @model ||= Eps::Model.load_pmml(File.read(model_file))
  end

  def model_file
    File.join(__dir__, "price_model.pmml")
  end
end

Build the model with:

PriceModel.build

This saves the model to price_model.pmml. Check this into source control or use a tool like Trove to store it.

Predict with:

PriceModel.predict(house)

We recommend monitoring how well your models perform over time. To do this, save your predictions to the database. Then, compare them with:

actual = houses.map(&:price)
predicted = houses.map(&:predicted_price)
Eps.metrics(actual, predicted)

For RMSE and MAE, alert if they rise above a certain threshold. For ME, alert if it moves too far away from 0. For accuracy, alert if it drops below a certain threshold.

Eps makes it easy to serve models from other languages. You can build models in Python, R, and others and serve them in Ruby without having to worry about how to deploy or run another language.

Eps can serve LightGBM, linear regression, and naive Bayes models. Check out ONNX Runtime and Scoruby to serve other models.

To create a model in Python, install the sklearn2pmml package

pip install sklearn2pmml

And check out the examples:

• LightGBM Regression
• LightGBM Classification
• Linear Regression
• Naive Bayes

To create a model in R, install the pmml package

install.packages("pmml")

And check out the examples:

• Linear Regression
• Naive Bayes

It’s important for features to be implemented consistently when serving models created in other languages. We highly recommend verifying this programmatically. Create a CSV file with ids and predictions from the original model.

Once the model is implemented in Ruby, confirm the predictions match.

model = Eps::Model.load_pmml("model.pmml")

# preload houses to prevent n+1
houses = House.all.index_by(&:id)

CSV.foreach("predictions.csv", headers: true, converters: :numeric) do |row|
  house = houses[row["house_id"]]
  expected = row["prediction"]

  actual = model.predict(bedrooms: house.bedrooms, bathrooms: house.bathrooms)

  success = actual.is_a?(String) ? actual == expected : (actual - expected).abs < 0.001
  raise "Bad prediction for house #{house.id} (exp: #{expected}, act: #{actual})" unless success

  putc "✓"
end

A number of data formats are supported. You can pass the target variable separately.

x = [{x: 1}, {x: 2}, {x: 3}]
y = [1, 2, 3]
Eps::Model.new(x, y)

Data can be an array of arrays

x = [[1, 2], [2, 0], [3, 1]]
y = [1, 2, 3]
Eps::Model.new(x, y)

Or Numo arrays

x = Numo::NArray.cast([[1, 2], [2, 0], [3, 1]])
y = Numo::NArray.cast([1, 2, 3])
Eps::Model.new(x, y)

Or a Rover data frame

df = Rover.read_csv("houses.csv")
Eps::Model.new(df, target: "price")

Or a Daru data frame

df = Daru::DataFrame.from_csv("houses.csv")
Eps::Model.new(df, target: "price")

When reading CSV files directly, be sure to convert numeric fields. The table method does this automatically.

CSV.table("data.csv").map { |row| row.to_h }

Pass an algorithm with:

Eps::Model.new(data, algorithm: :linear_regression)

Eps supports:

• LightGBM (default)
• Linear Regression
• Naive Bayes

Pass the learning rate with:

Eps::Model.new(data, learning_rate: 0.01)

By default, an intercept is included. Disable this with:

Eps::Model.new(data, intercept: false)

To speed up training on large datasets with linear regression, install GSL. With Homebrew, you can use:

brew install gsl

Then, add this line to your application’s Gemfile:

gem "gslr", group: :development

It only needs to be available in environments used to build the model.

To get the probability of each category for predictions with classification, use:

model.predict_probability(data)

Naive Bayes is known to produce poor probability estimates, so stick with LightGBM if you need this.

Pass your own validation set with:

Eps::Model.new(data, validation_set: validation_set)

Split on a specific value

Eps::Model.new(data, split: {column: :listed_at, value: Date.parse("2025-01-01")})

Specify the validation set size (the default is 0.25, which is 25%)

Eps::Model.new(data, split: {validation_size: 0.2})

Disable the validation set completely with:

Eps::Model.new(data, split: false)

The database is another place you can store models. It’s good if you retrain models automatically.

We recommend adding monitoring and guardrails as well if you retrain automatically

Create an Active Record model to store the predictive model.

rails generate model Model key:string:uniq data:text

Store the model with:

store = Model.where(key: "price").first_or_initialize
store.update(data: model.to_pmml)

Load the model with:

data = Model.find_by!(key: "price").data
model = Eps::Model.load_pmml(data)

You can use IRuby to run Eps in Jupyter notebooks. Here’s how to get IRuby working with Rails.

Specify a weight for each data point

Eps::Model.new(data, weight: :weight)

You can also pass an array

Eps::Model.new(data, weight: [1, 2, 3])

Weights are supported for metrics as well

Eps.metrics(actual, predicted, weight: weight)

Reweighing is one method to mitigate bias in training data

View the changelog

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

• Report bugs
• Fix bugs and submit pull requests
• Write, clarify, or fix documentation
• Suggest or add new features

To get started with development:

git clone https://github.com/ankane/eps.git
cd eps
bundle install
bundle exec rake test