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
Installation
Add this line to your application’s Gemfile:
gem "eps"
On Mac, also install OpenMP:
brew install libomp
Getting Started
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
Building Models
Goal
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.
Feature Engineering
Features are extremely important for model performance. Features can be:
- numeric
- categorical
- text
Numeric
For numeric features, use any numeric type.
{bedrooms: 4, bathrooms: 2.5}
Categorical
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}
Text
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,
max_features: 1000,
min_length: 1,
case_sensitive: true,
tokenizer: /\s+/, # how to tokenize the text, defaults to whitespace
stop_words: ["and", "the"] # words to exclude from the model
}
}
Full Example
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
data = houses.map { |v| features(v) }
model = Eps::Model.new(data, target: :price, split: :listed_at)
puts model.summary
File.write(model_file, model.to_pmml)
@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)
Monitoring
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.
Other Languages
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.
Python
To create a model in Python, install the sklearn2pmml package
pip install sklearn2pmml
And check out the examples:
R
To create a model in R, install the pmml package
install.packages("pmml")
And check out the examples:
Verifying
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.
house_id | prediction |
---|
1 | 145000 |
2 | 123000 |
3 | 250000 |
Once the model is implemented in Ruby, confirm the predictions match.
model = Eps::Model.load_pmml("model.pmml")
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
Data
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 }
Algorithms
Pass an algorithm with:
Eps::Model.new(data, algorithm: :linear_regression)
Eps supports:
- LightGBM (default)
- Linear Regression
- Naive Bayes
LightGBM
Pass the learning rate with:
Eps::Model.new(data, learning_rate: 0.01)
Linear Regression
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.
Probability
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.
Validation Options
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("2019-01-01")})
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)
Database Storage
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 ActiveRecord 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)
Jupyter & IRuby
You can use IRuby to run Eps in Jupyter notebooks. Here’s how to get IRuby working with Rails.
Weights
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
Upgrading
0.3.0
Eps 0.3.0 brings a number of improvements, including support for LightGBM and cross-validation. There are a number of breaking changes to be aware of:
-
LightGBM is now the default for new models. On Mac, run:
brew install libomp
Pass the algorithm
option to use linear regression or naive Bayes.
Eps::Model.new(data, algorithm: :linear_regression)
-
Cross-validation happens automatically by default. You no longer need to create training and test sets manually. If you were splitting on a time, use:
Eps::Model.new(data, split: {column: :listed_at, value: Date.parse("2019-01-01")})
Or randomly, use:
Eps::Model.new(data, split: {validation_size: 0.3})
To continue splitting manually, use:
Eps::Model.new(data, validation_set: test_set)
-
It’s no longer possible to load models in JSON or PFA formats. Retrain models and save them as PMML.
0.2.0
Eps 0.2.0 brings a number of improvements, including support for classification.
We recommend:
- Changing
Eps::Regressor
to Eps::Model
- Converting models from JSON to PMML
model = Eps::Model.load_json("model.json")
File.write("model.pmml", model.to_pmml)
- Renaming
app/stats_models
to app/ml_models
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/eps.git
cd eps
bundle install
bundle exec rake test