library(tidymodels)
# Create Recipe
model_recipe <- recipes::recipe(what-to-predict ~ data-features,
data = your-data)
# Define model using parsnip (e.x. linear regression)
model_type <- parsnip::linear_reg()
# Combine the two to create a workflow
model_wflow <- workflows::workflow(model_recipe, model_type)
# Fit workflow to data using parsnip
model_fit <- parsnip::fit(model_wflow, your-data)Tidymodels and Vetiver
Review
What’s the Goal?
What we have:
-
A beautiful - tidy - validated dataset
Historical food inspection results
Business information
What do we want:
- A MODEL that can predict if an establishment will fail an inspection
Where are we going
Review - Chicago Food Inspections Project
What is this project?
- Over 15,000 food establishments across the City of Chicago
- Only a handful of food inspectors currently employed
Introduction to Tidymodels
✋ If you have used tidymodels before?
Introduction to Tidymodels
Steps to create (most) models:
- Data - preprocessing?
- Create a formula - what does the model do?
- Define your model type - Linear regression, decision tree, xgboost, etc?
- Train
- Test
- Predict
Introduction to Tidymodels
Introduction to Tidymodels
Activity Time!
Important
We will be using the diamonds dataset for this activity!
We have a model…now what 🤷 ?
We have two questions:
- How can we save our model so others can use it?
- How can we serve our model so others can interact with it?
Vetiver
- Open source R (and Python) package.
- Deploy and maintain machine learning models –> MLOps
- Provides tools to version, deploy, and monitor a trained model.
Introduction to Pins & Plumber
🧰 Pins
- Open source R (and Python) package.
- Publishes data, models, and other R objects, making it easy to share them across projects and with your colleagues.
- Interoperability –> You can share an object built in R with Python users (and vice versa)!
- For our workflow:
- How can we save our model so that others (or other content) can use it?
🧰 Plumber
- Open source R package.
- Create a web Application Programming Interface (API) using only R code!
- Interoperability –> You can interact with a Plumber API using Python (or any other language/HTTP client).
- For our workflow:
- How can we serve our model as an API?
Example Prediction
Review