Machine learning workflow

Understanding Machine Learning

Lis Sulmont

Curriculum Manager, DataCamp

Machine learning workflow

Our scenario

Our dataset: NYC property sales from 2015-2019

Includes:

• Square feet
• Neighborhood
• Year built
• Sale price
• And more!

Our target: Sale price

Step 1: Extract features

Step 2: Split dataset

Step 3: Train model

Step 3: Train model

$$

Step 4: Evaluate

Step 4: Evaluate

• Test dataset: "unseen" data
• Many ways to evaluate:
  • What is the average error of the predictions?
  • What percent of apartments did the model accurately predict within a 10% margin?

Step 4: Evaluate

Step 4: Evaluate

Step 4: Evaluate

• If not, tune the model and re-train it:
  • e.g., change the model's options, add/remove features

Machine learning workflow

Summary of steps

1. Extract features
   • Choosing features and manipulating the dataset
2. Split dataset
   • Train and test dataset
3. Train model
   • Input train dataset into a machine learning model
4. Evaluate
   • If desired performance isn't reached: tune the model and repeat Step 3

Let's practice!

Understanding Machine Learning