Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Improve time-based features and handling #164

Open
3 tasks
dfsnow opened this issue Jan 14, 2024 · 0 comments
Open
3 tasks

Improve time-based features and handling #164

dfsnow opened this issue Jan 14, 2024 · 0 comments
Labels
method ML technique or method change

Comments

@dfsnow
Copy link
Member

dfsnow commented Jan 14, 2024

While I think the model's overall time-trending is performing pretty well, it never hurts to test some tweaks/improvements. One thing we can do to check the model's performance/understanding of time is look at the errors for each fold from rolling-origin CV. To improve performance, we can:

  • Try adding more complex features (lagged predictors, lagged outcome, etc.). See some threads on this.
  • Test creating a separate index to use as a model feature
    • Calculate the index by area (Census tract, neighborhood, etc.)
    • Allows you to use a dedicated time series model for better/more performant forecasting
    • Helps get overall price trends correct in the main model
  • We can also revisit something we tried in the past: weighting more recent data using a decay function
@dfsnow dfsnow added the method ML technique or method change label Jan 14, 2024
@dfsnow dfsnow added this to the 2024 model changes milestone Jan 14, 2024
@dfsnow dfsnow removed this from the 2024 model changes milestone Apr 16, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
method ML technique or method change
Projects
None yet
Development

No branches or pull requests

1 participant