Add rolling mean explainer and update README #109
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
jeancochrane
approved these changes
Feb 26, 2025
| We leverage the qualities above to produce a leave-one-out, time-weighted, rolling average sale price for | ||
| each building/sale. In layman's terms, we take the average of the sales in the same building from the prior 5 years, _excluding the current sale_. Here's what the rolling windows look like for sales in the training data, where the last row represents the actual assessment scenario (where the "sale" occurs on the lien date): | ||
|
|
||
|  |
There was a problem hiding this comment.
[Thought, non-blocking] One potential issue with this diagram is that it doesn't illustrate the time weighting. Not necessarily a problem as long as readers understand that the diagram is ignoring time weighting, but there's potential for misunderstanding.
There was a problem hiding this comment.
I think it's fine to exclude the time weights here, lest it become even more confusing.
Comment on lines
+70
to
+75
| Some additional technical notes on this feature: | ||
|
|
||
| - The time weights used to weight sales are _global_, rather than building-specific. They follow a simple logistic curve centered 3 years before the most recent sale i.e. sales close to the lien date are weighted most heavily. | ||
| - The feature is calculated using sales from the _entire range_ of the training data. This means that the test set version of the feature has seen training data sales from the same building, but not the sale being predicted. We contend that this is not data leakage, as it mirrors the real-world, production scenario where the models sees all sales in the building from the five years prior to the lien date. | ||
| - We contend that this feature is _not_ sales chasing (in the IAAO sense) because it excludes the _current_ sale from the average. This means that the model is not using the current sale to predict itself, but rather using the _prior_ sales to predict the current sale. | ||
| - The average excludes outlier sales and sales of non-livable units. |
There was a problem hiding this comment.
[Suggestion, non-blocking] Should we mention here that PIN10s don't always correspond exactly to buildings, although we use the term "building" for convenience?
There was a problem hiding this comment.
That might be too deep in the weeds even for this section.
Co-authored-by: Jean Cochrane <jeancochrane@users.noreply.github.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR tidies up the README and finalizes it for the 2025 model. It primarily adds a short section explaining the construction and purpose of the rolling mean feature.
Closes #105.