-
Notifications
You must be signed in to change notification settings - Fork 105
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
logit output of RandomForest #33
Comments
Unlike Gradient Boosted Trees learning algorithm, the Random Forest algorithm works with a "voting mechanism". In the case of classification, each tree casts a vote for one class (or multiple classes depending on the hyper-parameters). Therefore, the algorithm does not rely on any link functions / logits. This is why the argument I rarely saw logits being used with Random Forests. Out of curiosity, do you mind detailing your setup :) ? If a logit is what you need (i.e. the inverse of the logistic function), you can always compute it from the probabilities (be careful with numerical precision and proba=0 case). |
Thanks for the answer. This request is about the number of unique output values. I am trying to replace a library with the tfdf. As you can see below screen capture, predict, predict_proba & predict_log_proba give me different output values. I am talking about the number of unique output values. In terms of number of unique output values, computing logits from probabilities may be of no use. Because the number of unique output values will be the same after all |
Need to try more data-sets but Random Forest has a good enough number of unique output values on my current test data-sets. Thanks! |
Some quick remarks.
|
Differently from GBT, it seems that RandomForest does not have logit output
The logits are available in v0.1.7, but the signature is different from sklearn:
Trains a Gradient Boosted Trees that returns logits (assuming the dataset is a binary classification)
model = tfdf.keras.GradientBoostedTreesModel(apply_link_function=False)
The text was updated successfully, but these errors were encountered: