Random Forest model in ATLAS Prediction not working

Implementers
,
1

Hi,

I am trying to implement the Patient Prediction package after downloading from Atlas with a Random Forest model. Unfortunately, it is not generating any results and is also not giving any explicit errors - the model does not train once executed although I can see the model in the settings file. I tried 2 other models (Logistic Regression and Gradient Boosting Machine) with the same target and outcome cohorts, covariate settings and population settings and they worked.

Varying the model settings did not help either. Here are the model settings (from the Export tab) -

“RandomForestSettings”: {
“seed”: null,
“mtries”: [
-1,5,20
],
“ntrees”: [
500,100,50,1000
],
“maxDepth”: [
4,10,17,25,50
],
“varImp”: [
true
]}}],
“runPlpArgs”: {
“minCovariateFraction”: 0.001,
“normalizeData”: true,
“testSplit”: “person”,
“testFraction”: 0.25,
“splitSeed”: 42,
“nfold”: 5
}

I’ve also attached the log files generated from the Random Forest model run. Will be grateful for any insight on how to train the model and why it isn’t working. Thanks! plplog.pdf (9.3 KB)

2

Hi,
To my knowledge, If you already succeed model development with lasso and gbm, but just couldn’t using random forest, you can check your python environment. (lasso and gbm working based on r package, but random forest use python code)
you can find the package install guide in here

3

Thank you! I followed the instructions and the PLP Installation Check ran successfully (no errors). However, re-running the prediction package still did not generate results for Random Forest. I think you are correct that it’s an issue with the Python environment as plplogs generated from the PLP Installation check for random forest, neural network, AdaBoost, Naive Bayes do not show the model training.

4

It was a Python environment issue - I used the updated installation instructions here and it’s working now. Thanks again for your help!

1 Like