How to see which features are used in the trained machine learning model?

Question

After training a model with autoML tool of H2O, I can see the variable importance with saved_model.varimp_plot(). I am curious about the feature engineering part whic H2O claims to do.

I'm trying simple lines of code sapmles in the documentation of H2O.

import h2o
h2o.init()

train_data = h2o.import_file("../full_data.csv")
test_data = h2o.import_file("../201810_pca.csv")

from h2o.automl import H2OAutoML
y = "Label"
x = ['feature0','feature1','feature2','feature3','feature4','feature5','feature6','feature7','feature8','feature9','feature10',
'feature11','feature12','feature13','feature14','feature15','feature16','feature17','feature18','feature19','feature20',
'feature21','feature22','feature23','Amount','DateTime']


aml = H2OAutoML(max_models = 100, max_runtime_secs=100000, seed = 1)
aml.train(x = x, y = y, training_frame = train_data)

lb = aml.leaderboard
lb.head()
lb.head(rows=lb.nrows) # Entire leaderboard

preds = aml.predict(test_data)
h2o.save_model(aml.leader, path = "./Saved_Models")


saved_model = h2o.load_model("./Saved_Models/XGBoost_2_AutoML_20191018_174201")

training_frame = your_model.actual_params['training_frame'] #The part gives error
print(training_frame)

How do I see which features are being used in the trained model? I'd like to see if H2O is extracting and adding new features or not.

I've used my_training_frame = your_model.actual_params['training_frame'] as stated in another question but it gives error: "TypeError: 'property' object has no attribute 'getitem'".

@edge can you please post the full code snippet? did you provide a validation frame when you called `.train()` by any chance. Please also note DAI and H2O-3 automl are separate products. AutoML does not do feature engineering for you, that is something in the DAI product. Thanks! — Lauren, Oct 24 '19 at 18:09
@Lauren I didn't know that, thanks for the information. Do you know any kind of dıcumentation about this topic that I can get more information about the feature engineering? — Ege, Oct 25 '19 at 08:56

score 0 · Accepted Answer · answered Oct 24 '19 at 19:27

Quick Note H2O.ai has a few products. The open source platform is called H2O-3 and it contains the AutoML algorithm. AutoML does not currently do feature engineering for you. If you want automatic feature engineering, you might be thinking of H2O's product Driverless-AI.

As for the error you are seeing, this is a bug and you can track the fix here.

Depending on what you pass to the .train() method, you may or may not hit this bug.

How to see which features are used in the trained machine learning model?

1 Answers1