vlegoy / rcatt Goto Github PK

Where did you get your dataset to train the model?

Just labeled data from
https://github.com/mitre/cti ?
I see there are 1700 records in total, which is not enough to train a machine learning model I suppose.

Or have you ever thought about collecting threat reports from other sites and human label them?

When I try to run rcATT_gui.py, it runs fine. When I access the website, put some dummy text inside and press "Predict", the whole thing crashes and displays "ModuleNotFoundError: No module named 'sklearn.feature_selection.univariate_selection'", as seen in the screenshots below.

Is there anything that can be done?

I noticed that when running the train command, the "added data" CSV file classification_tools/data/training_data_added.csv is not actually appended to the "original" training dataset:

Note that the Pandas DataFrame.append function returns a new DataFrame and does not modify its input, so the corrected version would be:

	train_data_df = train_data_df.append(train_data_added, ignore_index = True)

I'd be happy to submit this as a merge request, if that'd work.

This is a basic requirements.txt just in case you want to add it for an easier installation via pip.

colorama == 0.4.3
flask == 1.1.2
joblib == 0.14.1
nltk == 3.4.5
numpy == 1.17.4
pandas == 0.25.3
sklearn <= 0.22
stix2 == 1.2.1

https://github.com/vlegoy/rcATT/blob/f82f7fd456279abefcd3e0b50e8056345c11aeb7/classification_tools/postprocessing.py#L92C7-L92C47

this should be:
predprob_techniques_corrected[i][j] = 0

k makes no sense. See confidence_propagation-function.

I also changed the effectiveness of the function with numpy arrays. See my pending pull request.