README.md

Up-to-date: 2023-12-11.

List a set of machine learning methods, pick the best one, train and test. The usual.

Usage

python3 ml/main.py --help

# Example call for 2023-11-22, all modes (everything)
python3 -m sklearnex main.py \
    -i "../data/csv/2023-11-22/all.csv" \
    -o "../data/results/2023-11-22/" \
    -oh "../data/hyperparameters/2023-11-22/everything-42.json" \
    -md "../data/models/2023-11-22/" \
    -ec "../pcap_manipulation/configs/extract-v4.0.json" \
    -rc "../pcap_manipulation/configs/resolvers.json" \
    -dc "../scripts/configs/devices-v4.0.json" \
    -r "42" \
    -m "everything"

# Example call to load 2023-10-19's model and use 2023-10-20's data 
python3 -m sklearnex main.py \
    -i "../data/csv/2023-10-20/all.csv" \
    -o "../data/results/2023-10-19/" \
    -md "../data/models/2023-11-19/" \
    -ec "../pcap_manipulation/configs/extract-v3.0.json" \
    -rc "../pcap_manipulation/configs/resolvers.json" \
    -dc "../scripts/configs/devices-v3.0.json" \
    -r "12" \
    -lm "../data/models/2023-10-19/RandomForestClassifier-doh_Quad9-all_both-12.pipeline" \
    -prid "2023-10-19" &

See the get_modes() function to check the list of all possible modes. Use -m "everything" to start everything, or use a subset like -m "length_only,by_time".

Using

From sklearn:

• Classifiers
  • RandomForestClassifier
  • KNeighborsClassifier
  • ComplementNB
  • LogisticRegression
  • LinearSVC
  • SVC One-Vs-One
  • LinearSVC One-Vs-The-Rest
• Hyperparameters search functions
  • HalvingRandomSearchCV
  • RandomizedSearchCV (because the other does not yet work on neural networks outputs)

From imblearn:

• Pipeline (to correctly sample only the training set)

From keras:

• A simple Sequential model from previous works

From SciKeras:

• KerasClassifier to use it as a wrapper for keras, in order to integrate it in a Pipeline