config.yaml

data_dir: data/ #/home/guru/ENViSEC/data/
data_file: processed/IoT-23.csv # processed/EdgeIIoTset.csv # relative to data_dir
result_dir: results/
debug: False # mode: turn debugging mode True for preprocessing on debug_rows samples.
debug_rows: 30000 # if debug: True, consider the number of data rows for the experiment.
minority_threshold: 500 # filter out the minority classes with threshold value.
apply_balancer: False # True if you want to apply the classes' balancer
train: True
test: True

preprocess:
  iot23_url: https://mcfp.felk.cvut.cz/publicDatasets/IoT-23-Dataset/iot_23_datasets_small.tar.gz
  edgeIIoT: https://ieee-dataport.org/documents/edge-iiotset-new-comprehensive-realistic-cyber-security-dataset-iot-and-iiot-applications#
  raw_dir: data/local/2022-04-19
  save_csv:
    chunk # If 'chunk', save converted rows from text files into the equal size of csv files specified by chunk_size,
    # If 'full', it converts each text file into csv file.
  chunk_size: 1000000
  # a single raw file or dir of all .csv files with path relative to data_dir
  raw_data: raw/EdgeIIoTset.csv
  normalized_data: normalized/EdgeIIoTset.csv # relative to data_dir
  processed_data: processed/EdgeIIoTset.csv # relative to data_dir
  include_history: False # include history column or not
  num_features: 30 # number of features to select

model: # settings for training and testing experiments
  seed: 41 # seed for reproduciability to generate same set of samples of data splits
  type: multiclass # binary or multiclass
  name: dnn # SVM or RandomForest = rf
  k-fold: 20 # number of cross-validation folds
  save: True # save the trained model (True) or not (False)
  split_ratio: 0.2 # ratio of training and testing ratio
  use_neptune:
    True # Boolean value to switch between enabling neptune usage or not
    # for tracking the experiments in neptune.ai,
    # If True, make sure you have correctly configured '.neptune_config.ini' file
dnn:
  epochs: 204 # default 200
  batch: 16
  lr: 0.01
  patience: 100
  optimizer: adam # optimizer can be adam or sgd or any other
  dropout: 0.2 # dropout rate
  verbose: 2 # display the running information of the trainig or hide, 2 means show the info
  loss: categorical_crossentropy # categorical_crossentropy for multiclass or binary_crossentropy for binary classification