anodilib 0.1.5

Python library ANODI for Time Series Anomaly Detection

ANODI

This is the Python library ANODI for Time Series Anomaly Detection. It offers easy access to algorithms and benchmark data.

Short Description

• the core of the ANODI library is the algorithm class (anodilib.algorithm) that wrapps an algorithm and the data that the algorithm should be fit on. It has - for example - the following attributes:
  • self.X_train_df, self.Y_train_df, self.X_test_df, self.Y_test_df: Dataframes for train and test data and labels
  • train_dls, test_dls: Dataloaders (fastAI Usage) for train and test data
  • learner: The Learner object for fastAI
  • most of the algorithms have additional attributes for window_length, stride and batch_size
• the data and some meta arguments are set up as a DatasetSpecification in the data module and given as argument to the algorithm object.
  • If not overriden by another test-DatasetSpecification in the predict function, the given data behind des DatasetSpecification is automatically split into the Dataframes and Dataloaders from above using the parameter test_fraction
• ANODI offers the following modules: algorithm, data, metrics, model(containing the underlying pytorch models), visualization

Example Usage

• Example Usage of IsolationForest Algorithm on the ECG200_TRAIN Dataset:

from anodilib.algorithm.IsolationForest import IsolationForest as IF
from anodilib.data.DatasetSpecification import ECG200_TRAIN

alg = IF(dataset_specification=ECG200_TRAIN, contamination="auto", test_fraction=0.2)
alg.fit()
alg.predict()
anoms = alg.getAnomalies()

print(anoms)

• Example Usage of ModifiedLSTM Algorithm on the ECG200_TRAIN Dataset using c_f1_score Metric:

from anodilib.algorithm.ModifiedLSTM import ModifiedLSTM
from anodilib.data.DatasetSpecification import ECG200_TRAIN
from metrics.cmetrics import c_f1_score
import numpy as np

alg = ModifiedLSTM(dataset_specification=ECG200_TRAIN, batch_size=28,window_len=4, stride=1)

alg.fit(epoch_num=50, learning_rate="lr_find",learning_rate_iteration=200)
alg.fit(epoch_num=50, learning_rate="lr_find",learning_rate_iteration=200)
alg.fit(epoch_num=50, learning_rate="lr_find",learning_rate_iteration=200)
alg.fit(epoch_num=50, learning_rate="lr_find",learning_rate_iteration=200)

alg.predict()

anoms, _ , _ = alg.getAnomalies()
print(anoms)
print(c_f1_score(anoms, np.array(alg.Y_test_df)))

• Example on how to put your local data into a DatasetSpecification (here: WADI Dataset)

from anodilib.data.DatasetSpecification import DatasetSpecification

WADI = DatasetSpecification(
    dataset_name="WADI.A2_19_Nov_2019_WADI_attackdataLABLE.csv",
    location=None,
    delimiter=",",
    label_column=130, # column containing the label
    is_data_column=lambda c: c != 130 and c >= 3, # use data of columns
    columns=None,
    skip_header=3,
    skip_footer=3,
    label_realizations=[1, -1],
    decimal=".",
    only_first_n_entries=-1, # if -1, use all entries
    predefined_dataset=None,
)

# use for algorithms ...

• for more information, have a look at the anodilib/tests/ or on our Githlib Repository

Dev Installation

This package is built using poetry, run the following code to install an editable version of the package for development

pip install poetry
poetry install