Getting started#

Installation#

Do you use pip?

anomalearn can be installed via pip since it is published on PyPI

pip install anomalearn

Do you use conda?

Currently, anomalearn has not been published on conda, but we plan to do so in the future.

Not yet available

Not satisfied with previous instructions?

Install from source? Do you want to know dependencies? Check the detailed installation tutorial.

Go to installation tutorial

Quick facts#

What is the purpose of anomalearn?

The purpose of anomalearn is to ship a complete structure to ease the development of new methods for anomaly detection. It wants to aid the process of creating new models, pre-processing operations, post-processing operations, transformers, pipelines or hyperparameter search algorithms.

To introduction tutorial

What kind of algorithms does anomalearn contain?

anomalearn will contain several types of algorithms that are directly used to perform anomaly detection (such as isolation forest), or algorithms which does not solve anomaly detection, but that are surrounded by processing objects or other algorithms to carry on the anomaly detection task.

To introduction tutorial

What kind of experiment helpers does anomalearn contain?

anomalearn will contain objects to aid the creation of experiments to aid scientific research in computer science. An example of helper for experiment creation is the loader of datasets, which enables to easily iterate over training/testing sets, either the default (when the dataset contain a pre-defined split) or a specifically user defined split.

To introduction tutorial

How do I read public datasets with anomalearn?

One of the features offered by anomalearn is that of dataset readers. These components enable the user to instantiate a data reader object to read the public dataset. These objects can be iterated and indexed with usual python syntax. The return type is always a pandas.DataFrame.

To introduction tutorial

How do I analyse data with anomalearn?

The analysis functions offered by anomalearn are all functions present in the analysis package. These functions needs data in input and have some default parameters to specify how the analysis must be performed. Some of the analysis functions are implemented using numba to be efficient.

To introduction tutorial

How do I perform hyperparameter search with anomalearn?

anomalearn contains several algorithms to do research on time series anomaly detection. The anomalearn.algorithms.tuning.hyperparameter package contains all the components related to hyperparameter tuning. Searchers can be instantiated to search for optimal parameters of any function.

To introduction tutorial

How do I integrate my legacy objects with anomalearn?

anomalearn follows a type-based approach with interfaces defining the methods. If your objects implement functions such as predict(), you can simply inherit from the function and it will be integrated. In some cases, interfaces also enable duck typing, which let any legacy object to work with anomalearn without the actual need of inheriting anomalearn’s objects.

To introduction tutorial

How do I create a model with anomalearn?

anomalearn models inherit from at least one of the classes (concrete or abstract) in anomalearn.algorithms. Each model in anomalearn must inherit from BaseModel, or from SavableModel if it can be safely serialized to file.

To introduction tutorial

How do I create a pre-processing object with anomalearn?

anomalearn pre-processing objects inherit from interfaces located in anomalearn.algorithms. These models typically inherit from the IShapeChanger interface, and should not be confused with transformers. They implement operations which happen always before a model and make sense only if coupled with a model.

To introduction tutorial

How do I create a post-processing object with anomalearn?

anomalearn post-processing objects inherit from interfaces located in anomalearn.algorithms. These models typically inherit from the IShapeChanger interface, and should not be confused with transformers. They implement operations which happen always after a model and make sense only if coupled with a model. Sometimes they also depend on the pre-processing placed before the model.

To introduction tutorial

How do I create a transformer with anomalearn?

anomalearn transformers are objects which transform data without changing the shape of data, e.g. anomalearn.algorithms.transformers.MinMaxScaler. These objects inherit from ITransformer.

To introduction tutorial

What is the difference between pre-, post-processing and transformers?

The first difference between processing and transformer objects is that the latter does not change the shape of the data in input. The former type of object explicitly changes the shape of the objects in input. However, both processing and transformer objects manipulate data. The difference between pre- and post- processing objects is that the former always happens before the model, the latter always happens after the model.

To introduction tutorial

How do I create a pipeline with anomalearn?

anomalearn contains the implementation of a sequential pipeline which can contain any sequence of objects following the layer interface. The Pipeline object can be created with a list of layers, and it is mutable. Any sequence of objects whose input/output shape are coherent can be created, and pipelines are layers themselves, i.e. a pipeline can contain another pipeline.

To introduction tutorial

Has anomalearn any helper function?

Yes, anomalearn contains several helper functions for the purpose of development of its main features. The helper packages are:

To introduction tutorial