Contributing to Astronomaly

Astronomaly runs in a fully object-oriented framework. The main advantage of this is a great deal of functionality can be inherited from the base classes. The base pipeline class takes care of the following: * Automatic logging of all function calls * Automatic checks of whether the same function with the same arguments has already been called for given data (can be overridden with force_rerun=True) * Automatic saving and loading of output files (overridden with save_output=False) * (Potential) automatic parallelisation

Contributions are incredibly welcome! If you already have write access to the repository, please create a new well-named branch with your contributions and create a pull request when ready. Otherwise, please fork the repository and create a pull request when ready.

Contributing a new Dataset class

If your data is an image, time series or raw feature type, it is preferable to add flexibility to these existing classes rather than write a new class. If you need to read in an entirely new type of data however, a new Dataset class can be added. The new module should be added to the dataset_management folder and the new class should inherit from the base Dataset class like this:

from astronomaly.base.base_dataset import Dataset

class NewDatasetClass(Dataset):
  def __init__(an_argument=2, another_argument='hello', **kwargs):
    super().__init__(an_argument=an_argument, another_argument=another_argument, **kwargs)

Arguments must be explictly passed to the parent class’ init function to ensure correct logging. The functions get_sample and get_display_data must be implemented. The class variable metadata must be defined and should be a DataFrame with the index corresponding to a unique key for each sample. The class variable data_type must also be defined and should correspond to a known string in the webapp in order for the data to be displayed correctly.

Contributing a new PipelineStage class

Every step in the Astronomaly framework after the data has been read in can be represented with a PipelineStage object. If you would like to contribute a new feature extraction method or anomaly detection algorithm, these must be coded up as a class. Similar to the Dataset class, the PipelineStage class follows the following structure:

from astronomaly.base.base_pipeline import PipelineStage

class NewPipelineStage(PipelineStage):
  def __init__(an_argument=2, another_argument='hello', **kwargs):
    super().__init__(an_argument=an_argument, another_argument=another_argument, **kwargs)

  def _excute_function(self, data):
    <your actual function here>
    return output

The only function that has to be implemented in the new class is the _execute_function. This function should take a single input and produce a single output. All arguments required should be declared in __init__ as class variables so that they can be logged properly. The input type depends on whether this PipelineStage is meant to operate on the raw data (i.e. doing feature extraction) or any stage after that. _execute_function should never be called directly, instead a user should call one of two functions in the base PipelineStage class: run and run_on_dataset.

run_on_dataset, as the name suggests, should be called with a Dataset argument as input. It will automatically repeatedly call _execute_function on each sample in the Dataset, producing a DataFrame as output.

run on the other hand takes a DataFrame as input and then runs _execute_function on the entire DataFrame. This would be used for dimensionality reduction, anomaly detection etc.

Finally the new PipelineStage class should be put in a well-named module and placed in the appropriate folder within astronomaly.

Contributing to the webapp

If you have expertise in JavaScript, particularly with React, we’d welcome contributions to the front end. The files can be compiled using npm run watch in the webapp folder where changes can be viewed immediately.