Butterfly AI (0.1.1)

Returns a paginated list of all created datasets

Create a brand new dataset. A dataset represents a collection labelled data (rows in a CSV file) that can be used to train models. Upon creation, the dataset is initially empty and in PENDING status. Data must be added in CSV format using the endpoint POST /datasets/{datasetKey}/data

Retrieve dataset details by dataset key

Deletes a dataset

Adds data in CSV format to an existing (empty) dataset. On success, it returns back the dataset in PROCESSING status. Data can only be added to datasets in PENDING status.

The CSV file should be structured as follows:

• A header row must be included at the top
• The first column is treated as theID, a unique identifier for each row in the file
• The last column(s) vary depending on the type of data:
  • For binary data: a single last column containing the labels for negative (0, 'N', ...) or positive (1, 'Y', ...) samples
  • For multi-class data: multiple columns labelled with the class names added to the rightmost area of the CSV file
• Between the first and last columns, an arbitrary number of feature columns can be included. These features should be relevant to the domain and nature of the predictions. While there is no initial limit, it is recommended not to exceed 200 feature columns.

The dataset is prepared for training and when ready its status will become COMPLETED. Progress can be monitored via the returned datasetCreationProgressUrl.

Important Note This endpoint can only be used as-is if the CSV file is of less or equal to 32MB. If the file size is bigger than 32MB, use the uploadUrl returned after successful creation of the empty dataset with the X-Goog-Content-Length-Range header to directly PUT a larger file for processing.

curl --location --request PUT \ 
'{signedUrl}\
--header 'X-Goog-Content-Length-Range: 10,534773760' \
--header 'Content-Type: text/csv' \
--header 'Authorization: Bearer eyJhbGciOiJIUzI1NiJ9.eyJpc3MiOiJidXR0ZXJmbHktYWkiLCJzdWIiOiJzYWllZGFiZWRpMTk5OUBnbWFpbC5jb20iLCJuYmYiOjE3NTgyMzY3MjYsImV4cCI6MTc1ODI0MDMyNiwiaWF0IjoxNzU4MjM2NzI2LCJyb2xlcyI6W119.aeL5R_mONXFo2_HIBicZN5_55sNFudy512qYwXt4RbU' \
--data-binary '@/path/to/LoanRisk.csv'

List all training

List all training done by a given dataset paginated.

Creates a new training job for the dataset. The dataset must be in COMPLETED state to be able to run training from. Training only needs 2 mandatory hyperparameters:

• scalingFactor: a number between 4 and 400, an internal parameter that vary by use case. A reference document can be found (here)[https://butterfly-ai.mathficast.com]
• performanceThreshold: a floating point number between 0 and 1, indicating the desired performance at which training will stop. It represents the desired percentage of correct predictions (0 = 0%, 0.5 = 50%, 1 = 100%). Both 0 and 1 are allowed but they are unrealistic values. The smaller this value is, the faster training will completed.

4 algorithms run in parallel and compete achieve the desired performanceThreshold. Once one of them reaches it, the process finishes in COMPLETED status. Progress of the training process can be obtained using endpoint GET /api/training/3a1a9a83-d549-4f25-b7ed-8174e0c955de/progress

Get training job detail by key

Cancel a training job given its key

Get training job progress by its key. The status indicates the progress.

Creates a new prediction using the specified model and blind CSV file

Creates a new prediction using the specified model and blind CSV file

List all predictions done using a given model, paginated

Get prediction progress. The status indicates the progress

Get prediction

List all predictions done for a given dataset, paginated

List all models trained against a given dataset, paginated.

Get model detail