BigML Bindings: 101 - Images Feature ExtractionΒΆ

Following the schema described in the prediction workflow, document, this is the code snippet that shows the minimal workflow to extract features from images and generate an enriched dataset that can be used to train any kind of model.

from bigml.api import BigML
# step 0: creating a connection to the service (default credentials)
api = BigML()
# step 1: creating a source from the data in your local
# "data/images/" file. The file contains two folders, each
# of which contains a collection of images. The folder name will be used
# as label for each image it contains.
# The source is created enabling image analysis and setting some of the
# available features (see the API documentation at
# for details). In particular, we extract histogram of gradients and
# average pixels.
extracted_features = ["average_pixels", "histogram_of_gradients"]
source = api.create_source("data/images/",
    args={"image_analysis": {"enabled": True,
                             "extracted_features": extracted_features}})
# waiting for the source to be finished. Results will be stored in `source`
# and the new extracted features will be generated.
# step 3: creating a dataset from the previously created `source`
dataset = api.create_dataset(source)
# waiting for the dataset to be finished
# step 5: creating an anomaly detector
anomaly = api.create_anomaly(dataset)
# waiting for the anomaly detector to be finished
# the new input data to predict for should contain the path to the
# new image to be used for testing
input_data = {"image_id": "data/images/f2/fruits2.png"}
# creating a single anomaly score: The image file is uploaded to BigML,
# a new source is created for it using the same image_analysis
# used in the image field, and its ID is used as value
# for the ``image_id`` field in the input data to generate the prediction
prediction = api.create_prediction(anomaly, input_data)

You can also create a local anomaly score using the Anomaly class in the anomaly module. A simple example of that is:

from bigml.anomaly import Anomaly
local_anomaly = Anomaly("anomaly/5968ec46983efc21b000001c")
# creating a pipeline to store the feature extraction transformations
feature_extraction_pipeline = local_anomaly.data_transformations()
# scoring for some input data. As pipelines transform lists of rows
# we build a list with the single input data and get the first
# element of the output list
input_data = feature_extraction_pipeline.transform(
    [{"image_id": "data/images/f2/fruits2.png"}])[0]