BigML Bindings: 101 - Using a Fusion Model

Following the schema described in the prediction workflow, document, this is the code snippet that shows the minimal workflow to create a fusion model (assuming that some component models have already been created) and produce a single prediction.

from bigml.api import BigML
# step 0: creating a connection to the service (default credentials)
api = BigML()
# step 5: creating a fusion model from a preexisting model and a logistic
#         regression with equal weight
fusion = api.create_fusion({["id": "model/1111111111111111111111111",
                             "weight": 1},
                            {"id": "logisticregression/222222222222222222222222",
                             "weight": 1}])
# waiting for the fusion to be finished
api.ok(fusion)
# the new input data to predict for
input_data = {"petal width": 1.75, "petal length": 2.45}
# creating a single prediction
prediction = api.create_prediction(fusion, input_data)

If you want to create predictions for many new inputs, you can do so by creating a batch_prediction resource. First, you will need to upload to the platform all the input data that you want to predict for and create the corresponding source and dataset resources. In the example, we’ll be assuming you already created a model following the steps 0 to 5 in the previous snippet.

# step 6: creating a source from the data in your local "data/test_iris.csv" file
test_source = api.create_source("data/test_iris.csv")
# waiting for the source to be finished. Results will be stored in `source`
api.ok(test_source)
# step 8: creating a dataset from the previously created `source`
test_dataset = api.create_dataset(test_source)
# waiting for the dataset to be finished
api.ok(test_dataset)
# step 10: creating a batch prediction
batch_prediction = api.create_batch_prediction(fusion, test_dataset)
# waiting for the batch_prediction to be finished
api.ok(batch_prediction)
# downloading the results to your computer
api.download_batch_prediction(batch_prediction,
                              filename='my_dir/my_predictions.csv')

The batch prediction output (as well as any of the resources created) can be configured using additional arguments in the corresponding create calls. For instance, to include all the information in the original dataset in the output you would change step 10 to:

batch_prediction = api.create_batch_prediction(fusion, test_dataset,
                                               {"all_fields": True})

Check the API documentation to learn about the available configuration options for any BigML resource.

You can also predict locally using the Fusion class in the fusion module. A simple example of that is:

from bigml.fusion import Fusion
local_fusion = Fusion("fusion/5968ec46983efc21b000001b")
# predicting for some input data
local_fusion.predict({"petal length": 2.45, "sepal length": 2,
                      "petal width": 1.75, "sepal witdh": 3})

Or you could store first your fusion information (together with the included models) in a file per model and use those files to create the local Fusion object:

# downloading the model JSON to a local file
from bigml.api import BigML
api = BigML()
api.export("fusion/5968ec46983efc21b000001b",
           "filename": "my_dir/my_fusion.json")
# creating the local fusion from the file
from bigml.fusion import Fusion
local_fusion = Model("my_dir/my_fusion.json")
# predicting for some input data
local_fusion.predict({"petal length": 2.45, "sepal length": 2,
                      "petal width": 1.75, "sepal witdh": 3})

And if you want to predict locally for all the rows in a CSV file (first line should contain the field headers):

import csv
from bigml.fusion import Fusion
local_fusion = Fusion("fusion/5a414c667811dd5057000ab5")
with open("test_data.csv") as test_handler:
    reader = csv.DictReader(test_handler)
    for input_data in reader:
    # predicting for all rows
        print local_fusion.predict(input_data)

Every modeling resource in BigML has its corresponding local class. Check the Local resources section of the documentation to learn more about them.