Getting started (Python)

Before we get started, ensure you have the required packages installed:

pip install reductionml scikit-learn matplotlib

from reductionml import Workspace, FormatType

First, we need to define the configuration we are going to use. You can think of this as the definition of the structure of the reduction stack, including all things that will affect learning and prediction. Things such as hyper parameters.

We will create an instance of Coin as the only reduction in our stack which solves for linear regression problems.

config = {
    "entryReduction": {
        "typename": "Coin"
    }, 
    "globalConfig": {}
}

workspace = Workspace.create_from_config(config)

We’re going to use the diabetes dataset from sklearn and plot the mean squared error after each example. We’re using the JSON format to provide the features to ReductionML.

We use predict_then_learn to predict the output of the model before learning and keep track of the progressive loss.s

import sklearn.datasets
import matplotlib.pyplot as plt


def format_input(features, label):
    return {
        "label": {
            "value": label,
            "weight": 1.0
        },
        "features": {
            ":default": features.tolist()
        }
    }

dataset = sklearn.datasets.load_diabetes()

parser = workspace.create_parser(FormatType.Json)

mse_error = 0
count = 0
mse_errors = []
for (input_features, input_label) in zip(dataset.data, dataset.target):
    features, label = parser.parse(format_input(input_features, input_label))
    prediction = workspace.predict_then_learn(features, label)
    mse_error += (label.value - prediction.prediction) ** 2
    count += 1
    mse_errors.append(mse_error / count)

plt.plot(mse_errors)

[<matplotlib.lines.Line2D at 0x17f36c9a0>]