Custom Transformer

You can define Transformers in addition to the ones provided by GraphLab Create by inheriting from TransformerBase. You must implement the init, fit, and transform methods.

Introductory Example

import graphlab
from graphlab.toolkits.feature_engineering import TransformerBase

class MyTransformer(TransformerBase):

    def __init__(self):

    def fit(self, dataset):
        ''' Learn means during the fit stage.'''
        self.mean = {}
        for col in dataset.column_names():
            self.mean[col] = dataset[col].mean()
        return self

    def transform(self, dataset):
        ''' Subtract means during the transform stage.'''
        for col in dataset.column_names():
            dataset[col] = dataset[col] - dataset[col].mean()
        return dataset

# Dataset
dataset = graphlab.SFrame({"a": range(100) })            

# Create the model
model = graphlab.feature_engineering.create(dataset, MyTransformer())

# Transform new data
transformed_sf = model.transform(dataset)

# Save and load this model.'foo-bar')