Pipelines

EvalML pipelines represent a sequence of operations to be applied to data, where each operation is either a data transformation or an ML modeling algorithm.

A pipeline holds a combination of one or more components, which will be applied to new input data in sequence.

Each component and pipeline supports a set of parameters which configure its behavior. The AutoML search process seeks to find the combination of pipeline structure and pipeline parameters which perform the best on the data.

Defining a Pipeline Instance

Pipeline instances can be instantiated using any of the following classes:

  • RegressionPipeline

  • BinaryClassificationPipeline

  • MulticlassClassificationPipeline

  • TimeSeriesRegressionPipeline

  • TimeSeriesBinaryClassificationPipeline

  • TimeSeriesMulticlassClassificationPipeline

The class you want to use will depend on your problem type. The only required parameter input for instantiating a pipeline instance is component_graph, which can be a ComponentGraph instance, a list, or a dictionary containing a sequence of components to be fit and evaluated.

A component_graph list is the default representation, which represents a linear order of transforming components with an estimator as the final component. A component_graph dictionary is used to represent a non-linear graph of components, where the key is a unique name for each component and the value is a list with the component’s class as the first element and any parents of the component as the following element(s). For these two component_graph formats, each component can be provided as a reference to the component class for custom components, and as either a string name or as a reference to the component class for components defined in EvalML.

If you choose to provide a ComponentGraph instance and want to set custom parameters for your pipeline, set it through the pipeline initialization rather than ComponentGraph.instantiate().

[1]:
from evalml.pipelines import MulticlassClassificationPipeline, ComponentGraph

component_graph_as_list = ['Imputer', 'Random Forest Classifier']
MulticlassClassificationPipeline(component_graph=component_graph_as_list)
[1]:
pipeline = MulticlassClassificationPipeline(component_graph={'Imputer': ['Imputer', 'X', 'y'], 'Random Forest Classifier': ['Random Forest Classifier', 'Imputer.x', 'y']}, parameters={'Imputer':{'categorical_impute_strategy': 'most_frequent', 'numeric_impute_strategy': 'mean', 'categorical_fill_value': None, 'numeric_fill_value': None}, 'Random Forest Classifier':{'n_estimators': 100, 'max_depth': 6, 'n_jobs': -1}}, random_seed=0)
[2]:
component_graph_as_dict = {
    'Imputer': ['Imputer', 'X', 'y'],
    'Encoder': ['One Hot Encoder', 'Imputer.x', 'y'],
    'Random Forest Clf': ['Random Forest Classifier', 'Encoder.x', 'y'],
    'Elastic Net Clf': ['Elastic Net Classifier', 'Encoder.x', 'y'],
    'Final Estimator': ['Logistic Regression Classifier', 'Random Forest Clf.x', 'Elastic Net Clf.x', 'y']
}

MulticlassClassificationPipeline(component_graph=component_graph_as_dict)
[2]:
pipeline = MulticlassClassificationPipeline(component_graph={'Imputer': ['Imputer', 'X', 'y'], 'Encoder': ['One Hot Encoder', 'Imputer.x', 'y'], 'Random Forest Clf': ['Random Forest Classifier', 'Encoder.x', 'y'], 'Elastic Net Clf': ['Elastic Net Classifier', 'Encoder.x', 'y'], 'Final Estimator': ['Logistic Regression Classifier', 'Random Forest Clf.x', 'Elastic Net Clf.x', 'y']}, parameters={'Imputer':{'categorical_impute_strategy': 'most_frequent', 'numeric_impute_strategy': 'mean', 'categorical_fill_value': None, 'numeric_fill_value': None}, 'Encoder':{'top_n': 10, 'features_to_encode': None, 'categories': None, 'drop': 'if_binary', 'handle_unknown': 'ignore', 'handle_missing': 'error'}, 'Random Forest Clf':{'n_estimators': 100, 'max_depth': 6, 'n_jobs': -1}, 'Elastic Net Clf':{'penalty': 'elasticnet', 'C': 1.0, 'l1_ratio': 0.15, 'n_jobs': -1, 'multi_class': 'auto', 'solver': 'saga'}, 'Final Estimator':{'penalty': 'l2', 'C': 1.0, 'n_jobs': -1, 'multi_class': 'auto', 'solver': 'lbfgs'}}, random_seed=0)
[3]:
cg = ComponentGraph(component_graph_as_dict)

# set parameters in the pipeline rather than through cg.instantiate()
MulticlassClassificationPipeline(component_graph=cg, parameters={})
[3]:
pipeline = MulticlassClassificationPipeline(component_graph={'Imputer': ['Imputer', 'X', 'y'], 'Encoder': ['One Hot Encoder', 'Imputer.x', 'y'], 'Random Forest Clf': ['Random Forest Classifier', 'Encoder.x', 'y'], 'Elastic Net Clf': ['Elastic Net Classifier', 'Encoder.x', 'y'], 'Final Estimator': ['Logistic Regression Classifier', 'Random Forest Clf.x', 'Elastic Net Clf.x', 'y']}, parameters={'Imputer':{'categorical_impute_strategy': 'most_frequent', 'numeric_impute_strategy': 'mean', 'categorical_fill_value': None, 'numeric_fill_value': None}, 'Encoder':{'top_n': 10, 'features_to_encode': None, 'categories': None, 'drop': 'if_binary', 'handle_unknown': 'ignore', 'handle_missing': 'error'}, 'Random Forest Clf':{'n_estimators': 100, 'max_depth': 6, 'n_jobs': -1}, 'Elastic Net Clf':{'penalty': 'elasticnet', 'C': 1.0, 'l1_ratio': 0.15, 'n_jobs': -1, 'multi_class': 'auto', 'solver': 'saga'}, 'Final Estimator':{'penalty': 'l2', 'C': 1.0, 'n_jobs': -1, 'multi_class': 'auto', 'solver': 'lbfgs'}}, random_seed=0)

If you’re using your own custom components you can refer to them like so:

[4]:
from evalml.pipelines.components import Transformer

class NewTransformer(Transformer):
    name = 'New Transformer'
    hyperparameter_ranges = {
        "parameter_1":['a', 'b', 'c']
    }

    def __init__(self, parameter_1=1, random_seed=0):
        parameters = {"parameter_1": parameter_1}
        super().__init__(parameters=parameters,
                         random_seed=random_seed)

    def transform(self, X, y=None):
        # Your code here!
        return X

MulticlassClassificationPipeline([NewTransformer, 'Random Forest Classifier'])
[4]:
pipeline = MulticlassClassificationPipeline(component_graph={'New Transformer': [NewTransformer, 'X', 'y'], 'Random Forest Classifier': ['Random Forest Classifier', 'New Transformer.x', 'y']}, parameters={'New Transformer':{'parameter_1': 1}, 'Random Forest Classifier':{'n_estimators': 100, 'max_depth': 6, 'n_jobs': -1}}, random_seed=0)

Pipeline Usage

All pipelines define the following methods:

  • fit fits each component on the provided training data, in order.

  • predict computes the predictions of the component graph on the provided data.

  • score computes the value of an objective on the provided data.

[5]:
from evalml.demos import load_wine
X, y = load_wine()

pipeline = MulticlassClassificationPipeline(['Imputer', 'Random Forest Classifier'])
pipeline.fit(X, y)
print(pipeline.predict(X))
print(pipeline.score(X, y, objectives=['log loss multiclass']))
         Number of Features
Numeric                  13

Number of training examples: 178
Targets
class_1    39.89%
class_0    33.15%
class_2    26.97%
Name: target, dtype: object
0      class_0
1      class_0
2      class_0
3      class_0
4      class_0
        ...
173    class_2
174    class_2
175    class_2
176    class_2
177    class_2
Length: 178, dtype: category
Categories (3, object): ['class_0', 'class_1', 'class_2']
OrderedDict([('Log Loss Multiclass', 0.04132737017536148)])

Custom Name

By default, a pipeline’s name is created using the component graph that makes up the pipeline. E.g. A pipeline with an imputer, one-hot encoder, and logistic regression classifier will have the name ‘Logistic Regression Classifier w/ Imputer + One Hot Encoder’.

If you’d like to override the pipeline’s name attribute, you can set the custom_name parameter when initalizing a pipeline, like so:

[6]:
component_graph = ['Imputer', 'One Hot Encoder', 'Logistic Regression Classifier']
pipeline = MulticlassClassificationPipeline(component_graph)
print("Pipeline with default name:", pipeline.name)


pipeline_with_name = MulticlassClassificationPipeline(component_graph, custom_name="My cool custom pipeline")
print("Pipeline with custom name:", pipeline_with_name.name)
Pipeline with default name: Logistic Regression Classifier w/ Imputer + One Hot Encoder
Pipeline with custom name: My cool custom pipeline

Pipeline Parameters

You can also pass in custom parameters by using the parameters parameter, which will then be used when instantiating each component in component_graph. The parameters dictionary needs to be in the format of a two-layered dictionary where the key-value pairs are the component name and corresponding component parameters dictionary. The component parameters dictionary consists of (parameter name, parameter values) key-value pairs.

An example will be shown below. The API reference for component parameters can also be found [here] (../api_reference.rst#components).

[7]:
parameters = {
        'Imputer': {
            "categorical_impute_strategy": "most_frequent",
            "numeric_impute_strategy": "median"
        },
        'Logistic Regression Classifier': {
            'penalty': 'l2',
            'C': 1.0,
        }
    }
component_graph = ['Imputer', 'One Hot Encoder', 'Standard Scaler', 'Logistic Regression Classifier']
MulticlassClassificationPipeline(component_graph=component_graph, parameters=parameters)
[7]:
pipeline = MulticlassClassificationPipeline(component_graph={'Imputer': ['Imputer', 'X', 'y'], 'One Hot Encoder': ['One Hot Encoder', 'Imputer.x', 'y'], 'Standard Scaler': ['Standard Scaler', 'One Hot Encoder.x', 'y'], 'Logistic Regression Classifier': ['Logistic Regression Classifier', 'Standard Scaler.x', 'y']}, parameters={'Imputer':{'categorical_impute_strategy': 'most_frequent', 'numeric_impute_strategy': 'median', 'categorical_fill_value': None, 'numeric_fill_value': None}, 'One Hot Encoder':{'top_n': 10, 'features_to_encode': None, 'categories': None, 'drop': 'if_binary', 'handle_unknown': 'ignore', 'handle_missing': 'error'}, 'Logistic Regression Classifier':{'penalty': 'l2', 'C': 1.0, 'n_jobs': -1, 'multi_class': 'auto', 'solver': 'lbfgs'}}, random_seed=0)

Pipeline Description

You can call .graph() to see each component and its parameters. Each component takes in data and feeds it to the next.

[8]:
component_graph = ['Imputer', 'One Hot Encoder', 'Standard Scaler', 'Logistic Regression Classifier']
pipeline = MulticlassClassificationPipeline(component_graph=component_graph, parameters=parameters)
pipeline.graph()
[8]:
../_images/user_guide_pipelines_14_0.svg
[9]:
component_graph_as_dict = {
    'Imputer': ['Imputer', 'X', 'y'],
    'Encoder': ['One Hot Encoder', 'Imputer.x', 'y'],
    'Random Forest Clf': ['Random Forest Classifier', 'Encoder.x', 'y'],
    'Elastic Net Clf': ['Elastic Net Classifier', 'Encoder.x', 'y'],
    'Final Estimator': ['Logistic Regression Classifier', 'Random Forest Clf.x', 'Elastic Net Clf.x', 'y']
}

nonlinear_pipeline = MulticlassClassificationPipeline(component_graph=component_graph_as_dict)
nonlinear_pipeline.graph()
[9]:
../_images/user_guide_pipelines_15_0.svg

You can see a textual representation of the pipeline by calling .describe():

[10]:
pipeline.describe()

*********************************************************************************
* Logistic Regression Classifier w/ Imputer + One Hot Encoder + Standard Scaler *
*********************************************************************************

Problem Type: multiclass
Model Family: Linear

Pipeline Steps
==============
1. Imputer
         * categorical_impute_strategy : most_frequent
         * numeric_impute_strategy : median
         * categorical_fill_value : None
         * numeric_fill_value : None
2. One Hot Encoder
         * top_n : 10
         * features_to_encode : None
         * categories : None
         * drop : if_binary
         * handle_unknown : ignore
         * handle_missing : error
3. Standard Scaler
4. Logistic Regression Classifier
         * penalty : l2
         * C : 1.0
         * n_jobs : -1
         * multi_class : auto
         * solver : lbfgs
[11]:
nonlinear_pipeline.describe()

*******************************************************************************************************************
* Logistic Regression Classifier w/ Imputer + One Hot Encoder + Random Forest Classifier + Elastic Net Classifier *
*******************************************************************************************************************

Problem Type: multiclass
Model Family: Linear

Pipeline Steps
==============
1. Imputer
         * categorical_impute_strategy : most_frequent
         * numeric_impute_strategy : mean
         * categorical_fill_value : None
         * numeric_fill_value : None
2. One Hot Encoder
         * top_n : 10
         * features_to_encode : None
         * categories : None
         * drop : if_binary
         * handle_unknown : ignore
         * handle_missing : error
3. Random Forest Classifier
         * n_estimators : 100
         * max_depth : 6
         * n_jobs : -1
4. Elastic Net Classifier
         * penalty : elasticnet
         * C : 1.0
         * l1_ratio : 0.15
         * n_jobs : -1
         * multi_class : auto
         * solver : saga
5. Logistic Regression Classifier
         * penalty : l2
         * C : 1.0
         * n_jobs : -1
         * multi_class : auto
         * solver : lbfgs

Component Graph

You can use pipeline.get_component(name) and provide the component name to access any component (API reference here):

[12]:
pipeline.get_component('Imputer')
[12]:
Imputer(categorical_impute_strategy='most_frequent', numeric_impute_strategy='median', categorical_fill_value=None, numeric_fill_value=None)
[13]:
nonlinear_pipeline.get_component('Elastic Net Clf')
[13]:
ElasticNetClassifier(penalty='elasticnet', C=1.0, l1_ratio=0.15, n_jobs=-1, multi_class='auto', solver='saga')

Alternatively, you can index directly into the pipeline to get a component

[14]:
first_component = pipeline[0]
print(first_component.name)
Imputer
[15]:
nonlinear_pipeline['Final Estimator']
[15]:
LogisticRegressionClassifier(penalty='l2', C=1.0, n_jobs=-1, multi_class='auto', solver='lbfgs')

Pipeline Estimator

EvalML enforces that the last component of a linear pipeline is an estimator. You can access this estimator directly by using pipeline.estimator.

[16]:
pipeline.estimator
[16]:
LogisticRegressionClassifier(penalty='l2', C=1.0, n_jobs=-1, multi_class='auto', solver='lbfgs')

Input Feature Names

After a pipeline is fitted, you can access a pipeline’s input_feature_names attribute to obtain a dictionary containing a list of feature names passed to each component of the pipeline. This could be especially useful for debugging where a feature might have been dropped or detecting unexpected behavior.

[17]:
pipeline = MulticlassClassificationPipeline(['Imputer', 'Random Forest Classifier'])
pipeline.fit(X, y)
pipeline.input_feature_names
[17]:
{'Imputer': ['alcohol',
  'malic_acid',
  'ash',
  'alcalinity_of_ash',
  'magnesium',
  'total_phenols',
  'flavanoids',
  'nonflavanoid_phenols',
  'proanthocyanins',
  'color_intensity',
  'hue',
  'od280/od315_of_diluted_wines',
  'proline'],
 'Random Forest Classifier': ['alcohol',
  'malic_acid',
  'ash',
  'alcalinity_of_ash',
  'magnesium',
  'total_phenols',
  'flavanoids',
  'nonflavanoid_phenols',
  'proanthocyanins',
  'color_intensity',
  'hue',
  'od280/od315_of_diluted_wines',
  'proline']}

Saving and Loading Pipelines

You can save and load trained or untrained pipeline instances using the Python pickle format, like so:

[18]:
import pickle

pipeline_to_pickle = MulticlassClassificationPipeline(['Imputer', 'Random Forest Classifier'])

with open("pipeline.pkl", 'wb') as f:
    pickle.dump(pipeline_to_pickle, f)

pickled_pipeline = None
with open('pipeline.pkl', 'rb') as f:
    pickled_pipeline = pickle.load(f)

assert pickled_pipeline == pipeline_to_pickle
pickled_pipeline.fit(X, y)
[18]:
pipeline = MulticlassClassificationPipeline(component_graph={'Imputer': ['Imputer', 'X', 'y'], 'Random Forest Classifier': ['Random Forest Classifier', 'Imputer.x', 'y']}, parameters={'Imputer':{'categorical_impute_strategy': 'most_frequent', 'numeric_impute_strategy': 'mean', 'categorical_fill_value': None, 'numeric_fill_value': None}, 'Random Forest Classifier':{'n_estimators': 100, 'max_depth': 6, 'n_jobs': -1}}, random_seed=0)

Generate Code

Once you have instantiated a pipeline, you can generate string Python code to recreate this pipeline, which can then be saved and run elsewhere with EvalML. generate_pipeline_code requires a pipeline instance as the input. It can also handle custom components, but it won’t return the code required to define the component. Note that any external libraries used in creating the pipeline instance will also need to be imported to execute the returned code.

Code generation is not yet supported for nonlinear pipelines.

[19]:
from evalml.pipelines.utils import generate_pipeline_code
from evalml.pipelines import MulticlassClassificationPipeline
import pandas as pd
from evalml.utils import infer_feature_types
from skopt.space import Integer

class MyDropNullColumns(Transformer):
    """Transformer to drop features whose percentage of NaN values exceeds a specified threshold"""
    name = "My Drop Null Columns Transformer"
    hyperparameter_ranges = {}

    def __init__(self, pct_null_threshold=1.0, random_seed=0, **kwargs):
        """Initalizes an transformer to drop features whose percentage of NaN values exceeds a specified threshold.

        Args:
            pct_null_threshold(float): The percentage of NaN values in an input feature to drop.
                Must be a value between [0, 1] inclusive. If equal to 0.0, will drop columns with any null values.
                If equal to 1.0, will drop columns with all null values. Defaults to 0.95.
        """
        if pct_null_threshold < 0 or pct_null_threshold > 1:
            raise ValueError("pct_null_threshold must be a float between 0 and 1, inclusive.")
        parameters = {"pct_null_threshold": pct_null_threshold}
        parameters.update(kwargs)

        self._cols_to_drop = None
        super().__init__(parameters=parameters,
                         component_obj=None,
                         random_seed=random_seed)

    def fit(self, X, y=None):
        pct_null_threshold = self.parameters["pct_null_threshold"]
        X = infer_feature_types(X)
        percent_null = X.isnull().mean()
        if pct_null_threshold == 0.0:
            null_cols = percent_null[percent_null > 0]
        else:
            null_cols = percent_null[percent_null >= pct_null_threshold]
        self._cols_to_drop = list(null_cols.index)
        return self

    def transform(self, X, y=None):
        """Transforms data X by dropping columns that exceed the threshold of null values.
        Args:
            X (pd.DataFrame): Data to transform
            y (pd.Series, optional): Targets
        Returns:
            pd.DataFrame: Transformed X
        """

        X = infer_feature_types(X)
        return X.drop(columns=self._cols_to_drop)


pipeline_instance = MulticlassClassificationPipeline(['Imputer', MyDropNullColumns,
                                                      'DateTime Featurization Component',
                                                      'Text Featurization Component',
                                                      'One Hot Encoder', 'Random Forest Classifier'],
                                                     custom_name="Pipeline with Custom Component",
                                                     random_seed=20)

code = generate_pipeline_code(pipeline_instance)
print(code)

# This string can then be pasted into a separate window and run, although since the pipeline has custom component `MyDropNullColumns`,
#      the code for that component must also be included
from evalml.demos import load_fraud
X, y = load_fraud(1000)
exec(code)
pipeline.fit(X, y)
from evalml.pipelines.multiclass_classification_pipeline import MulticlassClassificationPipeline
pipeline = MulticlassClassificationPipeline(component_graph={'Imputer': ['Imputer', 'X', 'y'], 'My Drop Null Columns Transformer': [MyDropNullColumns, 'Imputer.x', 'y'], 'DateTime Featurization Component': ['DateTime Featurization Component', 'My Drop Null Columns Transformer.x', 'y'], 'Text Featurization Component': ['Text Featurization Component', 'DateTime Featurization Component.x', 'y'], 'One Hot Encoder': ['One Hot Encoder', 'Text Featurization Component.x', 'y'], 'Random Forest Classifier': ['Random Forest Classifier', 'One Hot Encoder.x', 'y']}, parameters={'Imputer':{'categorical_impute_strategy': 'most_frequent', 'numeric_impute_strategy': 'mean', 'categorical_fill_value': None, 'numeric_fill_value': None}, 'My Drop Null Columns Transformer':{'pct_null_threshold': 1.0}, 'DateTime Featurization Component':{'features_to_extract': ['year', 'month', 'day_of_week', 'hour'], 'encode_as_categories': False, 'date_index': None}, 'One Hot Encoder':{'top_n': 10, 'features_to_encode': None, 'categories': None, 'drop': 'if_binary', 'handle_unknown': 'ignore', 'handle_missing': 'error'}, 'Random Forest Classifier':{'n_estimators': 100, 'max_depth': 6, 'n_jobs': -1}}, custom_name='Pipeline with Custom Component', random_seed=20)
             Number of Features
Boolean                       1
Categorical                   6
Numeric                       5

Number of training examples: 1000
Targets
False    85.90%
True     14.10%
Name: fraud, dtype: object
[19]:
pipeline = MulticlassClassificationPipeline(component_graph={'Imputer': ['Imputer', 'X', 'y'], 'My Drop Null Columns Transformer': [MyDropNullColumns, 'Imputer.x', 'y'], 'DateTime Featurization Component': ['DateTime Featurization Component', 'My Drop Null Columns Transformer.x', 'y'], 'Text Featurization Component': ['Text Featurization Component', 'DateTime Featurization Component.x', 'y'], 'One Hot Encoder': ['One Hot Encoder', 'Text Featurization Component.x', 'y'], 'Random Forest Classifier': ['Random Forest Classifier', 'One Hot Encoder.x', 'y']}, parameters={'Imputer':{'categorical_impute_strategy': 'most_frequent', 'numeric_impute_strategy': 'mean', 'categorical_fill_value': None, 'numeric_fill_value': None}, 'My Drop Null Columns Transformer':{'pct_null_threshold': 1.0}, 'DateTime Featurization Component':{'features_to_extract': ['year', 'month', 'day_of_week', 'hour'], 'encode_as_categories': False, 'date_index': None}, 'One Hot Encoder':{'top_n': 10, 'features_to_encode': None, 'categories': None, 'drop': 'if_binary', 'handle_unknown': 'ignore', 'handle_missing': 'error'}, 'Random Forest Classifier':{'n_estimators': 100, 'max_depth': 6, 'n_jobs': -1}}, custom_name='Pipeline with Custom Component', random_seed=20)