# data nodes
P.add_nodes(
# node for training data
("Training Data", LabeledTrainingDataNode(digits.data[:-test],
digits.target[:-test])),
# node for validation data
("Validation Data", LabeledTestDataNode(digits.data[-test:],
digits.target[-test:])),
)
# create nodes using scikit-learn's support vector classification
# over logarithmic range of C from 0.01 to 1000
for exp in range(-1, 3):
# use one vs all node for multi-classification
# note use of node naming
P.add_node("SVM %s" % 10 ** exp, SklearnOneVsRestNode(svm.SVC, C=10 ** exp))
# create node for scoring the classifers
P.add_node("Metrics", MetricNode([accuracy_score, f1_score],
verbose=True))
# construct the DAG of the pipeline by defining layers using lists
# that are connected in full feedforward fashion, note the use of regexes
# in order to find nodes by their name
P << [["Training Data", "Validation Data"], "SVM\d*", "Metrics"]
P << ["Validation Data", "Metrics"]
# Dag Representation:
# ----> SVM 0.1 ---->
# Training Data ----> SVM 1 ----> Metrics <-------|
# Validation Data ----> SVM 10 ----> |
# | ----> SVM 1000 ----> |
# node for training data
("Training Data", LabeledTrainingDataNode(ft["pca.*"][:-test],
digits.target[:-test])),
# node for validation data
("Validation Data", LabeledTestDataNode(ft["pca.*"][-test:],
digits.target[-test:])),
)
# <codecell>
# create nodes using scikit-learn's support vector classification
# over logarithmic range of C from 0.01 to 1000
for exp in range(-1, 3):
# use one vs all node for multi-classification
# note use of node naming
P.add_node("Machine SVM %s" % 10 ** exp, SklearnOneVsRestNode(svm.SVC, C=10 ** exp))
# scikit's K-nearest neighbors and K-means algorithms
P.add_node("Machine KNN" , SklearnOneVsRestNode(neighbors.KNeighborsClassifier,warn_on_equidistant=False))
P.add_node("Machine K-means", EstimatorNode(cluster.KMeans,n_clusters=2))
# mass evaluator
P.add_node("Evaluator", MachineEvaluatorNode())
# <codecell>
# create cross validation, 3-fold
P.add_node("Cross Validation", SklearnCrossValidationNode(accuracy_score, score_weight=-1, verbose=True,top_k=4))
P.add_node("Visualize CV", CrossValidationVisualizationNode()) # see scores
# final metric