Examples¶
The following examples are meant to be executed in sequence, as they rely on previous steps, e.g., on data present.
Start the Java Virtual Machine¶
sklearn-weka-plugin simply imports everything from weka.core.jvm into sklweka.jvm for convenience:
import sklweka.jvm as jvm
# start JVM with Weka package support
jvm.start(packages=True)
Location of the datasets¶
The following examples assume the datasets to be present in the data_dir directory. For instance, this could be the following directory:
data_dir = "/my/datasets/"
Loading data¶
There are two methods available from the sklweka.dataset module for loading data:
load_arff - uses the loadarff method from the module scipy.io.arff (cannot handle string attributes). Nominal classes have to be converted using the to_nominal_labels method.
load_dataset - uses Weka’s own data loading functionality before converting it into sklearn data structures, i.e., numpy matrices. Bit slower due to data conversion, but handles string attributes. Also not limited to ARFF files. The data can be either returned with mixed types (not necessary to use the to_nominal_labels method then) or in Weka’s internal, numeric-only data format.
Regression¶
Regression algorithms are available through the sklweka.classifiers.WekaEstimator wrapper:
from sklweka.dataset import load_arff
from sklweka.classifiers import WekaEstimator
from sklearn.model_selection import cross_val_score
X, y, meta = load_arff(data_dir + "/bolts.arff", class_index="last")
lr = WekaEstimator(classname="weka.classifiers.functions.LinearRegression")
scores = cross_val_score(lr, X, y, cv=10, scoring='neg_root_mean_squared_error')
print("Cross-validating LR on bolts (negRMSE)\n", scores)
Classification¶
Classification algorithms are also available through the sklweka.classifiers.WekaEstimator wrapper:
from sklweka.dataset import load_arff, to_nominal_labels, load_dataset
from sklweka.classifiers import WekaEstimator
from sklearn.model_selection import cross_val_score
# using the load_arff method:
X, y, meta = load_arff(data_dir + "/iris.arff", class_index="last")
y = to_nominal_labels(y)
# using the load_dataset method:
# X, y = load_dataset(data_dir + "/iris.arff", class_index="last")
j48 = WekaEstimator(classname="weka.classifiers.trees.J48", options=["-M", "3"])
j48.fit(X, y)
scores = j48.predict(X)
probas = j48.predict_proba(X)
print("\nJ48 on iris\nactual label -> predicted label, probabilities")
for i in range(len(y)):
print(y[i], "->", scores[i], probas[i])
Alternatively to manually converting labels and nominal attributes, you can use the nominal_input_vars (list of 0-based indices or range string with 1-based indices) and nominal_output_var (bool) parameters in the constructor:
from sklweka.dataset import load_arff
from sklweka.classifiers import WekaEstimator
from sklearn.model_selection import cross_val_score
# using the load_arff method:
X, y, meta = load_arff(data_dir + "/vote.arff", class_index="last")
j48 = WekaEstimator(classname="weka.classifiers.trees.J48", options=["-M", "3"],
nominal_input_vars="first-last", nominal_output_var=True)
j48.fit(X, y)
scores = j48.predict(X)
probas = j48.predict_proba(X)
print("\nJ48 on iris\nactual label -> predicted label, probabilities")
for i in range(len(y)):
print(y[i], "->", scores[i], probas[i])
Clustering¶
Clustering algorithms are available through the sklweka.clusters.WekaCluster wrapper:
from sklweka.dataset import load_arff
from sklweka.clusters import WekaCluster
X, y, meta = load_arff(data_dir + "/iris.arff", class_index="last")
cl = WekaCluster(classname="weka.clusterers.SimpleKMeans", options=["-N", "3"])
clusters = cl.fit_predict(X)
print("\nSimpleKMeans on iris\nclass label -> cluster")
for i in range(len(y)):
print(y[i], "->", clusters[i])
Alternatively to manually converting labels and nominal attributes, you can use the nominal_input_vars (list of 0-based indices or range string with 1-based indices) parameter in the constructor:
from sklweka.dataset import load_arff
from sklweka.clusters import WekaCluster
X, y, meta = load_arff(data_dir + "/vote.arff", class_index="last")
cl = WekaCluster(classname="weka.clusterers.SimpleKMeans", options=["-N", "3"],
nominal_input_vars=[x for x in range(X.shape[1])])
clusters = cl.fit_predict(X)
print("\nSimpleKMeans on iris\nclass label -> cluster")
for i in range(len(y)):
print(y[i], "->", clusters[i])
Preprocessing¶
Weka filters can be applied by using the sklweka.preprocessing.WekaTransformer wrapper:
from sklweka.dataset import load_arff
from sklweka.preprocessing import WekaTransformer
X, y, meta = load_arff(data_dir + "/bolts.arff", class_index="last")
tr = WekaTransformer(classname="weka.filters.unsupervised.attribute.Standardize", options=["-unset-class-temporarily"])
X_new, y_new = tr.fit(X, y).transform(X, y)
print("\nStandardize filter")
print("\ntransformed X:\n", X_new)
print("\ntransformed y:\n", y_new)
Data generators¶
Weka’s data generators can be used for generating numpy arrays as well:
from sklweka.datagenerators import DataGenerator, generate_data
gen = DataGenerator(
classname="weka.datagenerators.classifiers.classification.BayesNet",
options=["-S", "2", "-n", "10", "-C", "10"])
X, y, X_names, y_name = generate_data(gen, att_names=True)
print("X:", X_names)
print(X)
print("y:", y_name)
print(y)
Stop the Java Virtual Machine¶
At end of your Python script, stop the JVM as follows:
jvm.stop()
NB: The JVM cannot be restarted within the same Python process, a drawback of the underlying javabridge library.
Additional examples¶
More examples can be found at:
