DL8.5 Predictor used for task involving label

PyDL8.5 allows users to write their own error function. This example shows how to write an error function based on transaction identifiers. PyDL8.5 will determine these transaction identifiers (tids) based on the occurrences of an itemset in the training data.

The error function is called very often, and calculating an error score based on tids can be time consuming. For classification tasks, it is highly recommended not to write an error function in Python that operators on lists of tids. check the plot_classifier_user_1.py example for a more efficient user-written error function in classification settings.

Moreover, this example shows how to use the DL85Predictor class. Using this class, the labels do not need to be passed to DL8.5. For classification tasks, the error function can also be specififed as a parameter of the DL85Classifier class. In this case, a standard implementation is used for filling in the class labels for leafs in the tree.

Another example of a user-specified error function is given in plot_cluster_user.py.

#####################################################################################
#      DL8.5 Predictor used for classification : user specific error function       #
#####################################################################################
Model building...
Model built. Duration of building = 0.6249
Accuracy DL8.5 on training set = 0.8243

import numpy as np
from sklearn.metrics import confusion_matrix
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
import time
from dl85 import DL85Predictor

dataset = np.genfromtxt("../datasets/anneal.txt", delimiter=' ')
X, y = dataset[:, 1:], dataset[:, 0]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=0)


print("#####################################################################################\n"
      "#      DL8.5 Predictor used for classification : user specific error function       #\n"
      "#####################################################################################")


# return the error and the majority class
def error(tids):
    classes, supports = np.unique(y_train.take(list(tids)), return_counts=True)
    maxindex = np.argmax(supports)
    return sum(supports) - supports[maxindex], classes[maxindex]


clf = DL85Predictor(max_depth=2, error_function=error, time_limit=600)
start = time.perf_counter()
print("Model building...")
clf.fit(X_train)
duration = time.perf_counter() - start
print("Model built. Duration of building =", round(duration, 4))
print("Accuracy DL8.5 on training set =", round(clf.accuracy_, 4))