def makeMaxHeightTree(X_train, y_train, X_test, y_test, attrs):
clf = DecisionTreeClassifier(max_height=-1)
clf.fit(X_train, y_train, attrs)
y_pred = clf.predict(X_train)
train_accuracy = accuracy(y_train, y_pred)
y_pred = clf.predict(X_test)
test_accuracy = accuracy(y_test, y_pred)
return train_accuracy, test_accuracy, clf
def makeTrees(X_train, y_train, X_test, y_test, attrs, start, stop):
train_accuracies = OrderedDict()
test_accuracies = OrderedDict()
clfs = OrderedDict()
for i in range(start, stop):
print("MaxHeight =", i)
clf = DecisionTreeClassifier(max_height=i)
clf.fit(X_train, y_train, attrs)
clfs[i] = clf
y_pred = clf.predict(X_train)
train_accuracies[i] = accuracy(y_train, y_pred)
CI = confidence_interval(y_train, y_pred)
print("Training Accuracy:", train_accuracies[i])
print("Confidence interval on training sample:", CI)
y_pred = clf.predict(X_test)
test_accuracies[i] = accuracy(y_test, y_pred)
CI = confidence_interval(y_test, y_pred)
print("Test Accuracy:", test_accuracies[i])
print("Confidence interval on test sample:", CI)
return train_accuracies, test_accuracies, clfs
def runAll():
print(
"Reading data and splitting into training, validation and test sets.")
df = pd.read_csv(DATA)
y = df["label"]
X = df.drop(columns=["label"])
attrs = [(a, "Continuous") for a in X.columns.tolist()]
data_splits = []
for _ in range(NUM_SPLITS):
data_splits.append(
test_train_split(X, y, 0.8, random_state=random.randint(100,
1000)))
print("Part 1: Tree with full height")
results = []
i = 0
for X_train, y_train, X_test, y_test in data_splits:
print("Generating for training set:", i)
i += 1
results.append(
makeMaxHeightTree(X_train, y_train, X_test, y_test, attrs))
train_accs, test_accs, clfs = zip(*results)
print("Average training accuracy:", sum(train_accs) / len(train_accs))
print("Max training accuracy:", max(train_accs))
print("Average test accuracy:", sum(test_accs) / len(test_accs))
print("Max test accuracy:", max(test_accs))
print("Saving tree with best test accuracy")
maxIdx = test_accs.index(max(test_accs))
clfs[maxIdx].plot(name="Max_height", storepath=OUTPUT_DIR)
print("Part 1 End")
print("Part 2: Best Depth Selection")
X_train, y_train, X_test, y_test = data_splits[0]
X_grow, y_grow, X_val, y_val = test_train_split(
X_train, y_train, 0.8, random_state=random.randint(100, 1000))
train_accuracy, test_accuracy, clfs = makeTrees(X_grow, y_grow, X_test,
y_test, attrs, 2, 16)
plt.xlabel("Max Height")
plt.ylabel("Accuracy")
px1, py1 = zip(*list(train_accuracy.items()))
plt.plot(px1, py1, label="Training Accuracy")
px2, py2 = zip(*list(test_accuracy.items()))
plt.plot(px2, py2, label="Test Accuracy")
plt.legend()
print("Saving accuracy vs height plot")
plt.savefig(os.path.join(OUTPUT_DIR, "accuracy.png"))
print("Part 2 End")
BEST_HEIGHT = 10 # Obtained from the accuracy graph
print("Part 3: Pruning the best height tree")
clf = clfs[BEST_HEIGHT]
clf.prune(X_val, y_val)
y_pred = clf.predict(X_test)
print("Accuracy after pruning:", accuracy(y_test, y_pred))
print("Print 3 End")
print("Part 4: Printing the Decision Tree")
print("Saving the tree as Final_tree.pdf")
clf.plot(name="Final_tree", storepath=OUTPUT_DIR)
print("Part 4 End")