Generalization Error

MA CH IN E LEA RN IN G W ITH TREE-BA S ED MODELS IN P YTH ON

Elie Kawerk
Data Scientist
Supervised Learning - Under the Hood
Supervised Learning: y = f(x), f is unknown.

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Goals of Supervised Learning
Find a model f^ that best approximates f : f^ ≈ f
f^ can be Logistic Regression, Decision Tree, Neural Network ...
Discard noise as much as possible.

End goal: f^ should acheive a low predictive error on unseen datasets.

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Dif culties in Approximating f
Over tting:f^(x) ts the training set noise.
Under tting: f^ is not exible enough to approximate f .

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Over tting

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Under tting

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Generalization Error
Generalization Error of f^: Does f^ generalize well on unseen data?
It can be decomposed as follows: Generalization Error of
f^ = bias2 + variance + irreducible error

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Bias
Bias: error term that tells you, on average, how much f^ ≠ f .

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Variance
Variance: tells you how much f^ is inconsistent over different training sets.

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Model Complexity
Model Complexity: sets the exibility of f^.
Example: Maximum tree depth, Minimum samples per leaf, ...

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Bias-Variance Tradeoff

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Bias-Variance Tradeoff: A Visual Explanation

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

 Let's practice!
MA CH IN E LEA RN IN G W ITH TREE-BA S ED MODELS IN P YTH ON
 Diagnosing Bias and
Variance Problems
MA CH IN E LEA RN IN G W ITH TREE-BA S ED MODELS IN P YTH ON

Elie Kawerk
Data Scientist
Estimating the Generalization Error
How do we estimate the generalization error of a model?

Cannot be done directly because:

f is unknown,
usually you only have one dataset,

noise is unpredictable.

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Estimating the Generalization Error
Solution:

split the data to training and test sets,

tf^ to the training set,

evaluate the error of f^ on the unseen test set.

generalization error of f^ ≈ test set error of f^.

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Better Model Evaluation with Cross-Validation
Test set should not be touched until we are con dent about f^'s performance.
Evaluating f^ on training set: biased estimate, f^ has already seen all training points.
Solution → Cross-Validation (CV):
K-Fold CV,

Hold-Out CV.

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

K-Fold CV

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

K-Fold CV

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Diagnose Variance Problems
If f^ suffers from high variance: CV error of f^ > training set error of f^.
f^ is said to over t the training set. To remedy over tting:
decrease model complexity,

for ex: decrease max depth, increase min samples per leaf, ...

gather more data, ..

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Diagnose Bias Problems
iff^ suffers from high bias: CV error of f^ ≈ training set error of f^ >> desired error.
f^ is said to under t the training set. To remedy under tting:
increase model complexity

for ex: increase max depth, decrease min samples per leaf, ...

gather more relevant features

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

K-Fold CV in sklearn on the Auto Dataset
from sklearn.tree import DecisionTreeRegressor
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error as MSE
from sklearn.model_selection import cross_val_score

# Set seed for reproducibility

SEED = 123
# Split data into 70% train and 30% test
X_train, X_test, y_train, y_test = train_test_split(X,y,
test_size=0.3,
random_state=SEED)

# Instantiate decision tree regressor and assign it to 'dt'

dt = DecisionTreeRegressor(max_depth=4,
min_samples_leaf=0.14,
random_state=SEED)

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

K-Fold CV in sklearn on the Auto Dataset
# Evaluate the list of MSE ontained by 10-fold CV
# Set n_jobs to -1 in order to exploit all CPU cores in computation
MSE_CV = - cross_val_score(dt, X_train, y_train, cv= 10,
scoring='neg_mean_squared_error',
n_jobs = -1)

# Fit 'dt' to the training set

dt.fit(X_train, y_train)
# Predict the labels of training set
y_predict_train = dt.predict(X_train)
# Predict the labels of test set
y_predict_test = dt.predict(X_test)

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

# CV MSE
print('CV MSE: {:.2f}'.format(MSE_CV.mean()))

CV MSE: 20.51

# Training set MSE

print('Train MSE: {:.2f}'.format(MSE(y_train, y_predict_train)))

Train MSE: 15.30

# Test set MSE

print('Test MSE: {:.2f}'.format(MSE(y_test, y_predict_test)))

Test MSE: 20.92

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

 Let's practice!
MA CH IN E LEA RN IN G W ITH TREE-BA S ED MODELS IN P YTH ON
 Ensemble Learning
MA CH IN E LEA RN IN G W ITH TREE-BA S ED MODELS IN P YTH ON

Elie Kawerk
Data Scientist
Advantages of CARTs
Simple to understand.

Simple to interpret.

Easy to use.

Flexibility: ability to describe non-linear dependencies.

Preprocessing: no need to standardize or normalize features, ...

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Limitations of CARTs
Classi cation: can only produce orthogonal decision boundaries.

Sensitive to small variations in the training set.

High variance: unconstrained CARTs may over t the training set.

Solution: ensemble learning.

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Ensemble Learning
Train different models on the same dataset.

Let each model make its predictions.

Meta-model: aggregates predictions of individual models.

Final prediction: more robust and less prone to errors.

Best results: models are skillful in different ways.

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Ensemble Learning: A Visual Explanation

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Ensemble Learning in Practice: Voting Classi er
Binary classi cation task.

N classi ers make predictions: P1 , P2 , ..., PN with Pi = 0 or 1.

Meta-model prediction: hard voting.

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Hard Voting

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Voting Classi er in sklearn (Breast-Cancer
dataset)
# Import functions to compute accuracy and split data
from sklearn.metrics import accuracy_score
from sklearn.model_selection import train_test_split

# Import models, including VotingClassifier meta-model

from sklearn.linear_model import LogisticRegression
from sklearn.tree import DecisionTreeClassifier
from sklearn.neighbors import KNeighborsClassifier as KNN
from sklearn.ensemble import VotingClassifier

# Set seed for reproducibility

SEED = 1

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Voting Classi er in sklearn (Breast-Cancer
dataset)
# Split data into 70% train and 30% test
X_train, X_test, y_train, y_test = train_test_split(X, y,
test_size= 0.3,
random_state= SEED)
# Instantiate individual classifiers
lr = LogisticRegression(random_state=SEED)
knn = KNN()
dt = DecisionTreeClassifier(random_state=SEED)

# Define a list called classifier that contains the tuples (classifier_name, classifier)
classifiers = [('Logistic Regression', lr),
('K Nearest Neighbours', knn),
('Classification Tree', dt)]

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

# Iterate over the defined list of tuples containing the classifiers
for clf_name, clf in classifiers:
#fit clf to the training set
clf.fit(X_train, y_train)

# Predict the labels of the test set

y_pred = clf.predict(X_test)

# Evaluate the accuracy of clf on the test set

print('{:s} : {:.3f}'.format(clf_name, accuracy_score(y_test, y_pred)))

Logistic Regression: 0.947

K Nearest Neighbours: 0.930
Classification Tree: 0.930

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

Voting Classi er in sklearn (Breast-Cancer
dataset)
# Instantiate a VotingClassifier 'vc'
vc = VotingClassifier(estimators=classifiers)

# Fit 'vc' to the traing set and predict test set labels
vc.fit(X_train, y_train)
y_pred = vc.predict(X_test)

# Evaluate the test-set accuracy of 'vc'

print('Voting Classifier: {.3f}'.format(accuracy_score(y_test, y_pred)))

Voting Classifier: 0.953

MACHINE LEARNING WITH TREE-BASED MODELS IN PYTHON

 Let's practice!
MA CH IN E LEA RN IN G W ITH TREE-BA S ED MODELS IN P YTH ON