Machine Learning
What is scikit-learn?
scikit-learn is the core library for classic ML algorithms in Python.
The design of scikit-learn
scikit-learn is the standard library of classic machine learning. Its strength comes less from the number of algorithms than from its consistent interface: every model shares the same few methods, so swapping one model for another is one line.
from sklearn.svm import SVC
from sklearn.ensemble import RandomForestClassifier
for m in [SVC(), RandomForestClassifier(random_state=0)]:
m.fit(X_train, y_train)
print(type(m).__name__, m.score(X_test, y_test))
Modules are split by job: linear_model, tree, ensemble, svm, preprocessing, model_selection, metrics.
Built-in datasets
sklearn.datasets offers small real datasets (load_*) and synthetic data generators (make_*). Generators are ideal for trying out a particular situation (imbalanced classes, noise, a non-linear boundary):
from sklearn.datasets import make_classification, make_moons
X, y = make_classification(n_samples=200, n_features=5, random_state=0)
Xm, ym = make_moons(n_samples=200, noise=0.2, random_state=0)
print(X.shape, Xm.shape) # (200, 5) (200, 2)
Without random_state, different data is generated on every run.
Exercise Three models, same code
Train LogisticRegression(max_iter=1000), SVC() and RandomForestClassifier(random_state=0) in a single loop. On each line print the class name (type(m).__name__) and the test accuracy with round(..., 3).
Show example solution
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
X, y = load_iris(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=0, stratify=y)
from sklearn.linear_model import LogisticRegression
from sklearn.svm import SVC
from sklearn.ensemble import RandomForestClassifier
for m in [LogisticRegression(max_iter=1000), SVC(), RandomForestClassifier(random_state=0)]:
m.fit(X_train, y_train)
print(type(m).__name__, round(m.score(X_test, y_test), 3))
Test yourself
-
Why is replacing one model with another easy in scikit-learn?
- All models are the same algorithm
- They take no parameters
- It isn't easy
- They all share the same fit/predict/score interface
Show answer
They all share the same fit/predict/score interface — Thanks to the shared interface, many models can be tried with the same code in a loop.
-
Generate synthetic classification data.
X, y = ____(n_samples=100, n_features=4, random_state=0)
Show answer
X, y = make_classification(n_samples=100, n_features=4, random_state=0)
-
Match the modules.
- linear_model
- Linear models
- ensemble
- Ensemble models
- model_selection
- Splitting and search
- preprocessing
- Scaling and encoding
Next lesson: Estimator API
