Training model
- Machine Learning course
- Intermediate
- Lesson 32/112
- 7 min
Train an SVM model.
The chain and the search
The model chain: median filling → scaling → SVM. Against imbalance, class_weight="balanced" weights the minority class's errors by the inverse of the class share. Settings are searched by F1:
pipe = make_pipeline(SimpleImputer(strategy="median"), StandardScaler(),
SVC(class_weight="balanced"))
izgara = {"svc__C": [0.1, 1, 10], "svc__gamma": ["scale", 0.1]}
gs = GridSearchCV(pipe, izgara, cv=5, scoring="f1").fit(X_train, y_train)The median is more robust to outliers than the mean for filling missing values.
The baseline
To interpret the search result, we first set up an untuned baseline: the same chain, without class_weight and with default settings. Every step of the project is judged by what it adds over this baseline:
temel = make_pipeline(SimpleImputer(strategy="median"), StandardScaler(), SVC())
print(cross_val_score(temel, X_train, y_train, cv=5, scoring="f1").mean())
print(gs.best_score_)The comparison is done with CV on the training set; the test set is still untouched.
Practice Baseline and search
Compute the 5-fold F1 mean of the untuned baseline chain on the training set. Then search the class_weight="balanced" chain with the grid {"svc__C": [0.1, 1, 10], "svc__gamma": ["scale", 0.1]} using scoring="f1". Print the baseline score, best_score_ (both round(..., 3)) and best_params_ on separate lines.
Starter code
import numpy as np
from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split
X, y = make_classification(n_samples=600, n_features=8, n_informative=4, weights=[0.8, 0.2], random_state=7)
rng = np.random.default_rng(7)
X[rng.random(X.shape) < 0.05] = np.nan
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.25, random_state=0, stratify=y)
from sklearn.pipeline import make_pipeline
from sklearn.impute import SimpleImputer
from sklearn.preprocessing import StandardScaler
from sklearn.svm import SVC
from sklearn.model_selection import GridSearchCV, cross_val_score
# buraya yazSample solution
import numpy as np
from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split
X, y = make_classification(n_samples=600, n_features=8, n_informative=4, weights=[0.8, 0.2], random_state=7)
rng = np.random.default_rng(7)
X[rng.random(X.shape) < 0.05] = np.nan
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.25, random_state=0, stratify=y)
from sklearn.pipeline import make_pipeline
from sklearn.impute import SimpleImputer
from sklearn.preprocessing import StandardScaler
from sklearn.svm import SVC
from sklearn.model_selection import GridSearchCV, cross_val_score
temel = make_pipeline(SimpleImputer(strategy="median"), StandardScaler(), SVC())
print(round(cross_val_score(temel, X_train, y_train, cv=5, scoring="f1").mean(), 3))
pipe = make_pipeline(SimpleImputer(strategy="median"), StandardScaler(), SVC(class_weight="balanced"))
gs = GridSearchCV(pipe, {"svc__C": [0.1, 1, 10], "svc__gamma": ["scale", 0.1]}, cv=5, scoring="f1").fit(X_train, y_train)
print(round(gs.best_score_, 3))
print(gs.best_params_)Expected output
0.753
0.772
{'svc__C': 1, 'svc__gamma': 'scale'}
Questions and answers
Review the lesson questions together with their answers. To test yourself, solve the lesson interactively.
1 Roughly what weight does <code>class_weight="balanced"</code> give a 20% minority class (relative to the majority)?
- Equal
- 0.25 times
- 20 times
- 4 times Correct answer
Explanation: The weight is proportional to the inverse of the class share: 0.8 / 0.2 = 4.
2 Balance the class weights automatically.
SVC(class_weight=____)SVC(class_weight="balanced")Explanation: "balanced" computes the weights as n / (k · class_count).
3 Match the chain steps.
| Term | Meaning |
|---|---|
SimpleImputer(median) |
Fills missing values |
StandardScaler |
Scales |
SVC(class_weight) |
Classifies with balance |
scoring="f1" |
Selection criterion |