A Crash Course in Scikit-learn: Your First Steps in Machine Learning

1
from sklearn.datasets import load_iris
2
 
3
iris = load_iris()
4
X, y = iris.data, iris.target

1
from sklearn.model_selection import train_test_split
2
 
3
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=42)

1
from sklearn.neighbors import KNeighborsClassifier
2
 
3
# Create an instance of the model
4
knn = KNeighborsClassifier(n_neighbors=3)
5
 
6
# Train the model
7
knn.fit(X_train, y_train)

1
y_pred = knn.predict(X_test)

1
from sklearn.metrics import accuracy_score
2
 
3
accuracy = accuracy_score(y_test, y_pred)
4
print(f"Model Accuracy: {accuracy:.2f}")
5
# Expected Output: Model Accuracy: 1.00

1
import pandas as pd
2
import numpy as np
3
from sklearn.model_selection import train_test_split, cross_val_score
4
from sklearn.feature_extraction.text import TfidfVectorizer
5
from sklearn.naive_bayes import MultinomialNB
6
from sklearn.svm import LinearSVC
7
from sklearn.ensemble import RandomForestClassifier
8
from sklearn.pipeline import Pipeline
9
from sklearn.metrics import classification_report, confusion_matrix
10
import matplotlib.pyplot as plt
11
import seaborn as sns
12
 
13
# Create a more realistic dataset
14
messages = [
15
    # Ham messages
16
    "Hey, are you free for lunch tomorrow?",
17
    "I'll be home late tonight, traffic is terrible",
18
    "Can you pick up milk on your way home?",
19
    "Meeting rescheduled to 3pm. See you there!",
20
    "Thanks for the birthday wishes!",
21
    "The project deadline has been extended to Friday",
22
    "Great presentation today!",
23
    "Let me know when you arrive",
24
    "Happy anniversary! Love you",
25
    "Don't forget the team lunch at noon",
26
    "Your package was delivered",
27
    "Reminder: Doctor appointment at 2pm",
28
    "Nice seeing you yesterday!",
29
    "Can we talk later?",
30
    "Running 10 minutes late",
31
    
32
    # Spam messages
33
    "WINNER!! You've won £1000000! This is definitely real! -Nigerian Prince",
34
    "Hot singles in your area want to discuss your car's extended warranty",
35
    "Doctors HATE this one weird trick! (It's called eating vegetables)",
36
    "You've inherited $50M from a long-lost uncle you never knew existed!!!",
37
    "URGENT: Your computer has virus. Click here to download more virus",
38
    "Congratulations! You're our 1,000,000th visitor! So is everyone else!",
39
    "Make $$$ working from home! Requires: time machine, unicorn, PhD in wizardry",
40
    "Your package is waiting! We don't know what package. Just click the link",
41
    "Free iPhone 15! Just pay shipping of $999.99",
42
    "Your bank account will be suspended! -Definitely Your Real Bank (trust us)",
43
    "Bitcoin opportunity! Turn $1 into $1M! Math professors HATE this!",
44
    "Free cruise to the Bahamas! Departs from Nebraska. Seems legit",
45
    "IRS here. We only accept iTunes gift cards for tax payment now",
46
    "Your crush wants to meet you! They're definitely real and not a bot",
47
    "Lose 50 pounds in 2 days with this miracle pill! Side effects include levitation"
48
]
49
 
50
labels = ['ham'] * 15 + ['spam'] * 15
51
 
52
# Create DataFrame
53
df = pd.DataFrame({'message': messages, 'label': labels})
54
 
55
# Step 1: Create a pipeline with preprocessing and model
56
pipeline = Pipeline([
57
    ('tfidf', TfidfVectorizer(stop_words='english', max_features=1000)),
58
    ('classifier', MultinomialNB())
59
])
60
 
61
# Step 2: Split data
62
X_train, X_test, y_train, y_test = train_test_split(
63
    df['message'], df['label'], test_size=0.3, stratify=df['label'], random_state=42
64
)
65
 
66
# Step 3: Train the model
67
pipeline.fit(X_train, y_train)
68
 
69
# Step 4: Make predictions
70
y_pred = pipeline.predict(X_test)
71
 
72
# Step 5: Evaluate performance
73
print("Classification Report:")
74
print(classification_report(y_test, y_pred))
75
 
76
# Step 6: Visualize confusion matrix
77
cm = confusion_matrix(y_test, y_pred, labels=['ham', 'spam'])
78
plt.figure(figsize=(8, 6))
79
sns.heatmap(cm, annot=True, fmt='d', cmap='Blues', 
80
            xticklabels=['ham', 'spam'], yticklabels=['ham', 'spam'])
81
plt.title('Confusion Matrix')
82
plt.ylabel('True Label')
83
plt.xlabel('Predicted Label')
84
plt.tight_layout()
85
plt.show()
86
 
87
# Step 7: Compare different models
88
models = {
89
    'Naive Bayes': MultinomialNB(),
90
    'SVM': LinearSVC(random_state=42),
91
    'Random Forest': RandomForestClassifier(n_estimators=100, random_state=42)
92
}
93
 
94
for name, model in models.items():
95
    pipeline = Pipeline([
96
        ('tfidf', TfidfVectorizer(stop_words='english', max_features=1000)),
97
        ('classifier', model)
98
    ])
99
    scores = cross_val_score(pipeline, df['message'], df['label'], cv=5, scoring='f1_macro')
100
    print(f"{name}: {scores.mean():.3f} (+/- {scores.std() * 2:.3f})")
101
 
102
# Step 8: Feature importance - What words indicate spam?
103
pipeline.fit(df['message'], df['label'])
104
feature_names = pipeline.named_steps['tfidf'].get_feature_names_out()
105
coefficients = pipeline.named_steps['classifier'].feature_log_prob_[1] - \
106
               pipeline.named_steps['classifier'].feature_log_prob_[0]
107
 
108
# Get top spam indicators
109
top_spam_idx = coefficients.argsort()[-10:][::-1]
110
top_spam_words = [(feature_names[i], coefficients[i]) for i in top_spam_idx]
111
 
112
print("\nTop spam indicators:")
113
for word, score in top_spam_words:
114
    print(f"  {word}: {score:.3f}")
115
 
116
# Step 9: Test on new messages
117
new_messages = [
118
    "Meeting at 3pm in conference room",
119
    "URGENT!!! Your goldfish has won the lottery!!!",
120
    "Can you send me the report?", 
121
    "Hello, I am a normal human. Please click this legitimate link: totallynotascam.virus"
122
]
123
 
124
predictions = pipeline.predict(new_messages)
125
for msg, pred in zip(new_messages, predictions):
126
    print(f"\nMessage: '{msg[:50]}...'\nPrediction: {pred}")

1
from sklearn.cluster import KMeans
2
 
3
kmeans = KMeans(n_clusters=3, random_state=42, n_init=10)
4
kmeans.fit(X)
5
print(kmeans.labels_)

1
from sklearn.pipeline import Pipeline
2
from sklearn.preprocessing import StandardScaler
3
from sklearn.svm import SVC
4
 
5
# Create a pipeline
6
pipe = Pipeline([
7
    ('scaler', StandardScaler()),  # Step 1: Scale the data
8
    ('svm', SVC())                 # Step 2: Apply the classifier
9
])
10
 
11
# The pipeline can be used like any other estimator
12
pipe.fit(X_train, y_train)
13
print(f"Pipeline Accuracy: {pipe.score(X_test, y_test):.2f}")

1
# ❌ Bad: My model is a gold digger
2
X_train = [[25, 50000], [30, 60000], [35, 70000]]  # age, salary
3
 
4
# ✅ Good: Teaching your model that all features matter
5
from sklearn.preprocessing import StandardScaler
6
scaler = StandardScaler()
7
X_train_scaled = scaler.fit_transform(X_train)
8
X_test_scaled = scaler.transform(X_test)  # Use transform, not fit_transform!

1
# ❌ Bad: "I'll just scale everything together, what could go wrong?"
2
X_scaled = StandardScaler().fit_transform(X)
3
X_train, X_test, y_train, y_test = train_test_split(X_scaled, y)
4
 
5
# ✅ Good: Keep your test data pure and innocent
6
X_train, X_test, y_train, y_test = train_test_split(X, y)
7
scaler = StandardScaler()
8
X_train_scaled = scaler.fit_transform(X_train)
9
X_test_scaled = scaler.transform(X_test)  # No peeking at test data stats!

1
from collections import Counter
2
print(Counter(y_train))  # {0: 9900, 1: 100} - Uh oh
3
 
4
# Your "amazing" model:
5
print("Accuracy: 99%!")  # Just predicts the majority class
6
print("Fraud caught: 0%")  # Completely useless
7
 
8
# Fix it with stratified splitting and better metrics:
9
X_train, X_test, y_train, y_test = train_test_split(
10
    X, y, test_size=0.3, stratify=y, random_state=42
11
)
12
 
13
# Look at metrics that actually matter:
14
from sklearn.metrics import classification_report
15
print(classification_report(y_test, y_pred))  # Shows the ugly truth

1
# ❌ Bad: The "It'll probably work" approach
2
model = RandomForestClassifier()  # n_estimators=100, max_depth=None, etc.
3
 
4
# ✅ Good: Actually finding what works
5
from sklearn.model_selection import GridSearchCV
6
 
7
param_grid = {
8
    'n_neighbors': [3, 5, 7, 9],
9
    'weights': ['uniform', 'distance']
10
}
11
 
12
grid_search = GridSearchCV(KNeighborsClassifier(), param_grid, cv=5)
13
grid_search.fit(X_train, y_train)  # Warning: Your laptop fan will sound like a jet engine
14
 
15
print(f"Best parameters: {grid_search.best_params_}")
16
print(f"Time spent: 3 hours")
17
print(f"Improvement: 0.02%")  # But hey, it's optimized!

1
# The error that makes you question your dataset
2
model.fit(X_train, y_train)  # ValueError: Input contains NaN
3
 
4
# Check the damage:
5
print(f"Missing values: {X.isna().sum().sum()}")  # 47... oops
6
 
7
# The "good enough" fix:
8
from sklearn.impute import SimpleImputer
9
imputer = SimpleImputer(strategy='mean')  # When in doubt, use the average!
10
X_imputed = imputer.fit_transform(X)  # Now with 100% made-up values!