5.1.1 Machine Learning Basics Roadmap: Task, Data, Model, Score

Machine learning starts when you stop hand-writing every rule and let a model learn patterns from data. The first habit is not algorithm memorization. It is a small project loop.

Look at the Map First

Keep this loop:

define task→split data→fit model→predict→score→decide next step

Word	First meaning
feature	input column used by the model
label / target	answer the model learns to predict
train set	data used to learn
test set	data kept aside to check generalization
baseline	a simple first model used for comparison

Run the Smallest sklearn Loop

Create ml_first_loop.py and run it after installing scikit-learn.

from sklearn.datasets import load_iris
from sklearn.linear_model import LogisticRegression
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=0.2, random_state=42, stratify=y
)

model = LogisticRegression(max_iter=200)
model.fit(X_train, y_train)
predictions = model.predict(X_test)

print("task: classification")
print("test_accuracy:", round(model.score(X_test, y_test), 3))
print("prediction_count:", len(predictions))

Expected output:

task: classification
test_accuracy: 0.967
prediction_count: 30

This is the smallest useful machine learning loop: split first, train only on the training set, evaluate on the test set.

Learn in This Order

Order	Read	What to practice
1	5.1.2 What Is Machine Learning?	task types, features, labels
2	5.1.3 Scikit-learn Introduction	fit, predict, score
3	5.1.4 How Math Flows Into ML	vectors, probability, loss, optimization
4	5.1.5 Machine Learning History	why major algorithms appeared
5	5.1.6 sklearn and Matplotlib Workshop	run, plot, explain a baseline

Evidence to Keep

Keep this page’s proof of learning as a small evidence card:

Ml Problem

supervised, unsupervised, evaluation, or feature-engineering task

Baseline

simplest sklearn/modeling loop and fixed train/test split

Output

prediction, metric, chart, or model decision note

Failure Check

data leakage, unclear target, weak baseline, or metric mismatch

Expected Output

minimal ML loop with metric and one failure observation

Pass Check

You pass this roadmap when you can name the task type, identify X and y, explain why train/test split matters, and keep one baseline score as evidence.

Check reasoning and explanation

1. X is the feature matrix: rows are samples and columns are inputs the model can use. y is the label or target the model learns to predict.
2. Train/test split matters because the test set simulates new data. If the model sees test information during training, the score is no longer evidence of generalization.
3. A passing baseline record should include the task type, split settings, model, metric, and one sentence about what could still fail.