Evaluation Metrics

Implement classification evaluation metrics from scratch.

Functions to implement

1. accuracy(y_true, y_pred)

Compute accuracy (fraction correct).

• Input: Lists of true and predicted labels
• Output: Fraction of correct predictions

2. confusion_matrix(y_true, y_pred)

Compute the 2x2 confusion matrix for binary classification.

• Input: Lists of true (0/1) and predicted (0/1) labels
• Output: Dictionary with keys 'TP', 'TN', 'FP', 'FN'

3. precision(y_true, y_pred)

Compute precision = TP / (TP + FP).

• Input: Lists of true and predicted labels
• Output: Precision score

4. recall(y_true, y_pred)

Compute recall = TP / (TP + FN).

• Input: Lists of true and predicted labels
• Output: Recall score

5. f1_score(y_true, y_pred)

Compute F1 = 2 * (precision * recall) / (precision + recall).

• Input: Lists of true and predicted labels
• Output: F1 score

Examples

y_true = [1, 1, 0, 0, 1]
y_pred = [1, 0, 0, 0, 1]

accuracy(y_true, y_pred)         # 0.8
confusion_matrix(y_true, y_pred) # {'TP': 2, 'TN': 2, 'FP': 0, 'FN': 1}
precision(y_true, y_pred)        # 1.0 (no false positives)
recall(y_true, y_pred)           # 0.667 (missed one positive)
f1_score(y_true, y_pred)         # 0.8

Input domain and boundary policy

Inputs are equal-length, nonempty lists of binary labels (0 or 1), with 1 as the positive class. confusion_matrix returns the four-key dictionary, not a nested list. Precision returns 0.0 if TP+FP is zero; recall returns 0.0 if TP+FN is zero; F1 returns 0.0 if precision+recall is zero. These are explicit exercise conventions for undefined ratios. Do not mutate inputs. Threshold selection belongs outside these functions: they receive final labels.