Metrics

from sklearn.metrics import (
    accuracy_score, precision_score, recall_score, f1_score,
    roc_auc_score, average_precision_score, log_loss, confusion_matrix,
    classification_report,
)
import numpy as np

# Classification
y_true = np.array([0, 0, 1, 1, 0, 1])
y_pred = np.array([0, 1, 1, 1, 0, 0])     # hard labels
y_prob = np.array([0.1, 0.6, 0.8, 0.7, 0.3, 0.45])   # probabilities

accuracy_score(y_true, y_pred)        # OK on balanced data
precision_score(y_true, y_pred)        # of predicted positives, how many right
recall_score(y_true, y_pred)           # of actual positives, how many caught
f1_score(y_true, y_pred)               # harmonic mean of precision + recall
log_loss(y_true, y_prob)               # penalises confident-wrong predictions
roc_auc_score(y_true, y_prob)          # ranking quality across thresholds
average_precision_score(y_true, y_prob) # better for imbalanced data than AUC

print(classification_report(y_true, y_pred))
print(confusion_matrix(y_true, y_pred))

# Regression
from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score
mean_squared_error(y_true, y_pred)         # MSE
mean_squared_error(y_true, y_pred, squared=False)   # RMSE
mean_absolute_error(y_true, y_pred)         # MAE — robust to outliers
r2_score(y_true, y_pred)

# Multiclass — pick averaging
f1_score(y_true_multi, y_pred_multi, average='macro')      # unweighted mean
f1_score(y_true_multi, y_pred_multi, average='weighted')   # weighted by support
f1_score(y_true_multi, y_pred_multi, average='micro')      # global TP/FP/FN

from sklearn.metrics import accuracy_score, f1_score, roc_auc_score
print(accuracy_score(yte, pred))
print(f1_score(yte, pred, average='macro'))
print(roc_auc_score(yte, prob))