What are the sample_weight parameter in various performance metrics

Question

Can someone explain what is the sample_weight parameter for in various scikit-learn performance metrics such as:

• accuracy score (https://scikit-learn.org/stable/modules/generated/sklearn.metrics.accuracy_score.html)
• Jaccard_score (https://scikit-learn.org/stable/modules/generated/sklearn.metrics.jaccard_score.html#sklearn.metrics.jaccard_score)
• F1-score (https://scikit-learn.org/stable/modules/generated/sklearn.metrics.f1_score.html) etc.

Many thanks.

Answer 1

It simply indicates how much each smaple affects the metric. Look at this example:

With below y_true and y_pred the accuracy_score would be 0.6666666:

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

accuracy_score(y_true, y_pred) # 0.6666666666666666

But if second sample is more important to us than other two, we can enforce it's significance with sample_weight:

accuracy_score(y_true, y_pred, sample_weight=[1, 2, 1]) # 0.5