Python problem track

ML metrics

Losses and scores a scientist or ML engineer must implement without sklearn: MSE, R², precision, recall, F1, BCE.

7 problems · ~78 min total · Data Scientist · ML Engineer · MLOps Engineer · AI Engineer

1. Step 1 · ~8 min · Google, OpenAI

   Mean squared error

   Implement mse(y_true, y_pred): the mean of squared residuals, rounded to 4 decimal places.

   PythonNumPyPyTorch

   mediumProNot started

2. Step 2 · ~6 min · Amazon, Google

   Mean absolute error

   Implement mae(y_true, y_pred): the mean of absolute residuals, rounded to 4 decimal places.

   PythonNumPyPyTorch

   mediumProNot started

3. Step 3 · ~12 min · Databricks, Google

   R-squared

   Return 1 - SS_res / SS_tot for y_true vs y_pred, rounded to 4 decimals. SS_tot uses the mean of y_true.

   PythonNumPyPyTorch

   hardProNot started

4. Step 4 · ~12 min · Meta, Google

   Precision

   Binary precision: TP / (TP + FP) for labels y and predictions pred in {0,1}. Round to 4 decimals. If there are no predicted positives, return 0.

   PythonNumPyPyTorch

   hardProNot started

5. Step 5 · ~12 min · Meta, Google

   Recall

   Binary recall: TP / (TP + FN) for labels y and predictions pred in {0,1}. Round to 4 decimals. If there are no actual positives, return 0.

   PythonNumPyPyTorch

   hardProNot started

6. Step 6 · ~14 min · OpenAI, Anthropic

   F1 score

   Binary F1: the harmonic mean of precision and recall for labels y and predictions pred in {0,1}. Round to 4 decimals. If precision + recall is 0, return 0.

   PythonNumPyPyTorch

   hardProNot started

7. Step 7 · ~14 min · OpenAI, Anthropic

   Binary cross-entropy

   Mean binary cross-entropy of labels y in {0,1} and probabilities p in (0,1), rounded to 4 decimals. Use natural log.

   PythonNumPyPyTorch

   hardProNot started

Other tracks