Medium Machine Learning Engineer questions
34 questions · 10 points each for a correct first answer.
- 1.Scaling before splittingdata-leakage, scikit-learnmedium
- 2.Validating a forecasting modelcross-validation, time-seriesmedium
- 3.Different results every inference runpytorch, deep-learningmedium
- 4.Handling class imbalanceimbalanced-data, model-evaluationmedium
- 5.Which regularisation gives sparse weightsregularisation, linear-modelsmedium
- 6.Early layers stop learningdeep-learning, optimisationmedium
- 7.Probabilities that cannot be trustedcalibration, model-evaluationmedium
- 8.Millions of product IDs as a featureembeddings, feature-engineeringmedium
- 9.F1 score from a confusion matrixclassification-metricsmedium
- 10.Log loss of a poor predictionloss-functions, classification-metricsmedium
- 11.Positive test for a rare conditionprobability, statisticsmedium
- 12.AUC from four scoresclassification-metricsmedium
- 13.Conv output size with stride 2cnnmedium
- 14.Parameters in a conv layercnnmedium
- 15.Reading a logistic regression weightlogistic-regressionmedium
- 16.Gradient of a sum of squarespytorch, autogradmedium
- 17.Training loop without zero_gradpytorch, trainingmedium
- 18.Softmax before CrossEntropyLosspytorch, loss-functionsmedium
- 19.Shape after a batched matmulpytorch, tensor-shapesmedium
- 20.Rows after a merge on duplicate keyspandas, data-cleaningmedium
- 21.Precision at 5 for a ranked listranking-metrics, recommender-systemsmedium
- 22.NDCG for a three-item rankingranking-metricsmedium
- 23.Effective batch size across GPUsdistributed-trainingmedium
- 24.Why use learning rate warmupoptimization, trainingmedium
- 25.Random search versus grid searchhyperparameter-tuningmedium
- 26.Initialisation for a deep ReLU networkinitialization, deep-learningmedium
- 27.Raising C in a soft-margin SVMsvmmedium
- 28.Loss turns NaN in an RNNrnn, trainingmedium
- 29.Scaling factor in attentiontransformers, attentionmedium
- 30.Trainable parameters in a LoRA adapterlora, fine-tuningmedium
- 31.Memory for 4-bit weightsquantization, llm-servingmedium
- 32.RAG misses exact part numbersrag, retrievalmedium
- 33.Target encoding without leakagefeature-engineering, data-leakagemedium
- 34.Several scans per patientcross-validation, data-leakagemedium