Interview Flashcards 🃏

Global Interpreter Lock — mutex allowing only one thread to execute Python bytecode at a time. Use multiprocessing for CPU-bound parallelism.

Click to see question →

Bias = underfitting (too simple). Variance = overfitting (too complex). Total Error = Bias² + Variance + Noise.

Click to see question →

L1 (Lasso): Sparse weights, feature selection. L2 (Ridge): Small weights, keeps all features. ElasticNet = both.

Click to see question →

INNER: Only matching rows from both. LEFT: All from left table + matching from right (NULL if no match).

Click to see question →

Architecture using self-attention (not recurrence) for parallel processing. Key: Multi-Head Attention + Positional Encoding + FFN + Layer Norm.

Click to see question →

Atomicity (all or nothing), Consistency (valid state), Isolation (concurrent txn independence), Durability (committed = permanent).

Click to see question →

Retrieval-Augmented Generation: Retrieve relevant docs from vector DB → inject as context into LLM prompt → generate grounded answer.

Click to see question →

Precision = TP/(TP+FP) "how accurate are positive predictions". Recall = TP/(TP+FN) "how many positives did we find".

Click to see question →