Day 60 of 133

Training pipelines & experiment tracking

Airflow vs Kubeflow vs Prefect; MLflow / W&B; reproducibility traps.

DSA · NeetCode 1-D DP

Climbing StairsDSA · 1-D DP
Interview questions to prep
1. Recognize this as Fibonacci. Walk through DP → space-optimized to O(1).
2. Generalize to 'k steps at a time' — how does the recurrence change?
Min Cost Climbing StairsDSA · 1-D DP
Interview questions to prep
1. State the DP: define the state, the transition, and the base case explicitly.
2. Top-down (memoized recursion) vs bottom-up (tabulation) — which is more natural here, and why?
3. Can you space-optimize from O(n) to O(1)? Show the rolling-window trick.

Pipeline orchestrators: Airflow, Prefect, Kubeflow, MetaflowMLOpsKubeflow
Interview questions to prep
1. Compare Airflow vs Prefect vs Kubeflow for ML pipelines.
2. When does an ML team outgrow Airflow?
Experiment tracking: MLflow, W&B, CometMLOpsW&B
Interview questions to prep
1. What metadata do you log for every training run?
2. How would you reconstruct a 6-month-old training run from your tracking metadata?
Reproducibility: seeds, environments, lockfilesMLOpsHF Trainer
Interview questions to prep
1. Why does setting a seed not guarantee reproducibility on GPUs?
2. What three artifacts MUST you pin to make an ML training run actually reproducible?

References & further reading