Important Exam Clarification
💡 SageMaker includes built-in algorithms for traditional and common ML scenarios and business solutions. While Frank Kane goes into quite a bit of depth on each algorithm, this MLA-C01 certification only expects you to know these algorithms at a high-level.
- Make sure you understand what each algorithm does, memorize its use cases, and know whether it's powered by NNs and DL (which means
batch_size, learning_rate, and num_epochs are important hyperparameters) or if it's a simpler, more traditional ML algorithm that doesn't leverage NNs and DL.
- Don't worry about memorizing every detail on training data, hyperparameters, and appropriate training and inference instance types. Knowing those more in depth is expected in the Machine Learning Specialty certification (MLS-C01), but not in the MLA-C01 certification.
Training and Inference in Supervised & Unsupervised Algorithms
💡 ”Training” is quite different between supervised and unsupervised algorithms:
- Supervised algorithms train on labeled data by “learning” what the expected output should be from a given input. Once training is complete, they will infer predictions on new data.
- Unsupervised algorithms don't “train” the model before inferencing in the same way. Instead they attempt to discern inherent patterns in the dataset they ingest. For new datapoints they will evaluate/infer them based on the learned patterns.
- Training vs Inference:
- Training in supervised ML involves mapping inputs to labels, while in unsupervised ML, it’s about uncovering hidden patterns.
- Inference in supervised ML applies the learned mapping to predict labels, whereas in unsupervised ML, it applies the learned structure to process new data.
- ‼️ In both cases, training updates the model’s parameters, while inference does NOT!
Detailed pages
Supervised Algorithms in SageMaker
Unsupervised Algorithms in SageMaker