Training the Model
- The model learns by updating weights (parameters) iteratively to reduce error.
- This continues until the error is minimal or a set number of iterations is reached.
Running Experiments
- Multiple algorithms and settings are tested to find the best model.
- Experiments help you find the most effective solution.
Hyperparameters
- These are the settings that affect model performance (e.g., number of layers in a deep learning model).
- Hyperparameters are fine-tuned through experimentation to improve accuracy.
Using SageMaker for Training
- A training job is created in SageMaker where you:
- Specify the S3 bucket with training data.
- Select the compute resources for training.
- Choose the training algorithm and set its hyperparameters.
- The model is trained on SageMaker’s compute instances, and results are saved in S3.
Iterative Process
- Model training is an ongoing process of trying different data, algorithms, and settings to optimize performance.
- Thousands of training runs may be necessary to find the best solution.
SageMaker Experiments
- A tool to manage and analyze machine learning experiments.
- Helps track multiple training runs, compare results, and identify the best-performing models.
Automatic Model Tuning (AMT)
- Also known as hyperparameter tuning, AMT automates the process of selecting the best hyperparameters by running multiple training jobs.
- It stops once the improvement in the model’s performance plateaus.
This iterative cycle of training, tuning, and evaluating ensures the best possible model is achieved.
