Path3.Mod1.a - Automated Machine Learning - What is it?

Question 1

AutomatedML defined

Answer 1

The process of automating the time-consuming, iterative tasks of machine learning model development. It allows us to try different model training algorithms with a variety of hyperparameters in parallel in order to find the one that best fits your data

See What is AutoML

Question 2

NoCo Save BP Ag OF

AutomatedML Advantages

Answer 2

• Implementing ML solutions without extensive programming knowledge
• Saves time and resources
• Leverage Data Science Best Practices
• Provides Agile problem-solving
• Provides charts and metrics to help identify risks related to overfitting and helps to mitigate them

Question 3

Id CoNoCo DS CMLP Job Rev

Six steps for designing and running an AutoML experiment

Answer 3

1. Identify the type of ML Problem: Classification, Time-Series Forecasting, Regression, Computer Vision or NLP?
2. Choose Code-First Experience (SDKs) or No-Code Studio Web Experience (ML Studio)
3. Specify the source of your labeled training data
4. Configure ML parameters; # of iterations, what models to try, hyperparameter tuning, advanced preprocessing/featurization, and selecting metrics
5. Submit the training job
6. Review the results in the logged job info (metrics)

Question 4

C R TSF CV NLP

The five ML Workloads supported by AutoML
(you know this from AI-900)

Answer 4

• Classification: Predict which categories new data will fall into, based on learnings from its training data.
• Regression: Predict numeric values based on independent predictors, in order to establish a relationship between those predictors, by estimating how each variable impacts others.
• Time-Series Forecasting: A multivariant regression problem where past values are pivoted to become additional dimensions for the regressor with other predictors. It naturally incorporates multiple contextual variables and their relationship to one another.
• Computer Vision: Generate models trained on image data for Vision-related scenarios
• Natural Language Processing: Generate models trained on text for classification and named entity

Question 5

Two options for Code-First AzureM ML experiences

Answer 5

1. Azure ML Python SDKv2
2. Azure ML CLIv2 (REST APIs)

Question 6

MCC MLC OD IS

Four AutoML-supported Vision tasks

Answer 6

• Multi-class image Classification: Images are classified with a single label from a set of possible class/categorization values
• Multi-label image Classification: Images are classified with one or more labels from a set of possible class/categorization values
• Object Detection: Objects are identified and located using bounding boxes within an image
• Instance Segmentation: Identify object at the pixel level, drawing polygons around each object in the image

Question 7

MCTC MLTC NER

Three AutoML-supported NLP tasks and what file types work best with each

Answer 7

• Multi-Class Text Classification: multiple categories, each data sample can only be classified as exactly ONE class (find the best match). .CSV works for this task
• Multi-Label Text Classification: multiple categories, each data sample can be classsifeid as MORE THAN ONE class (find ALL that apply). .CSV works for this task
• Named Entity Recognition: multiple possible tags for tokens in sequence. The task at hand is to predict tags for tokens for each sequence. Requires .TXT files that use a space as the separator and adhere to the CoNLL format (Conference on Natural Language Learning)