Machine Learning Workflow is used in development and design of Machine Learning projects. However Machine Learning Development Lifecycle (MLDLC) is used in delivery of such projects in production and is part of MLOps.
Notes:
You don’t need ML until you can prove that you need ML. some problems may seem complex but they may have simple solutions that doesn't require ML.
Always set a baseline. then try to beat it. some common baselines:
Average human
Simple linear model
Results of an existing model that works well on similar data, with no tuning.
Random prediction for binary classification, and highest frequency class for none-binary.
Design your evaluation methodology beforehand, including:
evaluation criteria
Stopping criteria: maximum processing time or iteration count
Workflow
Defining and formulating a problem: define the problem and Identify best method for generating solution. I.e. classification, regression, or clustering
Data Mining: Select a datasets or collect data. E.g. Web Scrapping, Surveys, etc
Establishing a baseline: A baseline is the simplest model that can solve your problem with minimal requirements and works as a reference point when comparing the actual model with the baseline.
Selecting and training a model by choosing AI Algorithms.
Considerations in selecting a model:
The scope of the problem: Some specific problems work best with specific models
The size of the dataset: Some models don’t work well with too small or large data
The level of interpretability: Some models are hard to interpret and explain. E.g. ANN
Training time: Training time on different models differs