Problem Framing - Google Developers

This Google developers course is part of the foundation series of short courses on the subject of machine learning. Quick summary of this problem framing course:

• Problem Framing - Determine if machine learning is the right approach followed by framing the problem in machine learning language.
• Understand the problem - Can machine learning, predictive or generative be used or is a non machine learning solution required?
  • Predictive ML - Uses large amount of training data to predict outputs. Outputs can be checked against reality. Use supervised, unsupervised or reinforcement learning models.
  • Generative AI - Generates output based on the user's input. Large language model or image generator is trained with lots of unlabelled data.
  • Is ML the right approach? Perhaps you have some simple non-ML solution? That could be used to compare against a possible ML approach and decide if the ML solution warrants the cost and extra complications.
  • Must have good data! - You need abundant data, trusted and collected in similar ways and with features that are predictive
• Framing in ML terms
  • Define the ideal outcome and goal.
    • What is the output and type of output?
      • Predictive ML for classification or numerical prediction.
        • Classification Model - Output which class the input is in.
        • Regression Model - Output a numerical prediction.
      • Generative AI for generative new content.
• Define success metrics.
• Implementing a Model
  • There are pre-trained models you can use, such as at TensorFlow and Kaggle. Have to be careful that you have the exact data these models need.