• Contents:

Good data

• To train a ML model we must have good training data
  • 💡 Garbage in → Garbage out
  • ‼️ Most critical stage to build a good model!
  • What is good data? Depends on scenario and requirements!
• Several options to model our data, which will impact the types of algorithms we can use to train our models
  1. Labeled vs. Unlabeled Data
  2. Structured vs. Unstructured Data

Labeled vs Unlabeled Data

Labeled Data

• Data includes both input objects and corresponding output labels
  • Use case: Supervised Learning → map inputs to known outputs
• Example

Unlabeled Data

• Data includes only input objects (without any output labels)
  • Use case: Unsupervised Learning → model tries to find patterns or structures in data
• Example
• 💡 Models that are trained with labeled data are more precise than those trained with unlabeled data… so why use unlabeled data? Because often it can be a pain to label loads of unlabeled data, so it's more efficient to use unsupervised learning!

Structured vs Unstructured Data

Structured Data

• Data has structure (like in a DB)
• Examples
  • Tabular Data: tables with rows and columns
  • Time Series Data: sequential datapoints in time