The project has not been finished ...yet.
| Supervised | Unsupervised | Reinforcement | |
|---|---|---|---|
| Use Case | Supervised Learning can be used to classify into defined categories by training on data with x and y values present | Unsupervised Learning teaches itself to find categories in data with only x-values | Reinforcement Learning improves at finding y-values based on x-values |
| Requirements | A set of data with x and y values | A set of data with only x-values | A reward function for the neural network to measure its own progress |
| "The catch" | A complete set of data is needed | The neural network might find unexpected patterns and give unexpected y-values | Requires a precise reward function, Difficult (for me) |
- Data preprocessing is important
- In reinforcement, knowing when it works is harder than getting it to work
- Statistics are crucial
- Never use an old version
- Getting started
- Knowing when it works, especially in the Super Mario project
- Identifying why it doesn't work (Reinforcement)
- Python and PyCharm
- Udemy course "A Complete Guide on TensorFlow 2.0 using Keras API"