The learning will be continuous and you will probably be learning new things as you build out a project. This is very similar to applying computer science concepts to code. You can have a v1 codebase and after some time, you come back to it, see your mistakes and/or inefficiencies, and you make a v2 or a v1.1.
#Stuck in a loop of continuously learning
stable flame
IMO, you're "ready" for your own project (start small), when you understand the following:
- data
a) exploration
b) bias
c) augmentation
d) collection - models
a) classical ML models (SVM, linear regression, trees, KNN, etc)
b) simple neural networks (linear layer, RNN, CNN, transformer)
c) regularization and how to use it effectively
d) model training and how to detect overfitting/underfitting
This is the rough outline of general concepts you should definitely know before starting your project, as this will allow you to have a complete understanding of what to implement and why.
Your project itself should start small. Start with tabular datasets and then work up. Don't start with generative models. Doing that is the equivalent of trying to make GTA V as baby's first Unity game.
