I do mL for a living. Giving my two cents on things.
>>12416641At the level you're at, I would look at
fast.ai and Andrew Ng's courses. Both of those have you implement a lot of stuff in code. From there you can move to textbooks. Goodfellow book, Elements of Statistical Learning, Pattern Recognition and Machine Learning. You'll get more out of coding things up than reading books.
A lot of people suggest coding stuff from scratch, but I think that causes a lot of people to get lost in the weeds. My recommended approach is to start with a high-level library and to strip away the layers until you get to the from-scratch level.
>>12416759Depends on your system design, but from the sound of things you would have the output of the environment model be part of the input to the control model.
>>12416828For a chatbot you want a conversational language model. If you want to train one yourself, go with LSTMs and a standard chat dataset. You can do that on a single GPU. If you want something more performant, find a pre-trained transformer model that was fine-tuned from GPT-2 or something similar. You can probably get that from Huggingface.
>>12416813Tensorflow JS exists but I've heard it's a dumpster fire.