I need a program written in python3 (with Scikit/Keras/Tensorflow) that can take up to 10 binary inputs, and a decimal number. I want it to take data and learn from it unsupervised and record a reward for a win, and a negative-reward for a loss. The decimal number will likely be $5 - $2k and is related to the binary inputs somehow.
The algo should treat it no different than a game. The agent should have 3 actions "buy", "sell", "hold". The agent can buy as much as it can (a start balance should be configurable). A buy will buy, and reduce the balance, and increase inventory by 1, a sell should increase the balance and reduce the inventory, and a hold does nothing. The program should never buy if the balance cannot afford the price, and should also never sell if the inventory count is less than or equal to 0.
It should be able to get better at guessing price movements over time, with more and more price and binary data.
The reward will be the difference from the buy and the sell. If it loses money, a negative reward, and if positive, a positive reward.
Model should be saved as it trains so it never should restart from the bottom.
The data will be in CSV file and look like this:
14 pekerja bebas membida secara purata $193 untuk pekerjaan ini
HI there, We are experts in the python related frameworks solution. We are expert developers who have done many projects and have wide experience in python programming language.
Hi, I am an expert in Reinforcement learning. I am very familiar with Deep Q-Learing. I can do your project perfectly. Waiting for your response. Thanks.
Hi. I am an expert in machine learning. I have many experiences in such works. I am sure I can do it. We can discuss details via chat. I wait for you now. Thanks.