I am building a model where firms have to set prices and make production decisions. Prices are continuous and so are the decision variables. (inventory, last sales, prices...).
What reinforcement learning method can I use that maps continuous to continuous ? Which python packages are there? If there are no python packages, I could write a wrapper.
PyBrain machine learning library is what you're looking for. It is quite complex implementation of neural networks, but when you understand it you will get a really powerful tool.
PyBrain is a python library for neural network modeling. Overview of the PyBrain should get you a basic idea: at each timestamp you provides neural network with a set of continuous values and take out another set of continuous values. But more important is that you can evaluate the output and train your neural network.
All these steps - including network training - are already implemented in PyBrain.