Dynamic environment configurations for machine learning services

A system comprising:

• a plurality of processing devices corresponding to a simulation environment for a set of simulated, hosted customer networks;
• and
• a non-transitory computer readable medium storing instructions that, when executed by the at least a first processing device, cause the system to:
• receive, from a client device, a request to provide a generated trained reinforcement learning model on training data generated from an identified simulated, hosted customer network;
• responsive to the request from the client device, selecting a first simulated, hosted customer network from a set of simulated, hosted customer networks using the received request as selection criteria;
• identify, from a data store, an environment for generating the training data corresponding to a network simulation of an identified network and for training a reinforcement learning model based at least in part on the request;
• identify a reference reinforcement learning model for the training request;
• instantiate a simulation environment interface based at least in part on a parameter from an environment used to train the reference reinforcement learning model, wherein the environment interface communicates with the simulation environment;
• instantiate an agent in communication with the environment interface, the agent being configured using at least one parameter used to train the reference reinforcement learning model;
• activate the agent for a simulation period to form the trained reinforcement learning model, wherein each activation includes a state and a reward for a previous action taken by the agent;
• and
• provide to a client device the generated trained reinforcement learning model, of a simulated reinforcement learning model of the training data, in response to the request.

A computer-implemented method comprising:

• under control of a plurality of processing devices corresponding to a simulation environment for a set of simulated, hosted customer networks,
• identifying, from a data store, an environment for training a reinforcement learning model based at least in part on a training request;
• identifying a reference reinforcement learning model for generating the training data corresponding to a network simulation of an identified network and for the training request on the training data generated from an identified simulated, hosted customer network;
• responsive to the training request, selecting a first simulated, hosted customer network from a set of simulated, hosted customer networks using the training request as selection criteria;
• instantiating an environment interface based at least in part on a parameter from an environment used to train the reference reinforcement learning model, wherein the environment interface communicates with the environment;
• instantiating an agent in communication with the environment interface to generate a trained reinforcement learning model, the agent being configured using at least one parameter used to train the reference reinforcement learning model;
• activating the agent for a simulation period, wherein each activation includes a state and a reward for a previous action taken by the agent;
• and
• providing to a client device the generated trained reinforcement learning model, of a simulated reinforcement learning model of the training data, in response to the training request.