How to store, load and run model

[Samureimer]

Hi.

What is the indented way to store, load and run a model?
It seems trivial to save the model by calling ".save(path)" on the agent but loading the model again seems convoluted. Or maybe I just don't understand how to do it correctly.

I have to instantiate a options class for my agent with the same parameters used for the training.
I then have to instantiate a list of environments that I have no use for.
I can then instantiate the agent with the options and environments parameters.
I'm then able to load a model.

Running the model with GetActionsBatchAsync and isTraining=false then throws an exception:
System.AggregateException: 'One or more errors occurred. (mat1 and mat2 shapes cannot be multiplied (1x44 and 45x1024)

I am able to run the agent if I set isTraining=true. This I assume alters the model over time.

Isn't it possible to just load a model from path, feed it state data and get a action in return?

           var env = new List<IEnvironmentAsync<float[]>> {
                evalEnvironment
            };
            var evaluationAgent = new LocalDiscreteRolloutAgent<float[]>(learningSetup, env);
            await evaluationAgent.Load(path);

            for (int i = 0; i < 8596; i++)
            {
                float[] states = evalEnvironment.GetState();
                Guid guid = Guid.NewGuid();
                Dictionary<Guid, int[]> actions = await evaluationAgent.GetActionsBatchAsync(new List<(Guid, float[])>() { ( guid, states )}, true);
                evalEnvironment.Step(actions.FirstOrDefault().Value.FirstOrDefault());
            }