ModelCreator | A program to create JSON models for blocks | Runtime Evironment library

You only need df as a parameter to modelCreator(). You then access your norm method via self.norm. Below is your code updated.

I worked out the process and now have an AutoFac module. I was able to registermodule by instance of the class and also pass in the options when I instantiate. Because EntityDbContext implements the three interfaces I was registering separately in the Simple Injector scenario, AutoFac has the convenience of being able to just infer them and register with AsImplementedInterfaces()

Two options come in my mind, depending what fit better in your project:

1) Use a placeholder for return type:

This code is a bit convoluted - but it is doing nothing les than what you told it to do.

And what it does, besides the needed descriptor (i.e.: the class containing the __get__ and __set__ methods) and usual metaclass mechanisms is plugging in a __new__ method to the class that is wrong in several ways.

For one, the new method that is assigned to the class __new__ ends its executing by using a hardcoded call to type. - that is the most wrong thing of all - as type returns a new class - not an instance. The call at the end of the plugged new method should be to object.__new__ - or better yet, using a mechanism that would call the __new__ in the next class on its __mro__ (but that would not that be trivial - as you will have to find that in the metaclass __new__ code surrounding the new method you are plugging in).

Anyway - it only would make sense to call type there if you would want classes that use this metaclass to be themselves "class factories" - that would return whole new classes not only with the declared fields, but with the sent-in defaults. And calling type is why you are seeing type(st) returning type- which is your first question.

Then, it is still wrong: the new class method, which is called at each instantation, sets the default attribute to the descritor (i.e. "field") - and that default will apply to every other instantiation of the same class - or other classes that inherit from it. You should set the default, if any, on the call to your StringField class - and on the method that will become the __new__ on the class, set the value for the instance.

That can be done if you first call the superclass __new__ to obtain an actual instance, and then loop though the passed in keyword arguments, and use setattr as the mechanism to set the attribute. Using setattr will ensure the StringField __set__ method is properly called.

So, there are many things weird in this code, but an attempt to fix it go by re-writting your metaclass __new__ to be more or less:

Your network is not converging. Try changing the parameters. The loss should reduce consistently. Also initialize the parameters properly.