Non Sequential API Attempt

tensorlayer_api_test.py

import inspect
import time
import wrapt

import pprint

def get_caller(skip=2):
    try:
        stack = inspect.stack()

        if len(stack) < skip + 1:
            raise ValueError("The length of the inspection stack is shorter than the requested start position.")
        
        for currstack in stack[skip:]:
            
            try:
                args, _, _, values = inspect.getargvalues(currstack[0])

                if 'self' in values.keys() and isinstance(values['self'], Network):
                    return values['self']
                
            except:
                continue
            
        return None

    except:
        return None
    
@wrapt.decorator
def pass_through(wrapped, instance, args, kwargs):
    
    cls = wrapped(*args, **kwargs)        
    
    try:
        caller = get_caller()
        if isinstance(caller, Network):
            caller.add_layer(instance)
    except:
        pass
    
    return cls
            
class My_obj(object):
    
    @pass_through
    def __init__(self, name):
        self.name = name
        print("[TL] Layer:", name)

print('no context')
obj = My_obj("thomas")

print("\nwith context")        
class Network(object):
    def __init__(self):
        
        self.all_layer = dict()
        
        def define_add_layers():
            def relu_layer():
                obj = My_obj(name="ReLU")
            obj = My_obj(name="Dropout")
            obj = My_obj(name="LSTM")
            relu_layer()
            
            
        obj = My_obj(name="Dense")
        obj = My_obj(name="Conv2d")
        obj = My_obj(name="BatchNorm")
        
        define_add_layers()
        
    def add_layer(self, layer):
        self.all_layer[layer.name] = layer
        
net = Network()

pprint.pprint(net.all_layer)