Post History

How to execute and find gradients of a tensorflow (1) graph object

I have an "tensorflow.python.framework.ops.Graph" object loaded from a .pb file.
```
def load_pb(path_to_pb):
    with tf.compat.v1.gfile.GFile(path_to_pb, "rb") as f:
        graph_def = tf.compat.v1.GraphDef()
        graph_def.ParseFromString(f.read())
    with tf.Graph().as_default() as graph:
        tf.import_graph_def(graph_def, name='')
        return graph

g = load_pb("some_file.pb")
```
I also know that tensorflow models can be executed or have the gradient found through this process
```
x_tensor = tf.convert_to_tensor(inputs2, dtype=tf.float32)
    with tf.GradientTape() as t:
        t.watch(x_tensor)
        output = new_model(x_tensor)
    gradients = t.gradient(output, x_tensor)
    score = float(output)
``` 
However, I am having a hard time getting the former tensorflow graph "g" to accept any input. It cannot be called as a function, or subscripted.

I *think* the model takes a vector of arbitrary length as input, but I don't know how to find out.

I want to:

1. Be able to input and output data from this graph

2. Be able to find the gradient of the output with respect to the input, if possible.