This function loads three networks: the full network, the initial portion up to the intermediate layer, and the final portion beginning with

Function to find error between Marabou solver and TF/Nnet at a given point Args: inputValues (list of np arrays): Input values to

* Return the initial (bottom) scope */

* Get the ContextMemoryManager for this Scope */

Get the constraint in the score tracker with the highest score */

Returns a list of lower bounds for the given layer If b=True, b variables Otherwise, f variables Args:

Find the node in the graph corresponding to the given name Args: nodeName (str): Name of node to find in graph Returns:

Returns a list of upper bounds for the given layer If b=True, b variables Otherwise, f variables Args:

Options that are determined at compile time */

Function representing identity Args: node (node): ONNX node representing identity operation :meta private:

* Inserts an element into the map. * The key inserted must be not be currently mapped. * This is implemented using d_insertMap.push_back().

* Return true iff this Scope is the current top Scope */

Load one of the acas networks. This network is larger than fc1.onnx, making it a better test case for testing timeout.

The test network fc1.onnx is used, which has two input variables and two output variables. The network was trained such that the first output

Get total number of variables in network Returns: (int)

* Inserts an element into the map, and pushes its key to the front * of the stack. The key inserted must be not be currently mapped. */

Registers a new variable, grows the BoundManager size and bound vectors, initializes new bounds to +/-inf, and returns the index of the new

Recompute the attributes of the network from basic arguments Args: weights (list of list of list of float): Outer index correspon