16474aee0c575e615c78d32f2a170c3d7f8f0082,autokeras/graph.py,Graph,produce_model,#Graph#,75

Before Change

    def produce_model(self):
        input_shape = get_int_tuple(self.model.inputs[0].shape[1:])

        input_tensor = Input(shape=tuple(input_shape))
        input_id = self.node_to_id[self.model.inputs[0]]
        output_id = self.node_to_id[self.model.outputs[0]]

        id_to_tensor = {input_id: input_tensor}
        q = Queue()
        q.put(input_id)
        while not q.empty():
            u = q.get()
            for v, edge_id in self.adj_list[u]:
                layer = self.edge_list[edge_id]
                if edge_id in self.old_edge_ids:
                    copied_layer = copy_layer(layer, get_int_tuple(id_to_tensor[u].shape))
                    temp_tensor = copied_layer(id_to_tensor[u])
                else:
                    temp_tensor = layer(id_to_tensor[u])
                id_to_tensor[v] = temp_tensor
                q.put(v)
        return Model(input_tensor, id_to_tensor[output_id])

After Change

        output_id = self.node_to_id[self.model.outputs[0]]

        id_to_tensor = {input_id: input_tensor}
        for v in self._topological_order():
            for u, edge_id in self.reverse_adj_list[v]:
                layer = self.edge_list[edge_id]

                if isinstance(layer, WeightedAdd):
                    edge_input_tensor = list(map(lambda x: id_to_tensor[x], self.edge_id_to_input_ids[edge_id]))
                else:
                    edge_input_tensor = id_to_tensor[u]

                if edge_id in self.old_edge_ids:
                    new_layer = copy_layer(layer)
                else:
                    new_layer = layer

                temp_tensor = new_layer(edge_input_tensor)
                id_to_tensor[v] = temp_tensor
        return Model(input_tensor, id_to_tensor[output_id])

    def to_conv_wider_model(self, pre_layer, n_add):
        output_id = self.node_to_id[pre_layer.output]

Instances