[ { "function_name": "__init__", "file_path": "graphmetro/moe_model.py", "function_calls": [ { "function_call_name": "super().__init__", "functionality": "Initializes the base nn.Module class.", "input": "()", "output": "None", "location_or_namespace": "torch.nn.Module" }, { "function_call_name": "len", "functionality": "Returns the number of items in a dictionary.", "input": "trans_func_dict", "output": "int", "location_or_namespace": "builtins" }, { "function_call_name": "list", "functionality": "Converts dictionary values to a list.", "input": "trans_func_dict.values()", "output": "list", "location_or_namespace": "builtins" }, { "function_call_name": "list", "functionality": "Converts dictionary keys to a list.", "input": "trans_func_dict.keys()", "output": "list", "location_or_namespace": "builtins" }, { "function_call_name": "range", "functionality": "Generates a sequence of numbers.", "input": "self.num_experts", "output": "range", "location_or_namespace": "builtins" } ] }, { "function_name": "_aggregate", "file_path": "graphmetro/moe_model.py", "function_calls": [ { "function_call_name": "self.experts[i].encoder", "functionality": "Encodes the graph data for a specific expert.", "input": "data.x, data.edge_index, data.edge_attr, data.batch", "output": "torch.Tensor", "location_or_namespace": "" }, { "function_call_name": "unsqueeze", "functionality": "Adds a dimension to a tensor.", "input": "dim=-1", "output": "torch.Tensor", "location_or_namespace": "torch.Tensor" }, { "function_call_name": "append", "functionality": "Adds an element to a list.", "input": "emb", "output": "None", "location_or_namespace": "builtins.list" }, { "function_call_name": "torch.concat", "functionality": "Concatenates a sequence of tensors along a given dimension.", "input": "embs, dim=-1", "output": "torch.Tensor", "location_or_namespace": "torch" }, { "function_call_name": "size", "functionality": "Returns the size of the tensor.", "input": "0", "output": "torch.Size", "location_or_namespace": "torch.Tensor" }, { "function_call_name": "torch.unique", "functionality": "Returns the unique elements of a tensor.", "input": "data.batch", "output": "torch.Tensor", "location_or_namespace": "torch" }, { "function_call_name": "len", "functionality": "Returns the length of an object.", "input": "torch.unique(data.batch)", "output": "int", "location_or_namespace": "builtins" }, { "function_call_name": "isinstance", "functionality": "Checks if an object is an instance of a class.", "input": "weights_or_labels, (int, np.int64)", "output": "bool", "location_or_namespace": "builtins" }, { "function_call_name": "torch.LongTensor", "functionality": "Creates a LongTensor.", "input": "[weights_or_labels for _ in range(length)]", "output": "torch.Tensor", "location_or_namespace": "torch" }, { "function_call_name": "len", "functionality": "Returns the length of an object.", "input": "weights_or_labels.size()", "output": "int", "location_or_namespace": "builtins" }, { "function_call_name": "copy.deepcopy", "functionality": "Creates a deep copy of an object.", "input": "weights_or_labels", "output": "torch.Tensor", "location_or_namespace": "copy" }, { "function_call_name": "torch.zeros", "functionality": "Creates a tensor filled with zeros.", "input": "(length, self.num_experts)", "output": "torch.Tensor", "location_or_namespace": "torch" }, { "function_call_name": "to", "functionality": "Moves the tensor to a specified device.", "input": "cfg.device", "output": "torch.Tensor", "location_or_namespace": "torch.Tensor" }, { "function_call_name": "range", "functionality": "Generates a sequence of numbers.", "input": "length", "output": "range", "location_or_namespace": "builtins" }, { "function_call_name": "argmax", "functionality": "Returns the indices of the maximum values along a dimension.", "input": "dim=1", "output": "torch.Tensor", "location_or_namespace": "torch.Tensor" }, { "function_call_name": "bool", "functionality": "Casts the tensor to boolean type.", "input": "()", "output": "torch.Tensor", "location_or_namespace": "torch.Tensor" }, { "function_call_name": "torch.zeros_like", "functionality": "Creates a tensor filled with zeros, with the same size as input.", "input": "weights_or_labels", "output": "torch.Tensor", "location_or_namespace": "torch" }, { "function_call_name": "unsqueeze", "functionality": "Adds a dimension to a tensor.", "input": "dim=2", "output": "torch.Tensor", "location_or_namespace": "torch.Tensor" }, { "function_call_name": "torch.bmm", "functionality": "Performs batch matrix-matrix product.", "input": "embs, weights", "output": "torch.Tensor", "location_or_namespace": "torch" }, { "function_call_name": "squeeze", "functionality": "Removes a dimension from a tensor.", "input": "dim=-1", "output": "torch.Tensor", "location_or_namespace": "torch.Tensor" } ] }, { "function_name": "get_expert_weights", "file_path": "graphmetro/moe_model.py", "function_calls": [ { "function_call_name": "self.gating_model", "functionality": "Computes expert weights using the gating model.", "input": "trans_data.x, trans_data.edge_index, trans_data.edge_attr, trans_data.batch", "output": "torch.Tensor", "location_or_namespace": "" }, { "function_call_name": "softmax", "functionality": "Applies the softmax function.", "input": "dim=-1", "output": "torch.Tensor", "location_or_namespace": "torch.Tensor" } ] }, { "function_name": "forward", "file_path": "graphmetro/moe_model.py", "function_calls": [ { "function_call_name": "self.get_expert_weights", "functionality": "Computes expert weights using the gating model.", "input": "trans_data, softmax=True", "output": "torch.Tensor", "location_or_namespace": "graphmetro/moe_model.py" }, { "function_call_name": "argmax", "functionality": "Returns the indices of the maximum values along a dimension.", "input": "dim=-1", "output": "torch.Tensor", "location_or_namespace": "torch.Tensor" }, { "function_call_name": "view", "functionality": "Returns a new tensor with the same data but different size.", "input": "-1", "output": "torch.Tensor", "location_or_namespace": "torch.Tensor" }, { "function_call_name": "self._aggregate", "functionality": "Aggregates expert outputs based on weights.", "input": "trans_data, expert_weights, hard", "output": "torch.Tensor", "location_or_namespace": "graphmetro/moe_model.py" }, { "function_call_name": "self.classifier", "functionality": "Classifies the aggregated embedding.", "input": "new_z", "output": "torch.Tensor", "location_or_namespace": "" }, { "function_call_name": "torch.zeros_like", "functionality": "Creates a tensor filled with zeros, with the same size as input.", "input": "expert_weights", "output": "torch.Tensor", "location_or_namespace": "torch" }, { "function_call_name": "range", "functionality": "Generates a sequence of numbers.", "input": "expert_weights.size(0)", "output": "range", "location_or_namespace": "builtins" } ] }, { "function_name": "losses", "file_path": "graphmetro/moe_model.py", "function_calls": [ { "function_call_name": "self.get_expert_weights", "functionality": "Computes expert weights using the gating model.", "input": "trans_data, softmax=False", "output": "torch.Tensor", "location_or_namespace": "graphmetro/moe_model.py" }, { "function_call_name": "repeat", "functionality": "Repeats the tensor along specified dimensions.", "input": "expert_weights.size(0), 1", "output": "torch.Tensor", "location_or_namespace": "torch.Tensor" }, { "function_call_name": "to", "functionality": "Moves the tensor to a specified device.", "input": "cfg.device", "output": "torch.Tensor", "location_or_namespace": "torch.Tensor" }, { "function_call_name": "print", "functionality": "Prints objects to the text stream.", "input": "'Added pos_weight=5'", "output": "None", "location_or_namespace": "builtins" }, { "function_call_name": "F.binary_cross_entropy_with_logits", "functionality": "Computes Binary Cross Entropy loss with logits.", "input": "expert_weights, target, pos_weight=torch.Tensor([4.]).to(expert_weights.device)", "output": "torch.Tensor", "location_or_namespace": "torch.nn.functional" }, { "function_call_name": "torch.Tensor", "functionality": "Creates a Tensor.", "input": "[4.]", "output": "torch.Tensor", "location_or_namespace": "torch" }, { "function_call_name": "sigmoid", "functionality": "Applies the sigmoid function.", "input": "()", "output": "torch.Tensor", "location_or_namespace": "torch.Tensor" }, { "function_call_name": "float", "functionality": "Casts the tensor to float type.", "input": "()", "output": "torch.Tensor", "location_or_namespace": "torch.Tensor" }, { "function_call_name": "mean", "functionality": "Computes the mean of the tensor.", "input": "()", "output": "torch.Tensor", "location_or_namespace": "torch.Tensor" }, { "function_call_name": "detach", "functionality": "Detaches the tensor from the computation graph.", "input": "()", "output": "torch.Tensor", "location_or_namespace": "torch.Tensor" }, { "function_call_name": "softmax", "functionality": "Applies the softmax function.", "input": "dim=-1", "output": "torch.Tensor", "location_or_namespace": "torch.Tensor" }, { "function_call_name": "self._aggregate", "functionality": "Aggregates expert outputs based on weights.", "input": "trans_data, expert_weights_detached, hard", "output": "torch.Tensor", "location_or_namespace": "graphmetro/moe_model.py" }, { "function_call_name": "torch.unique", "functionality": "Returns the unique elements of a tensor.", "input": "data.batch", "output": "torch.Tensor", "location_or_namespace": "torch" }, { "function_call_name": "len", "functionality": "Returns the length of an object.", "input": "torch.unique(data.batch)", "output": "int", "location_or_namespace": "builtins" }, { "function_call_name": "torch.linalg.norm", "functionality": "Computes the matrix or vector norm.", "input": "new_zs - gt_zs_base, 'fro'", "output": "torch.Tensor", "location_or_namespace": "torch.linalg" } ] } ]