Source code for stgraph.dataset.temporal.metrla_dataloader

"""Traffic forecasting based on Los Angeles city."""

from __future__ import annotations

import numpy as np
import torch

from stgraph.dataset.temporal.stgraph_temporal_dataset import STGraphTemporalDataset


[docs]class METRLADataLoader(STGraphTemporalDataset): r"""Traffic forecasting dataset based on the Los Angeles city. A dataset for predicting traffic patterns in the Los Angeles Metropolitan area, comprising traffic data obtained from 207 loop detectors on highways in Los Angeles County. The dataset includes aggregated 5-minute interval readings spanning a four-month period from March 2012 to June 2012. This class provides functionality for loading, processing, and accessing the METRLA dataset for use in deep learning tasks such as traffic forecasting. .. list-table:: gdata :widths: 33 33 33 :header-rows: 1 * - num_nodes - num_edges - total_timestamps * - 207 - 1722 - 100 Example ------- .. code-block:: python from stgraph.dataset import METRLADataLoader metrla = METRLADataLoader(verbose=True) num_nodes = metrla.gdata["num_nodes"] num_edges = metrla.gdata["num_edges"] total_timestamps = metrla.gdata["total_timestamps"] edge_list = metrla.get_edges() edge_weights = metrla.get_edge_weights() feats = metrla.get_all_features() targets = metrla.get_all_targets() Parameters ---------- verbose : bool, optional Flag to control whether to display verbose info (default is False) num_timesteps_in : int, optional The number of timesteps the sequence model sees (default is 12) num_timesteps_out : int, optional The number of timesteps the sequence model has to predict (default is 12) cutoff_time : int, optional The cutoff timestamp for the temporal dataset (default is None) redownload : bool, optional (default is False) Redownload the dataset online and save to cache Attributes ---------- name : str The name of the dataset. gdata : dict Graph meta data. """ def __init__( self: METRLADataLoader, verbose: bool = False, num_timesteps_in: int = 12, num_timesteps_out: int = 12, cutoff_time: int | None = None, redownload: bool = False, ) -> None: """Traffic forecasting based on the Los Angeles city.""" super().__init__() if not isinstance(num_timesteps_in, int): raise TypeError("num_timesteps_in must be of type int") if num_timesteps_in < 0: raise ValueError("num_timesteps_in must be a positive integer") if not isinstance(num_timesteps_out, int): raise TypeError("num_timesteps_out must be of type int") if num_timesteps_out < 0: raise ValueError("num_timesteps_out must be a positive integer") if cutoff_time is not None and not isinstance(cutoff_time, int): raise TypeError("cutoff_time must be of type int") if cutoff_time is not None and cutoff_time < 0: raise ValueError("cutoff_time must be a positive integer") self.name = "METRLA" self._url = "https://raw.githubusercontent.com/bfGraph/STGraph-Datasets/main/METRLA.json" self._verbose = verbose self._num_timesteps_in = num_timesteps_in self._num_timesteps_out = num_timesteps_out self._cutoff_time = cutoff_time self._edge_list = None self._edge_weights = None self._all_features = None self._all_targets = None if redownload and self._has_dataset_cache(): self._delete_cached_dataset() if self._has_dataset_cache(): self._load_dataset() else: self._download_dataset() self._save_dataset() self._process_dataset() def _process_dataset(self: METRLADataLoader) -> None: self._set_total_timestamps() self._set_num_nodes() self._set_num_edges() self._set_edges() self._set_edge_weights() self._set_targets_and_features() def _set_total_timestamps(self: METRLADataLoader) -> None: r"""Set the total timestamps present in the dataset. It sets the total timestamps present in the dataset into the gdata attribute dictionary. It is the minimum of the cutoff time choosen by the user and the total time periods present in the original dataset. """ if self._cutoff_time is not None: self.gdata["total_timestamps"] = min( self._dataset["time_periods"], self._cutoff_time, ) else: self.gdata["total_timestamps"] = self._dataset["time_periods"] def _set_num_nodes(self: METRLADataLoader) -> None: r"""Set the total number of nodes present in the dataset.""" node_set = set() max_node_id = 0 for edge in self._dataset["edges"]: node_set.add(edge[0]) node_set.add(edge[1]) max_node_id = max(max_node_id, edge[0], edge[1]) if max_node_id != len(node_set) - 1: raise RuntimeError("Node ID labelling is not continuous") self.gdata["num_nodes"] = len(node_set) def _set_num_edges(self: METRLADataLoader) -> None: r"""Set the total number of edges present in the dataset.""" self.gdata["num_edges"] = len(self._dataset["edges"]) def _set_edges(self: METRLADataLoader) -> None: r"""Set the edge list of the dataset.""" self._edge_list = [(edge[0], edge[1]) for edge in self._dataset["edges"]] def _set_edge_weights(self: METRLADataLoader) -> None: r"""Set the edge weights of the dataset.""" edges = self._dataset["edges"] edge_weights = self._dataset["weights"] comb_edge_list = [ (edges[i][0], edges[i][1], edge_weights[i]) for i in range(len(edges)) ] comb_edge_list.sort(key=lambda x: (x[1], x[0])) self._edge_weights = np.array([edge_det[2] for edge_det in comb_edge_list]) def _set_targets_and_features(self: METRLADataLoader) -> None: r"""Calculate and set the target and feature attributes.""" x = [ self._dataset[str(timestamp)] for timestamp in range(self.gdata["total_timestamps"]) ] x = np.array(x).transpose(1, 2, 0).astype(np.float32) # x = x.transpose((1, 2, 0)) # x = x.astype(np.float32) # Normalise as in DCRNN paper (via Z-Score Method) means = np.mean(x, axis=(0, 2)) x = x - means.reshape(1, -1, 1) stds = np.std(x, axis=(0, 2)) x = x / stds.reshape(1, -1, 1) x = torch.from_numpy(x) indices = [ (i, i + (self._num_timesteps_in + self._num_timesteps_out)) for i in range( x.shape[2] - (self._num_timesteps_in + self._num_timesteps_out) + 1, ) ] # Generate observations features, target = [], [] for i, j in indices: features.append((x[:, :, i : i + self._num_timesteps_in]).numpy()) target.append((x[:, 0, i + self._num_timesteps_in : j]).numpy()) self._all_features = np.array(features) self._all_targets = np.array(target)
[docs] def get_edges(self: METRLADataLoader) -> list: r"""Return the edge list.""" return self._edge_list
[docs] def get_edge_weights(self: METRLADataLoader) -> np.ndarray: r"""Return the edge weights.""" return self._edge_weights
[docs] def get_all_targets(self: METRLADataLoader) -> np.ndarray: r"""Return the targets for each timestamp.""" return self._all_targets
[docs] def get_all_features(self: METRLADataLoader) -> np.ndarray: r"""Return the features for each timestamp.""" return self._all_features