google_solver/convert_data.py

import torch
import numpy as np


def convert_tensor(x):
    """
    Convert a PyTorch tensor to a nested Python list of integers.

    Args:
        x (torch.Tensor): Tensor to convert.

    Returns:
        list: Converted nested list.
    """
    x = x.long().cpu().numpy().astype(int)
    if x.ndim == 1:
        return list(x)
    return [list(row) for row in x]


def make_time_windows(start_time, end_time):
    """
    Concatenate start and end time tensors to form time windows.

    Args:
        start_time (torch.Tensor): Start times (B, N, 1)
        end_time (torch.Tensor): End times (B, N, 1)

    Returns:
        torch.Tensor: Time windows (B, N, 2)
    """
    return torch.cat([start_time, end_time], dim=2)


def convert_data(input_data, scale_factor):
    """
    Convert batched graph and fleet data to OR-Tools compatible format.

    Args:
        input_data (tuple): Tuple of (graph_data, fleet_data) as dictionaries.
        scale_factor (float): Scaling factor to convert float to integer.

    Returns:
        list: List of dictionaries, one per batch item, containing:
              - distance_matrix
              - time_matrix
              - time_windows
              - depot index (default 0)
              - num_vehicles
    """
    graph_data, fleet_data = input_data

    start_times = graph_data['start_times']
    end_times = graph_data['end_times']
    distance_matrix = graph_data['distance_matrix']
    time_matrix = graph_data['time_matrix']
    time_windows = make_time_windows(start_times, end_times)

    batch_size = distance_matrix.size(0)
    converted_data = []

    for i in range(batch_size):
        space_mat = (distance_matrix[i] * scale_factor)
        time_mat = (time_matrix[i] * scale_factor)
        windows = (time_windows[i] * scale_factor)

        sample_dict = {
            'distance_matrix': convert_tensor(space_mat),
            'time_matrix': convert_tensor(time_mat),
            'time_windows': convert_tensor(windows),
            'depot': 0,
            'num_vehicles': distance_matrix[i].shape[1]  # assuming square matrix
        }

        converted_data.append(sample_dict)

    return converted_data