RecursiveRouteChoice

RecursiveRouteChoice is an efficient Python implementation of the recursive logit model for route choice. This kind of model was first developed in Fosgerau, Frejinger & Karlstrom 2013

This implementation contains both prediction and estimation models. See the docs for a more detailed introduction. Here is a quick start overview.

Installation

The current most direct method of installation is to clone the repository and install using pip:

pip install -r requirements.txt

Note that (currently) on windows the 2004 update has broken some interactions with OpenBLAS. To avoid any problems it is recommended to use conda, which will use MKL BLAS and avoid any issues:

conda install --file (cat requirements.txt)

Note that there are no direct problems related to this in the code but pip will install numpy 1.19.4 by default, which will immediately crash to warn about this. Reverting to numpy 1.19.3 "fixes" this in that the warning isn't triggered, but other code which does require on the broken functionality may silently fail.

It is recommended to use some kind of virtual environment to avoid conflicting package versions. There is also a requirements_strict.txt which explicitly specifies package versions. This will likely contain less up to date versions, but should always work regardless of future api changes of the dependencies.

Example Usage

Note the tntp file here is not included in the repository data, download separately from TransportationNetworks

import numpy as np
from recursiveRouteChoice.data_loading import load_tntp_node_formulation
from recursiveRouteChoice import RecursiveLogitModelPrediction, ModelDataStruct, \
RecursiveLogitModelEstimation, optimisers

# DATA
network_file = "SiouxFalls_net.tntp"
node_max = 24  # from network file

data_list, data_list_names = load_tntp_node_formulation(
    network_file, columns_to_extract=["length", "capacity"], sparse_format=False)

distances = data_list[0]

incidence_mat = (distances > 0).astype(int)
network_struct = ModelDataStruct(data_list, incidence_mat)

beta_sim = np.array([-0.8, -0.00015])
model = RecursiveLogitModelPrediction(network_struct,
                                      initial_beta=beta_sim, mu=1)
print("Linear system size", model.get_exponential_utility_matrix().shape)

orig_indices = np.arange(0, node_max, 1)
# dest_indices = (orig_indices + 5) % node_max
dest_indices = np.arange(0, node_max, 1)  # sample every OD pair
obs_per_pair = 1
print(f"Generating {obs_per_pair * len(orig_indices) * len(dest_indices)} obs total per "
      f"configuration")
seed = 42
obs = model.generate_observations(origin_indices=orig_indices, dest_indices=dest_indices,
                                  num_obs_per_pair=obs_per_pair, iter_cap=2000, rng_seed=seed)

optimiser = optimisers.ScipyOptimiser(method='l-bfgs-b')
beta_est_init = [-5, -0.00001]
model_est = RecursiveLogitModelEstimation(network_struct, observations_record=obs,
                                          initial_beta=beta_est_init, mu=1,
                                          optimiser=optimiser)

beta = model_est.solve_for_optimal_beta(verbose=False)

print(f"beta expected: [{beta_sim[0]:6.4f}, {beta_sim[1]:6.4f}],"
              f" beta_actual: [{beta[0]:6.4f}, {beta[1]:6.4f}]")