Module qose.driver_code_par

"""
Driver code par
***************
This script launches the tree search with a given configuration. We use MPI4Py to parallelize calculating the circuits
 at the leaves of the tree. Can be launched from the command line
with `mpiexec -n 1 --oversubscribe python driver_code_par.py`.
"""
from qose.subarchitecture_tree_search_par import run_tree_architecture_search
import pennylane as qml

import os
import pickle



if __name__ == "__main__":
    # Create a unique name for your experiment
    EXPERIMENT_NAME = 'alpha'

    # Create a directory to store the data
    if not os.path.exists('data'):
        os.mkdir('data/')

    data_path = f'data/{EXPERIMENT_NAME}'
    if not os.path.exists(data_path):
        os.mkdir(data_path)

    # Create a configuration file for the tree prune algorithm
    config = {'nqubits': 3,
              'min_tree_depth': 4,
              'max_tree_depth': 10,
              'prune_rate': 0.15,
              'prune_step': 3,
              'plot_trees': False,
              'data_set': 'moons',
              'nsteps': 2,
              'optim': qml.AdamOptimizer,
              'batch_sizes': [8, 16, 32, 64],
              'n_samples': 1500,
              'learning_rates': [0.001, 0.005, 0.01],
              'save_frequency': 1,
              'save_path': data_path,
              'nprocesses': 50,
              'save_timing': False,
              'circuit_type': 'schuld',
              'Tmax': [100, 100, 100],
              'inf_time': 'numcnots',
              'fill': 'redundant',  # or 'pad'
              'rate_type': 'accuracy',  # or 'batch_cost'
              'readout_layer': 'one_hot',  # or 'weighted_neuron'
              }

    # Save the configuration file so that we can remember what we did
    with open(data_path + '/config.pickle', 'wb') as f:
        pickle.dump(config, f)

    # Execute the algorithm
    run_tree_architecture_search(config, 'local')