dnn_solver.py

from keras.models import load_model

# import cube
# import numpy as np
# n_moves = 1
# Order: Left,Front,Right,Back,Up,Down - [0,1,2,3,4,5] - [Orange (0), Green (1), Red (2), Yellow (3), White (4), Blue (5)]
# initial_state = np.asarray([
#     [[0,5], [3,3]], # Left
#     [[2,5], [1,0]], # Front
#     [[4,0], [4,2]], # Right
#     [[4,3], [1,1]], # Back
#     [[5,5], [3,2]], # Top
#     [[2,1], [0,4]], # Bottom
# ])


# j = 0
# moves_list = []
# while (cube.isSolved(C1.state) is False) and (j < 21):
#     cube_state = np.asarray([C1.state]).astype(np.int)
#     moves_encodings = model.predict_classes(cube_state)
#     moves = cube.decode_moves(moves_encodings, side)
#     moves_list = moves_list + [moves]
#     C1.apply_moves(moves)
#     #cube.display(C1.state, C1.side, C1.colormap)
#     j = j + 1

# Set these parameters
side = 2

def get_action(state, model_name):
    model = load_model(model_name)
    return model.predict_classes(state)

def dnn_step(state):
    if side == 2:
        model_name = "model_5moves.h5"
    else:
        exit(-1)
    return get_action(state, model_name)