def build_network():
network = nn.neural_network(LAYER_LEN + 2)
# input layer
network.create_neuron(0, [1, 0, 0, 0], 1, 'linear')
network.create_neuron(0, [0, 1, 0, 0], 1, 'linear')
network.create_neuron(0, [0, 0, 1, 0], 1, 'linear')
network.create_neuron(0, [0, 0, 0, 1], 1, 'linear')
# hidden layer
for l in range(1, LAYER_LEN + 1, 1):
lbc = len(network.layers[l - 1])
for _ in range(HIDDEN_LEN):
h = network.create_neuron(l, nn.rand_weights(lbc), 1, 'log')
for uid in network.layers[l - 1]:
network.create_connection(uid, h)
# output layer
lbc = len(network.layers[-2])
out1 = network.create_neuron(LAYER_LEN + 1, nn.rand_weights(lbc), 1, 'log')
out2 = network.create_neuron(LAYER_LEN + 1, nn.rand_weights(lbc), 1, 'log')
out3 = network.create_neuron(LAYER_LEN + 1, nn.rand_weights(lbc), 1, 'log')
for uid in network.layers[-2]:
network.create_connection(uid, out1)
network.create_connection(uid, out2)
network.create_connection(uid, out3)
return network
def __init__(self, x_offset, y_offset, template=None):
self.longest_chain = 0
self.time_alive = 0
self.finished = False
self.board = [[' ' for x in range(self.w)] for y in range(self.h)]
self.prev_board = [[' ' for x in range(self.w)] for y in range(self.h)]
self.offset = (x_offset, y_offset)
# create a neural network for the player
# first layer = size of input (ie: the size of the board)
# several internal layers, each with an experimental size
# finally an output layer, with 4 outputs (one corresponding to each action)
# TODO: calculate a list of available placements given the board, set these as the 'output' of NN
# I'm interested to see how this affects how the ai solves the problem
self.net = neuralnetwork.neural_network([(self.w * self.h) + 6 + 4,
250, 250, 125, 125, 4])
# scoring variables
self.chain = 0 # the current chain length
self.chain_size = 0 # the number of puyos in the current chain
self.score = 0 # the players score
self.diff_colors_in_chain = set(
) # the size of this will let us determine a bonus
self.chain_group_sizes = [] # the size of each group in a chain
self.puyo_to_remove = set()
self.falling = None
self.trigger = None
self.moves = [
((-1, 0), (0, 0)), ((-1, 1), (0, 1)), ((-1, 2), (0, 2)),
((-1, 3), (0, 3)), ((-1, 4), (0, 4)), ((-1, 5), (0, 5)),
((0, 0), (-1, 0)), ((0, 1), (-1, 1)), ((0, 2), (-1, 2)),
((0, 3), (-1, 3)), ((0, 4), (-1, 4)), ((0, 5), (-1, 5)),
((-1, 0), (-1, 1)), ((-1, 1), (-1, 2)), ((-1, 2), (-1, 3)),
((-1, 3), (-1, 4)), ((-1, 4), (-1, 5)), ((-1, 1), (-1, 0)),
((-1, 2), (-1, 1)), ((-1, 3), (-1, 2)), ((-1, 4), (-1, 3)),
((-1, 5), (-1, 4))
]
self.buffer = [self.next(), self.next()]
self.current = self.next()
row = 0
col = 0
if template:
for puyos in template:
if puyos == "\n":
col += 1
row = 0
continue
else:
self.board[col][row] = puyos
row += 1
def build_network():
network = nn.neural_network(2)
# input layer
network.create_neuron(0, [1, 0, 0, 0])
network.create_neuron(0, [0, 1, 0, 0])
network.create_neuron(0, [0, 0, 1, 0])
network.create_neuron(0, [0, 0, 0, 1])
# output layer
for _ in range(15):
uid = network.create_neuron(1, nn.rand_weights(4))
network.create_connection(0, uid)
network.create_connection(1, uid)
network.create_connection(2, uid)
network.create_connection(3, uid)
return network
def build_network():
network = nn.neural_network(3)
# input layer
in1 = network.create_neuron(0, [1, 0], 0.5, 'threshold')
in2 = network.create_neuron(0, [0, 1], 0.5, 'threshold')
# hidden layer
huid = []
for _ in range(HIDDEN_LEN):
h = network.create_neuron(1, nn.rand_weights(2), 1, 'log')
network.create_connection(in1, h)
network.create_connection(in2, h)
huid.append(h)
# output layer
out1 = network.create_neuron(2, nn.rand_weights(HIDDEN_LEN), 1, 'log')
for uid in huid:
network.create_connection(uid, out1)
return network
import numpy as np
import neuralnetwork as nn
from mnist import MNIST
print("Starting...")
mndata = MNIST('mnist-dataset')
images, labels = mndata.load_training()
print("Training...")
# train
num_classes = 10
targets = np.array([labels]).reshape(-1)
one_hot_targets = np.eye(num_classes)[labels]
net = nn.neural_network(3, [784, 20, 10], [None, "tanh", "softmax"],
cost_function="cross_entropy")
net.train(1,
inputs=images,
labels=one_hot_targets,
num_epochs=5,
learning_rate=0.001,
filename="savepoint.pkl")
print("Testing...")
# test
images, labels = mndata.load_testing()
targets = np.array([labels]).reshape(-1)
one_hot_targets = np.eye(num_classes)[labels]
net.check_accuracy("savepoint.pkl", images, one_hot_targets)