def main():
# Number of nodes
# Number of input pixels
input_nodes = 784
# Arbitrary
hidden_nodes = 100
# Number of options for output
output_nodes = 10
learning_rate = 0.3
network = nn.neuralNetwork(input_nodes, hidden_nodes, output_nodes, learning_rate)
print("Training network! May take some time...")
train(network, output_nodes)
print("Test the network!")
print("Options: ")
print(" Enter a number, 0-9, to test the image from mnist_dataset/images/test")
print(" Enter 'all' to test against all images in mnist_dataset/images/test")
print(" Type 'q' to quit")
choice = input("Input: ")
while choice is not 'q':
try:
input_num = int(choice)
test(network, input_num)
except ValueError:
if choice == 'all':
test_all(network)
else:
print("Error - unknown command")
choice = input("Input: ")
def __init__(self):
# Change this to whatever your emulator window name is
self.WINDOW_NAME = WINDOW_NAME = "Super Mario 64 (U)"
# Find window dimensions and apply padding
self.DIMENSIONS = self.get_window_dimensions()
PADDING_LEFT = 0
PADDING_TOP = 70
PADDING_RIGHT = 20
PADDING_BOTTOM = 70
self.DIMENSIONS = (self.DIMENSIONS[0] + PADDING_LEFT,
self.DIMENSIONS[1] + PADDING_TOP,
self.DIMENSIONS[2] - PADDING_LEFT - PADDING_RIGHT,
self.DIMENSIONS[3] - PADDING_TOP - PADDING_BOTTOM)
assert self.DIMENSIONS != None, "Error: Could not find game window " + WINDOW_NAME
#LEFT = self.DIMENSIONS[0]
#TOP = self.DIMENSIONS[1]
WIDTH = self.DIMENSIONS[2]
HEIGHT = self.DIMENSIONS[3]
# Neural network optimization.
# Use 1 for no compression. Higher numbers will reduce the size of the neural network and input nodes.
self.COMPRESSION = 8
# Determine, based on compression, the width and height of the compressed input image.
self.COMPRESSED_WIDTH = round(WIDTH / self.COMPRESSION)
self.COMPRESSED_HEIGHT = round(HEIGHT / self.COMPRESSION)
# Determine the size of each layer.
self.INPUT_NODES = self.COMPRESSED_WIDTH * self.COMPRESSED_HEIGHT
self.OUTPUT_NODES = 8
self.HIDDEN_NODES_1 = round(
(self.INPUT_NODES * 2 + self.OUTPUT_NODES) / 3)
self.HIDDEN_NODES_2 = round(
(self.INPUT_NODES + self.OUTPUT_NODES * 2) / 3)
# Neural network settings
self.LEARNING_RATE = 0.001 # How quickly nodes will be adjusted to compensate for error.
self.KEYPRESS_THRESHOLD = 0.5 # The neural network must output a value higher than this in order for it to be considered a press.
# Instantiate a neural network object
self.neural_network = neuralnetwork.neuralNetwork(
inputnodes=self.INPUT_NODES,
hiddennodes1=self.HIDDEN_NODES_1,
hiddennodes2=self.HIDDEN_NODES_2,
outputnodes=self.OUTPUT_NODES,
learningrate=self.LEARNING_RATE)
def initNetwork() -> neuralNetwork:
with open("config.yml", 'r') as ymlfile:
cfg = yaml.load(ymlfile, Loader=yaml.FullLoader)
# Default configuration for 28x28 pixel images.
input_nodes = cfg.get("input_nodes", 784)
hidden_nodes = cfg.get("hidden_nodes", 200)
output_nodes = cfg.get("output_nodes", 10)
learning_rate = cfg.get("learning_rate", 0.01)
n = neuralNetwork(input_nodes, hidden_nodes, output_nodes, learning_rate)
if cfg.get("kernel", None):
n.load(cfg.get("kernel"))
print(f"loaded kernel {n.kernel_name}")
return n
import numpy
import imageio
from tqdm import tqdm
from neuralnetwork import neuralNetwork
# ===学習モデルを読み込み開始====
parameters = numpy.load('trainedModel/parameters.npy')
# 入力層、隠れ層、出力層のノード数
input_nodes = int(parameters[0])
hidden_nodes = int(parameters[1])
output_nodes = int(parameters[2])
# 学習率
learning_rate = parameters[3]
# ニューラルネットワークのインスタンス生成
n = neuralNetwork(input_nodes, hidden_nodes, output_nodes, learning_rate)
# 保存した重みを読み込む
n.load_weights()
# ===学習モデルを読み込み終わり====
# run the network backwards, given a label, see what image it produces
# label to test
label = 6
# create the output signals for this label
targets = numpy.zeros(output_nodes) + 0.01
# all_values[0] is the target label for this record
targets[label] = 0.99
print(targets)
# get image data
image_data = n.backquery(targets)
import loadData as ld
import shutil
import os
import build_dataset as bd
import neuralnetwork as nn
import prediction as pr
# Check if data folder exists inside local git and delete it to have a clean folder available
if os.path.isdir('data'):
shutil.rmtree('data')
ld.resize_dataset()
bd.buildDataSet('data/*.jpg')
nn.neuralNetwork()
pr.prediction()
def neuralNetwork(self):
return neuralNetwork()