def __loadNNets(self, name, includeOptimisers=False):
nNetA = NeuralNet(in_channels=3).to(self.device)
nNetB = NeuralNet(in_channels=4).to(self.device)
nNetC = NeuralNet(in_channels=4).to(self.device)
if name is not None:
name = name.replace(".pth", "a.pth")
nNetA.load(name)
name = name.replace("a.pth", "b.pth")
nNetB.load(name)
name = name.replace("b.pth", "c.pth")
nNetC.load(name)
else:
nNetA.loadMostRecent("a.pth")
nNetB.loadMostRecent("b.pth")
nNetC.loadMostRecent("c.pth")
nnets = nNetA, nNetB, nNetC
if includeOptimisers:
optimisers = tuple(torch.optim.Adam(
N.parameters(), lr=0.001) for N in nnets)
return nnets, optimisers
else:
return nnets
def predict(args):
day = args[2]
num_lineups = int(args[3])
filename = FILENAME_USED_NN
if len(args) == 5:
filename = "trainedModels/" + args[4]
day_x, playerList = getInputForDay(day, True)
print(day_x.shape)
print(len(playerList))
N, M = day_x.shape
day_X = np.ones((N, M + 1))
day_X[:, 1:] = day_x
model = NeuralNet(NeuralNet.load(filename))
scores, predictedLineups = model.getPrediction(day_X, playerList,
num_lineups)
print()
print("No noise lineup : ")
for i in range(len(scores)):
predictedLineup = predictedLineups[i]
print("player selected:")
for player in predictedLineup:
print(player.toString())
print()
print("Expected total score :" + str(scores[i]))
print("---------------")
def test(train_X, test_X, train_y, test_y, args):
filename = FILENAME_USED_NN
if len(args) == 3:
filename = "trainedModels/" + args[2]
model_loaded = NeuralNet.load(filename)
model = NeuralNet(model_loaded)
print("final train accuracy:", model.score(train_X, train_y))
print("final test accuracy:", model.score(test_X, test_y))
def make_model(model=None,
hidden_layer_sizes=None,
learning_r=None,
batch_size_2=None,
dropout_rate=None,
epoch=None):
optimizer = DEFAULT_OPTIMIZER(learning_r)
loss = 'mean_squared_error'
return NeuralNet(model, INPUT_SIZE, OUTPUT_SIZE, hidden_layer_sizes,
optimizer, loss, batch_size_2, dropout_rate, epoch)
def convolutionTest(self): nn = NeuralNet() expected = np.array([[-1,0], [-3,-11], [-5,-17]]) # Convolve the vertical Sobel edge detector data = np.array([[1,1,1,0], [0,1,0,0], [2,1,3,4], [5,1,6,7], [8,9,10,11]]) kernel = np.array([[1,0,-1], [2,0,-2], [1,0,-1]]) output = nn.convolve(data, kernel) return self.arrayIsSame(output, expected)
def activationTest(self): nn = NeuralNet() data = np.array([[1,0,0,0], [1,1,1,0], [1,0,0,0], [1,0,0,0], [1,0,0,1]]) kernel = np.array([[1,0,-1], [2,0,-2], [1,0,-1]]) output = nn.convolve(data, kernel) expectedRelu = np.array([[2,2], [3,1], [4,0]]) expectedBin = np.array([[1,1], [1,1], [1,0]]) expectedSigmoid = np.array([[0.119,0.119], [0.047,0.269], [0.0180,0.731]]) expectedTanH = np.array([[0.964,0.964], [0.995,0.762], [0.999,-0.762]]) expected = np.array([[-1,0], [-3,-11], [-5,-17]]) reluOutput = nn.activationFunction(output, 'relu') binOutput = nn.activationFunction(output, 'binary') sigmoidOutput = nn.activationFunction(output, 'sigmoid') tanhOutput = nn.activationFunction(output, 'tanh') for index, value in np.ndenumerate(sigmoidOutput): sigmoidOutput[index] = round(value, 3) for index, value in np.ndenumerate(tanhOutput): tanhOutput[index] = round(value, 3) if not UnitTests.arrayIsSame(reluOutput, expectedRelu): print "Relu activationFunction test failed." print "Expected:" print expectedRelu print "Actual:" print reluOutput return False if not UnitTests.arrayIsSame(binOutput, expectedBin): print "Binary activationFunction test failed." print "Expected:" print expectedBin print "Actual:" print binOutput return False if not UnitTests.arrayIsSame(sigmoidOutput, expectedSigmoid): print "Sigmoid activationFunction test failed." print "Expected:" print expectedSigmoid print "Actual:" print sigmoidOutput return False if not UnitTests.arrayIsSame(tanhOutput, expectedTanH): print "tanh activationFunction test failed." print "Expected:" print expectedTanH print "Actual:" print tanhOutput return False return True
def main():
'''Experiment on NN tos simulate y= f(x1, x2) = sin(2*pi*x1)*sin(2*pi*x2)'''
'''Case 11: randomly generated inputs with random weights and biases
Optimized to yield the best network performance
'''
datafile = './data/perf/data.txt'
testfile = './data/perf/test.txt'
trainvcsv = './data/perf/training.csv'
testcsv = './data/perf/testing.csv'
imgfile = './data/perf/loss.png'
wbfile = './data/perf/weights_biases.txt'
dm.makeDir('./data/perf')
dm.genRandData(datafile, 50)
dm.genData(testfile, 10)
perf_number = 0
max_perf = 0
while perf_number < 75:
dataset = []
dm.parseData(datafile, dataset)
NN = NeuralNet(3, True) # create NN with 3 hidden neurons
NN.savewb('Weights and Biases before Training', wbfile)
NN.train(dataset, imgfile, trainvcsv, 500) # train the neural network
NN.savewb('Weights and Biases after Training', wbfile)
testset = []
dm.parseData(testfile, testset)
perf_number = NN.test(testset, testcsv)
if perf_number > max_perf:
max_perf = perf_number
print(f'\n\nMax perf: {max_perf}\n\n')
with open(wbfile, 'a+') as F:
F.write(f'\n\nMax perf: {perf_number}\n\n')
def maxpoolTest(self): nn = NeuralNet() # Should fail - Wrong size array expected = np.array([[4,4], [0,4]]) data = np.array([[4,0,1,3,4], [0,0,2,4,4], [0,0,4,4,4], [0,0,4,4,4]]) windowsize = 2 # 2x2 window output = nn.maxPool(data, windowsize) if output is not None: return False # Should pass expected = np.array([[4,4], [0,4]]) data = np.array([[4,0,1,3], [0,0,2,4], [0,0,4,4], [0,0,4,4]]) windowsize = 2 # 2x2 window output = nn.maxPool(data, windowsize) return self.arrayIsSame(output,expected)
c2 = c1[0] + t_size[0] + 3, c1[1] + t_size[1] + 4
cv2.rectangle(img, c1, c2, color, -1)
cv2.putText(img, label, (c1[0], c1[1] + t_size[1] + 4),
cv2.FONT_HERSHEY_PLAIN, 1, [225, 255, 255], 1)
return img
num_classes = 80
classes = loadClasses("data/coco.names")
confidence = 0.5
nms_thresh = 0.4
dimension = 416
print("loading network")
model = NeuralNet("yolov3.cfg")
model.loadWeights("yolov3.weights")
print("network loaded")
model.networkInfo["height"] = 416
model.cuda()
model.eval()
camera = cv2.VideoCapture(0)
while True:
ret, frame = camera.read()
image, orig_im = preprocess(frame, 416)
image = image.cuda()
def __init__(self):
self.net = NeuralNet(nodes=config["nodes"])
def trainTest(self, dataFilepath = None): nn = NeuralNet() nn.train() return
#!/usr/bin/env python
# -*- coding: utf-8 -*-
# These two lines are necessary to find source files!!!
import sys
sys.path.append('../src')
from neuralNet import NeuralNet, DataSet
from files import files
if __name__ == '__main__':
f = files["haberman"]
ds = DataSet(f)
n = NeuralNet([3, 1, 2], ds.dataMatrix, numericalEvaluation=True)
n.startTraining(1)
from optimizer import Optimizer from neuralNet import NeuralNet import functions size = [1, 2, 3] training_examples = [1, 4, 5] training_targets = [2, 8, 10] model = NeuralNet(size, last_activation_function=functions.relu) model.fit(10, training_examples, training_targets, 0.1)
