def analayzer(self):
vvec = vct.vectorizer
#Neural Net initializing
net = SNN.NetCnstr(21, 12, 2)
net.load_weights(
r'C:\Python34\Weights_store\Demands\input_to_hidden.npy',
'input_to_hidden')
net.load_weights(
r'C:\Python34\Weights_store\Demands\hidden_to_output.npy',
'hidden_to_output')
print('Weights are loaded successfully')
text = self.txt.get(index1='1.0', index2='end-1c')
self.txt.delete(index1='1.0', index2='end-1c')
splitted = text.split('\n')
estimation_holder = []
for i in splitted:
marks = net.query(vvec(i))
if marks[0] > marks[1]:
estimation_holder.append(marks[0])
else:
estimation_holder.append(0)
print(list(enumerate(estimation_holder)))
result = estimation_holder.index(max(estimation_holder))
splitted[result] = '<DEM ' + splitted[result] + ' /DEM>'
self.splitted = splitted
self.insert_tabs(mode='analyze')
self.txt.highlight_pattern('\<.+\>', 'green')
"train": {
"X": X[train_idxs],
"y_one_hot": y_one_hot[train_idxs],
"y": y[train_idxs],
},
"validation": {
"X": X[validation_idxs],
"y_one_hot": y_one_hot[validation_idxs],
"y": y[validation_idxs],
},
}
myNetwork = SNN.Network([
SNN.Layer(784, 100, activation="ReLU"),
SNN.Layer(100, 100, activation="ReLU"),
SNN.Layer(100, 10, activation="sigmoid")
],
loss="CategoricalCrossEntropyWithSoftmax",
optimizer="GradientDescentWithMomentum")
myNetwork.setParameters(lr=1)
myNetwork.fit(data['train']['X'],
data['train']['y_one_hot'],
epochs=200,
batch_size=1024)
y_test = data['validation']['y']
y_pred_onehot = myNetwork.predict(data['validation']['X'])
y_pred = np.argmax(y_pred_onehot, axis=1)
print(classification_report(y_test, y_pred))
'''
Model Test
'''
input = tf.placeholder(tf.float32)
input_exp = tf.exp(input)
groundtruth = tf.placeholder(tf.float32)
try:
w1 = np.load('weight_nslkdd11.npy')
w2 = np.load('weight_nslkdd12.npy')
w3 = np.load('weight_nslkdd13.npy')
layer_in = SNN.SNNLayer(312, 100, w1)
layer_out1 = SNN.SNNLayer(100, 100, w2)
layer_out2 = SNN.SNNLayer(100, 5, w3)
print('Weight loaded!')
except:
layer_in = SNN.SNNLayer(312, 100)
layer_out1 = SNN.SNNLayer(100, 100)
layer_out2 = SNN.SNNLayer(100, 5)
print('No weight file found, use random weight')
layerin_out = layer_in.forward(input_exp)
layerout_out1 = layer_out1.forward(layerin_out)
layerout_out2 = layer_out2.forward(layerout_out1)
# the time of output
nnout = tf.log(layerout_out2)
import numpy as np
import os
import matplotlib.pyplot as plt
import SNN
SAVE_PATH = os.getcwd() + '/weight_mnist'
mnist = SNN.MnistData(
path=["MNIST/t10k-images.idx3-ubyte", "MNIST/t10k-labels.idx1-ubyte"])
w1 = np.load(SAVE_PATH + '1.npy')
w2 = np.load(SAVE_PATH + '2.npy')
Ts = 1e-3
scale = 2
view_max = 2
l1 = SNN.SNNDiscrete(w1, Ts, scale)
l2 = SNN.SNNDiscrete(w2, Ts, scale)
xs, ys = mnist.next_batch(1, shuffle=True)
xs = (1 - xs[0, :]) / Ts
print(ys)
input_mat = np.zeros([784, int(1 / Ts * view_max)])
input_mat[range(784), xs.astype(int)] = 1
l1out = l1.forward(input_mat)
l2out = l2.forward(l1out)
fg, ax = plt.subplots(nrows=2, ncols=1)
for i in range(np.shape(l2.potential)[0]):
import matplotlib.pyplot as plt
import SNN
import data
SAVE_PATH = os.getcwd() + '/weight_mnist'
mnist = data.MNIST(
path=["MNIST/t10k-images.idx3-ubyte", "MNIST/t10k-labels.idx1-ubyte"])
w1 = np.load(SAVE_PATH + '1.npy')
w2 = np.load(SAVE_PATH + '2.npy')
Ts = 1e-3
scale = 2
view_max = 2
l1 = SNN.SNNDiscrete(w1, Ts, scale)
l2 = SNN.SNNDiscrete(w2, Ts, scale)
correct = 0
for i in range(mnist.datasize):
xs, ys = mnist.next_batch(1, shuffle=True)
xs = (1 - xs[0, :]) / Ts
input_mat = np.zeros([784, int(1 / Ts * view_max)])
input_mat[range(784), xs.astype(int)] = 1
l1out = l1.forward(input_mat)
l2out = l2.forward(l1out)
peak = np.argmax(l2out, axis=1)
import sys
sys.path.append("..")
import tensorflow as tf
import numpy as np
import os
import SNN
import data
input = tf.placeholder(tf.float32)
input_exp = tf.exp(input)
groundtruth = tf.placeholder(tf.float32)
layer_in = SNN.SNNLayer(784, 800)
layer_out = SNN.SNNLayer(800, 10)
layerin_out = layer_in.feedforward(input_exp)
layerout_out = layer_out.feedforward(layerin_out)
nnout = tf.log(layerout_out)
layerout_groundtruth = tf.concat([layerout_out, groundtruth], 1)
loss = tf.reduce_mean(tf.map_fn(SNN.loss_func, layerout_groundtruth))
wsc = layer_in.w_sum_cost() + layer_out.w_sum_cost()
l2c = layer_in.l2_cost() + layer_out.l2_cost()
K = 100
K2 = 1e-3
learning_rate = 1e-4
TRAINING_BATCH = 10
'''
Model Test
'''
input = tf.placeholder(tf.float32)
input_exp = tf.exp(input)
groundtruth = tf.placeholder(tf.float32)
try:
w1 = np.load('weight_awid1.npy')
w2 = np.load('weight_awid2.npy')
w3 = np.load('weight_awid3.npy')
layer_in = SNN.SNNLayer(206, 100, w1)
layer_out1 = SNN.SNNLayer(100, 100, w2)
layer_out2 = SNN.SNNLayer(100, 4, w3)
print('Weight loaded!')
except:
layer_in = SNN.SNNLayer(206, 100)
layer_out1 = SNN.SNNLayer(100, 100)
layer_out2 = SNN.SNNLayer(100, 4)
print('No weight file found, use random weight')
layerin_out = layer_in.forward(input_exp)
layerout_out1 = layer_out1.forward(layerin_out)
layerout_out2 = layer_out2.forward(layerout_out1)
# the time of output
nnout = tf.log(layerout_out2)
input_real = tf.placeholder(tf.float32)
output_real = tf.placeholder(tf.float32)
global_step = tf.Variable(1, dtype=tf.int64)
step_inc_op = tf.assign(global_step, global_step + 1)
'''
Here is a reshape, because TensorFlow DO NOT SUPPORT tf.extract_image_patches gradients operation for VARIABLE SIZE inputs
'''
input_real_resize = tf.reshape(tf.exp(input_real),[TRAINING_BATCH,28,28,1])
try:
w1 = np.load('weight_scnn1.npy')
w2 = np.load('weight_scnn2.npy')
w3 = np.load('weight_scnn3.npy')
print('Done')
layer1 = SNN.SCNNLayer(kernel_size=5,in_channel=1,out_channel=32,strides=2, w=w1)
layer2 = SNN.SCNNLayer(kernel_size=5,in_channel=32,out_channel=16,strides=2, w=w2)
layer3 = SNN.SNNLayer(in_size=784,out_size=10, w=w3)
print('Weight loaded!')
except:
layer1 = SNN.SCNNLayer(kernel_size=5,in_channel=1,out_channel=32,strides=2)
layer2 = SNN.SCNNLayer(kernel_size=5,in_channel=32,out_channel=16,strides=2)
layer3 = SNN.SNNLayer(in_size=784,out_size=10)
print('No weight file found, use random weight')
layerout1 = layer1.forward(input_real_resize)
layerout2 = layer2.forward(layerout1)
layerout3 = layer3.forward(tf.reshape(layerout2,[TRAINING_BATCH,784]))
wsc1 = layer1.kernel.w_sum_cost()
wsc2 = layer2.kernel.w_sum_cost()
import sys
sys.path.append("..")
import tensorflow as tf
import numpy as np
import os
import SNN
input = tf.placeholder(tf.float32)
input_exp = tf.exp(input)
groundtruth = tf.placeholder(tf.float32)
layer_in = SNN.SNNLayer(784, 800)
layer_out = SNN.SNNLayer(800, 10)
layerin_out = layer_in.forward(input_exp)
layerout_out = layer_out.forward(layerin_out)
nnout = tf.log(layerout_out)
layerout_groundtruth = tf.concat([layerout_out, groundtruth], 1)
loss = tf.reduce_mean(tf.map_fn(SNN.loss_func, layerout_groundtruth))
wsc = layer_in.w_sum_cost() + layer_out.w_sum_cost()
l2c = layer_in.l2_cost() + layer_out.l2_cost()
K = 100
K2 = 1e-3
learning_rate = 1e-4
TRAINING_BATCH = 10
SAVE_PATH = os.getcwd() + '/weight_mnist'
import sys
import matplotlib.pyplot as plt
sys.path.append("..")
import SNN
k = SNN.KittiData()
d = k.getdata(
"KITTI/2011_09_26/2011_09_26_drive_0005_sync/velodyne_points/data/0000000000.bin"
)
print(d)
plt.imshow(d)
plt.show()
import ribosome
import random as rd
Generation_Num = int(input("Input Generation Number"))
gene_list = []
for i in range(Generation_Num):
gene_list = Genetic.Read_Gene()
if (len(gene_list) == 0):
import startup
gene_score_list = []
selected_gene = []
descendants_gene = []
descendant_mutated = []
score_list = []
network_list = ribosome.Translate_Into_Networks(4, [1, 4], 2, -1)
score_list = SNN.return_score(network_list, i)
gene_score_list.append(score_list)
selected_gene: list = NEAT.calc_and_select_gene(gene_score_list, gene_list)
gene_num: int = int(len(gene_list) / 2)
for k in range(gene_num):
cross1 = rd.choice(selected_gene)
cross2 = rd.choice(selected_gene)
print(cross1, cross2)
Agene1, Agene2 = NEAT.Crossover(cross1, cross2)
descendants_gene.append(Agene1)
descendants_gene.append(Agene2)
for idx_gene, gene in enumerate(descendants_gene):
descendant_mutated.append(
NEAT.Mutate(gene, score_list[idx_gene], 1000000))
Genetic.Write_Gene(descendants_gene)
import sys
import matplotlib.pyplot as plt
import numpy as np
sys.path.append("..")
import SNN
mnist = SNN.MnistData(
path=["MNIST/train-images.idx3-ubyte", "MNIST/train-labels.idx1-ubyte"])
xs, ys = mnist.next_batch(batch_size=10, shuffle=True)
for i in range(10):
plt.imshow(np.reshape(xs[i], [28, 28]))
print(ys[i])
plt.show()