def conv_search(cost):
tf.reset_default_graph()
a = grid_param_conv['conv1_x']
f = open("scores\\scores_convxy", "a")
for i in range(len(grid_param_conv['conv1_x'])):
net_train([a[i], 32, 3], [42, 42, 14], cost, [1, 8],
[6 - a[i], math.floor(
(1280 - 32) / (13 * 8))], tf.orthogonal_initializer,
0.0005, "model_" + str(cost) + "_convxy_demo", 10,
"channels_586202_ordered_batch_3608.mat",
"labels_586202_batch_3608.mat")
tf.reset_default_graph()
[fpr, fnr, frar, nsrar, fsfpr
] = test_net([a[i], 32, 3], [42, 42, 14], cost, [1, 8],
[6 - a[i], math.floor(
(1280 - 32) / (13 * 8))], tf.orthogonal_initializer,
"model_" + str(cost) + "_convxy_demo", 11, 20,
"channels_586202_ordered_batch_3608.mat",
"labels_586202_batch_3608.mat")
print("convxy: ", a[i])
print("fpr: ", fpr)
print("fnr: ", fnr)
print("frar: ", frar)
f.write("convxy: " + str(a[i]) + "\tfpr: " + str(fpr) + "\tfnr: " +
str(fnr) + "\tfrar" + str(frar) + "\tnsrar" + str(nsrar) +
" \tfsfpr" + str(fsfpr) + "\n")
tf.reset_default_graph()
f.close()
def lr_search():
lrs = [
0.01, 0.005, 0.001, 0.0005, 0.0001, 0.00005, 0.00001, 0.000005,
0.000001, 0.0000005
]
for i in range(len(lrs)):
tf.reset_default_graph()
f = open("scores_lr", "a")
net_train([3, 48, 1], [39, 39, 13], 50, [1, 8], [3, 14],
tf.keras.initializers.glorot_normal, lrs[i],
"model_" + str(lrs[i]) + "_lr")
tf.reset_default_graph()
[fpr, fnr,
score] = test_net([3, 48, 1], [39, 39, 13], 50, [1, 8], [3, 14],
tf.keras.initializers.glorot_normal,
"model_" + str(lrs[i]) + "_lr")
print("lr: ", lrs[i])
print("fpr: ", fpr)
print("fnr: ", fnr)
print("score: ", score)
f.write("lr: " + str(lrs[i]) + "\tscore: " + str(score) + "\tfpr: " +
str(fpr) + "\tfnr: " + str(fnr) + "\n")
f.close()
def search_initializers(init, stri):
i = 0
tf.reset_default_graph()
f = open("scores\\scores_inits", "a")
net_train([4, 32, 3], [42, 42, 14], 50, [1, 8],
[2, math.floor((1280 - 32) / (13 * 8))], init, 0.0005,
"model_" + stri + "_inits", 10,
"channels_586202_ordered_batch_3608.mat",
"labels_586202_batch_3608.mat")
tf.reset_default_graph()
[fpr, fnr, frar, nsrar,
fsfpr] = test_net([4, 32, 3], [42, 42, 14], 50, [1, 8],
[2, math.floor((1280 - 32) / (13 * 8))], init,
"model_" + stri + "_inits", 11, 20,
"channels_586202_ordered_batch_3608.mat",
"labels_586202_batch_3608.mat")
print("init: ", stri)
print("fpr: ", fpr)
print("fnr: ", fnr)
print("frar: ", frar)
f.write("init: " + stri + "\tfpr: " + str(fpr) + "\tfnr: " + str(fnr) +
"\tfrar" + str(frar) + "\tnsrar" + str(nsrar) + " \tfsfpr" +
str(fsfpr) + "\n")
f.close()
def search_width(iters, cost, cx, val, init):
i = 0
tf.reset_default_graph()
neurons = init
prev = 9999
while (i < iters):
f = open("scores_neuron_width", "a")
net_train([cx, val, 1],
[neurons, neurons, math.floor(neurons / 3)], cost, [1, 8],
[6 - cx, math.floor((1280 - val) / (13 * 8))],
"model_" + str(neurons) + "_neuron_width")
tf.reset_default_graph()
[fpr, fnr, score] = test_net(
[cx, val, 1],
[neurons, neurons, math.floor(neurons / 3)], cost, [1, 8],
[6 - cx, math.floor((1280 - val) / (13 * 8))],
"model_" + str(neurons) + "_neuron_width")
print("width: ", neurons)
print("fpr: ", fpr)
print("fnr: ", fnr)
print("score: ", score)
f.write("cost: " + str(neurons) + "\tscore: " + str(score) +
"\tfpr: " + str(fpr) + "\tfnr: " + str(fnr) + "\n")
#if score >= prev:
# direction = direction * -1
# step1 = step1 - step2
#if step1 <= 0:
# break
neurons = neurons + 80
prev = score
tf.reset_default_graph()
f.close()
i = i + 1
def conv_search_z(iters, cost, cx, init, step1, step2, direction):
tf.reset_default_graph()
i = 0
f = open("scores_convz", "a")
val = init
prev = 99999
while i < iters:
net_train([cx, val, 1], [39, 39, 13], cost, [1, 8],
[6 - cx, math.floor((1280 - val) / (13 * 8))],
"model_" + str(val) + "_convz")
tf.reset_default_graph()
[fpr, fnr, score
] = test_net([cx, val, 1], [39, 39, 13], cost, [1, 8],
[6 - cx, math.floor((1280 - val) / (13 * 8))],
"model_" + str(val) + "_convz")
print("conv: ", val)
print("fpr: ", fpr)
print("fnr: ", fnr)
print("score: ", score)
f.write("convz: " + str(val) + "\tscore: " + str(score) + "\n")
if score >= prev:
direction = direction * -1
step1 = step1 - step2
if step1 <= 0:
break
val = val + step1 * direction
prev = score
tf.reset_default_graph()
i = i + 1
def search_square(init, fs, it):
tf.reset_default_graph()
f = open("scores_convxy_2", "a")
net_train([fs, 48, 1], [39, 39, 13], 50, [1, 8], [6 - fs, 14], init,
0.0005, "model_" + str(fs) + "_convxy_2_" + str(it))
tf.reset_default_graph()
[fpr, fnr,
score] = test_net([fs, 48, 1], [39, 39, 13], 50, [1, 8], [6 - fs, 14],
init, "model_" + str(fs) + "_convxy_2_" + str(it))
print("convxy: ", str(fs))
print("fpr: ", fpr)
print("fnr: ", fnr)
print("score: ", score)
f.write("convxy: " + str(fs) + "\tscore: " + str(score) + "\tfpr: " +
str(fpr) + "\tfnr: " + str(fnr) + "\n")
f.close()
def search_w(init, fs, w, it):
tf.reset_default_graph()
f = open("scores_width_2", "a")
net_train([4, fs, 1], [w, w, w / 3], 50, [1, 8],
[2, math.floor((1280 - fs) / (13 * 8))], init, 0.0005,
"model_" + str(w) + "_width_2_" + str(it))
tf.reset_default_graph()
[fpr, fnr, score] = test_net([4, fs, 1], [w, w, w / 3], 50, [1, 8],
[2, math.floor((1280 - fs) / (13 * 8))], init,
"model_" + str(w) + "_width_2_" + str(it))
print("width: ", str(w))
print("fpr: ", fpr)
print("fnr: ", fnr)
print("score: ", score)
f.write("width: " + str(w) + "\tscore: " + str(score) + "\tfpr: " +
str(fpr) + "\tfnr: " + str(fnr) + "\n")
f.close()
def optimize_cost(iters, cost, step1, step2, direction):
n_costs = iters
i = 0
tf.reset_default_graph()
while (i < n_costs):
f = open("scores\\scores_cost", "a")
net_train([4, 32, 3], [42, 42, 14], cost, [1, 8],
[2, math.floor(
(1280 - 32) / (13 * 8))], tf.orthogonal_initializer,
0.0005, "model_" + str(cost) + "_cost_demo", 10,
"channels_586202_ordered_batch_3608.mat",
"labels_586202_batch_3608.mat")
tf.reset_default_graph()
[fpr, fnr, frar, nsrar, fsfpr
] = test_net([4, 32, 3], [42, 42, 14], cost, [1, 8],
[2, math.floor(
(1280 - 32) / (13 * 8))], tf.orthogonal_initializer,
"model_" + str(cost) + "_cost_demo", 11, 20,
"channels_586202_ordered_batch_3608.mat",
"labels_586202_batch_3608.mat")
print("cost: ", cost)
print("fpr: ", fpr)
print("fnr: ", fnr)
print("frar: ", frar)
f.write("cost: " + str(cost) + "\tfpr: " + str(fpr) + "\tfnr: " +
str(fnr) + "\tfrar" + str(frar) + "\tnsrar" + str(nsrar) +
" \tfsfpr" + str(fsfpr) + "\n")
f.close()
#if score>=prev:
# direction = direction*-1
# step1=step1-step2
#if step1<=0:
# break
cost = cost + step1 * direction
tf.reset_default_graph()
f.close()
#prev=score
i = i + 1