def create_optimizer(dim, fitness, n_step, n_init_sample, model_type):
C = ContinuousSpace([-5, 5]) * 2
I = OrdinalSpace([-100, 100])
N = NominalSpace(['OK', 'A', 'B', 'C', 'D', 'E'])
search_space = C * I * N
levels = search_space.levels
if model_type == 'GP':
# thetaL = 1e-3 * (ub - lb) * np.ones(dim)
# thetaU = 10 * (ub - lb) * np.ones(dim)
# theta0 = np.random.rand(dim) * (thetaU - thetaL) + thetaL
#
# model = GaussianProcess(regr='constant', corr='matern',
# theta0=theta0, thetaL=thetaL,
# thetaU=thetaU, nugget=1e-5,
# nugget_estim=False, normalize=False,
# verbose=False, random_start = 15*dim,
# random_state=None)
pass
elif model_type == 'sklearn-RF':
min_samples_leaf = max(1, int(n_init_sample / 20.))
max_features = int(np.ceil(dim * 5 / 6.))
model = RandomForest(levels=levels, n_estimators=100,
max_features=max_features,
min_samples_leaf=min_samples_leaf)
elif model_type == 'R-RF':
min_samples_leaf = max(1, int(n_init_sample / 20.))
max_features = int(np.ceil(dim * 5 / 6.))
model = RrandomForest(levels=levels, n_estimators=100,
max_features=max_features,
min_samples_leaf=min_samples_leaf)
opt = mipego(search_space, fitness, model, max_iter=n_step, random_seed=None,
n_init_sample=n_init_sample, minimize=True, optimizer='MIES')
return opt
i += 1
except:
print("error in receiving answer from gpu " + str(gpu_no))
success = False
try:
tuple = (float(tuple_str1), float(tuple_str2), success)
except:
tuple = (0.0, 0.0, False)
#return outputval
return tuple
#define the search space.
objective = obj_func('./all-cnn_bi_msphere.py')
real_space = ContinuousSpace([0.0, 4.0], 'real_space') * 5
integer_space = OrdinalSpace([0, 4], 'integer_space') * 5
discrete_space = NominalSpace(['0', '1', '2', '3', '4'], 'discrete_space') * 5
search_space = real_space * integer_space * discrete_space
print('starting program...')
#available_gpus = gp.getAvailable(limit=2)
available_gpus = gp.getAvailable(limit=5)
#try:
#available_gpus.remove(0)#CHRIS gpu 0 and 5 are differen gpu types on duranium since they are faster, timing will be unreliable, so remove them from list
#except:
#pass
#try:
#available_gpus.remove(5)
#except:
kernel_size = OrdinalSpace([1, 6], 'k') * 7
strides = OrdinalSpace([1, 5], 's') * 3
stack_sizes = OrdinalSpace([1, 5], 'stack') * 3
#TODO_CHRIS these changes are just for cigar test function
#filters = OrdinalSpace([0, 5], 'filters') * 7
#kernel_size = OrdinalSpace([0, 5], 'k') * 7
#strides = OrdinalSpace([0, 5], 's') * 3
#stack_sizes = OrdinalSpace([0, 5], 'stack') * 3
#TODO_CHRIS these changes are just for cigar test function
activation = NominalSpace(activation_fun_conv, "activation") # activation function
activation_dense = NominalSpace(activation_fun, "activ_dense") # activation function for dense layer
step = NominalSpace([True, False], "step") # step
global_pooling = NominalSpace([True, False], "global_pooling") # global_pooling
drop_out = ContinuousSpace([1e-5, .9], 'dropout') * 4 # drop_out rate
lr_rate = ContinuousSpace([1e-4, 1.0e-0], 'lr') # learning rate
l2_regularizer = ContinuousSpace([1e-5, 1e-2], 'l2')# l2_regularizer
#TODO_CHRIS these changes are just for cigar test function
#drop_out = ContinuousSpace([0.0, .9], 'dropout') * 4 # drop_out rate
#lr_rate = ContinuousSpace([0.0, 1.0e-0], 'lr') # learning rate
#l2_regularizer = ContinuousSpace([0.0, 1e-2], 'l2')# l2_regularizer
#TODO_CHRIS these changes are just for cigar test function
search_space = stack_sizes * strides * filters * kernel_size * activation * activation_dense * drop_out * lr_rate * l2_regularizer * step * global_pooling
print('starting program...')
#available_gpus = gp.getAvailable(limit=2)
available_gpus = gp.getAvailable(limit=5)
#try:
eval_n = 0
def obj_func(x, gpu_no=None):
global eval_n
print("Eval # " + str(eval_n) + " (Gpu " + str(gpu_no) + ")")
eval_n += 1
x_r, x_i, x_d = np.array([x['C_0'], x['C_1']]), x['I'], x['N']
if x_d == 'OK':
tmp = 0
else:
tmp = 1
return np.sum(x_r**2.) + abs(x_i - 10) / 123. + tmp * 2.
C = ContinuousSpace([-5, 5], 'C') * 2
I = OrdinalSpace([-100, 100], 'I')
N = NominalSpace(['OK', 'A', 'B', 'C', 'D', 'E'], 'N')
search_space = C * I * N
model = RandomForest(levels=search_space.levels)
# model = RrandomForest(levels=search_space.levels, seed=1, max_features='sqrt')
opt = mipego(
search_space,
obj_func,
model,
max_iter=n_step,
random_seed=None,
n_init_sample=n_init_sample,
def test_skippy():
from mipego.mipego import Solution #TODO remove this, only for testing
from mipego.SearchSpace import ContinuousSpace, NominalSpace, OrdinalSpace
from keras.utils import plot_model
#define the search space.
#objective = obj_func('./all-cnn_bi.py')
activation_fun = ["softmax"]
activation_fun_conv = ["elu", "relu", "tanh", "sigmoid", "selu"]
filters = OrdinalSpace([10, 100],
'filters') * 14 #TODO [0,100] should be [0,600]
kernel_size = OrdinalSpace([1, 8], 'k') * 14
strides = OrdinalSpace([1, 5], 's') * 7
stack_sizes = OrdinalSpace([0, 4],
'stack') * 7 #TODO [0,4] should be [0,7]
activation = NominalSpace(activation_fun_conv,
"activation") # activation function
activation_dense = NominalSpace(
activation_fun, "activ_dense") # activation function for dense layer
step = NominalSpace([True, False], "step") # step
global_pooling = NominalSpace([True, False],
"global_pooling") # global_pooling
#skippy parameters
skints = OrdinalSpace([0, 2**50 - 1], 'skint') * 3 #CHRIS TODO tweak this
skst = OrdinalSpace([2, 10], 'skst') * 3
dense_size = OrdinalSpace([0, 1200], 'dense_size') * 2
no_pooling = NominalSpace([True, False], "no_pooling")
#skippy parameters
drop_out = ContinuousSpace([1e-5, .9], 'dropout') * 8 # drop_out rate
lr_rate = ContinuousSpace([1e-4, 1.0e-0], 'lr') # learning rate
l2_regularizer = ContinuousSpace([1e-5, 1e-2], 'l2') # l2_regularizer
search_space = stack_sizes * strides * filters * kernel_size * activation * activation_dense * drop_out * lr_rate * l2_regularizer * step * global_pooling * skints * skst * dense_size * no_pooling
n_init_sample = 1
samples = search_space.sampling(n_init_sample)
print(samples)
var_names = search_space.var_name.tolist()
print(var_names)
#a sample
#samples = [[1, 1, 1, 1, 2, 3, 10, 10, 5, 10, 10, 10, 10, 3, 4, 2, 1, 3, 1, 3, 'relu', 'softmax', 0.7105013348601977, 0.24225495530708516, 0.5278997344637044, 0.7264822991098491, 0.0072338759099408985, 0.00010867041652507452, False, True]]
#test parameters
#original parameters
#RESnet-34-like
stack_0 = 1
stack_1 = 6
stack_2 = 4
stack_3 = 4
stack_4 = 6
stack_5 = 6
stack_6 = 6
s_0 = 2
s_1 = 2
s_2 = 1
s_3 = 2
s_4 = 1
s_5 = 2
s_6 = 1
filters_0 = 64 * 2
filters_1 = 64 * 2
filters_2 = 64 * 2
filters_3 = 64 * 2
filters_4 = 128 * 2
filters_5 = 128 * 2
filters_6 = 128 * 2
filters_7 = 128 * 2
filters_8 = 256 * 2
filters_9 = 256 * 2
filters_10 = 256 * 2
filters_11 = 256 * 2
filters_12 = 512 * 2
filters_13 = 512 * 2
k_0 = 7
k_1 = 3
k_2 = 3
k_3 = 3
k_4 = 3
k_5 = 3
k_6 = 3
k_7 = 3
k_8 = 3
k_9 = 3
k_10 = 3
k_11 = 3
k_12 = 3
k_13 = 3
activation = 'relu'
activ_dense = 'softmax'
dropout_0 = 0.001
dropout_1 = 0.001
dropout_2 = 0.001
dropout_3 = 0.001
dropout_4 = 0.001
dropout_5 = 0.001
dropout_6 = 0.001
dropout_7 = 0.001
lr = 0.1
l2 = 0.0001
step = True
global_pooling = True
#skippy parameters
om_en_om = 1
ranges = [stack_6, stack_5, stack_4, stack_3, stack_2, stack_1, stack_0]
for w in range(len(ranges)): #TODO testcode: remove
om_en_om = om_en_om << 1
for z in range(ranges[w] // 2):
om_en_om = om_en_om << 2
om_en_om += 1
om_en_om = om_en_om << 1
skint_0 = inv_gray(om_en_om) #3826103921638#2**30-1
skint_1 = 0 #19283461627361826#2**30-1
skint_2 = 0 #473829102637452916#2**30-1
skst_0 = 2
skst_1 = 0
skst_2 = 0
dense_size_0 = 1000 * 2
dense_size_1 = 0
no_pooling = False
#skippy parameters
#assembling parameters
samples = [[
stack_0, stack_1, stack_2, stack_3, stack_4, stack_5, stack_6, s_0,
s_1, s_2, s_3, s_4, s_5, s_6, filters_0, filters_1, filters_2,
filters_3, filters_4, filters_5, filters_6, filters_7, filters_8,
filters_9, filters_10, filters_11, filters_12, filters_13, k_0, k_1,
k_2, k_3, k_4, k_5, k_6, k_7, k_8, k_9, k_10, k_11, k_12, k_13,
activation, activ_dense, dropout_0, dropout_1, dropout_2, dropout_3,
dropout_4, dropout_5, dropout_6, dropout_7, lr, l2, step,
global_pooling, skint_0, skint_1, skint_2, skst_0, skst_1, skst_2,
dense_size_0, dense_size_1, no_pooling
]]
#var_names
#['stack_0', 'stack_1', 'stack_2', 's_0', 's_1', 's_2', 'filters_0', 'filters_1', 'filters_2', 'filters_3', 'filters_4', 'filters_5', 'filters_6', 'k_0', 'k_1', 'k_2', 'k_3', 'k_4', 'k_5', 'k_6', 'activation', 'activ_dense', 'dropout_0', 'dropout_1', 'dropout_2', 'dropout_3', 'lr', 'l2', 'step', 'global_pooling']
X = [
Solution(s, index=k, var_name=var_names) for k, s in enumerate(samples)
]
vla = {
's_0': 3,
'l2': 4.4274387289657325e-05,
'filters_7': 423,
'dense_size_1': 992,
'filters_12': 295,
'stack_0': 0,
'filters_2': 53,
'global_pooling': True,
'dropout_6': 0.5606577615096975,
'filters_13': 115,
'filters_4': 396,
'stack_4': 0,
'k_9': 6,
'activation': 'tanh',
'dropout_1': 0.07267176147234225,
'filters_5': 405,
'filters_1': 250,
'k_7': 7,
'filters_3': 408,
'stack_2': 0,
'no_pooling': False,
'dropout_7': 0.12689965102483852,
's_2': 4,
'filters_8': 455,
'dropout_4': 0.8991002969243431,
'k_11': 5,
'skst_0': 7,
'k_4': 3,
'dropout_3': 0.5966691482903116,
'step': False,
'dense_size_0': 583,
'stack_1': 1,
'k_0': 3,
'skint_1': 505527202345094,
'k_1': 5,
'k_8': 1,
'stack_6': 0,
'lr': 0.6919959357016345,
'activ_dense': 'softmax',
'filters_6': 305,
's_1': 3,
'filters_9': 226,
's_4': 2,
'stack_3': 1,
'skst_1': 5,
'skst_2': 6,
'dropout_2': 0.039087410518674766,
'k_12': 4,
'k_3': 6,
'dropout_5': 0.46057411289276423,
'skint_0': 957176709324259,
'k_5': 4,
'k_2': 3,
's_3': 1,
'filters_0': 195,
'k_6': 1,
'k_13': 2,
'skint_2': 1098454353499063,
'filters_11': 107,
'filters_10': 257,
'k_10': 4,
'stack_5': 0,
's_6': 1,
's_5': 2,
'dropout_0': 0.6293343140822664
}
print(X)
print(X[0].to_dict())
#cfg = [Solution(x, index=len(self.data) + i, var_name=self.var_names) for i, x in enumerate(X)]
test = True
if test:
#model = CNN_conf(X[0].to_dict(),test=test)
model = CNN_conf(vla, test=test)
plot_model(model,
to_file='model_skippy_test.png',
show_shapes=True,
show_layer_names=True)
model.summary()
print(model.count_params())
print(str(model.count_params() * 4 * 2 / 1024 / 1024 / 1024) + ' Gb')
else:
timer, loss = CNN_conf(X[0].to_dict(), test=test)
print('timer, loss:')
print(timer, loss)
def test_skippy():
from mipego.mipego import Solution #TODO remove this, only for testing
from mipego.SearchSpace import ContinuousSpace, NominalSpace, OrdinalSpace
from keras.utils import plot_model
with open('skippy_test_train_hist' + '_eval_train_hist.json', 'w') as f:
f.write('')
#define the search space.
#objective = obj_func('./all-cnn_bi.py')
activation_fun = ["softmax"]
activation_fun_conv = ["elu", "relu", "tanh", "sigmoid", "selu"]
filters = OrdinalSpace([10, 100],
'filters') * 14 #TODO [0,100] should be [0,600]
kernel_size = OrdinalSpace([1, 8], 'k') * 14
strides = OrdinalSpace([1, 5], 's') * 7
stack_sizes = OrdinalSpace([0, 4],
'stack') * 7 #TODO [0,4] should be [0,7]
activation = NominalSpace(activation_fun_conv,
"activation") # activation function
activation_dense = NominalSpace(
activation_fun, "activ_dense") # activation function for dense layer
step = NominalSpace([True, False], "step") # step
global_pooling = NominalSpace([True, False],
"global_pooling") # global_pooling
#skippy parameters
skstart = OrdinalSpace([0, 50], 'skstart') * 5
skstep = OrdinalSpace([1, 50], 'skstep') * 5
max_pooling = NominalSpace([True, False], "max_pooling")
dense_size = OrdinalSpace([0, 2000], 'dense_size') * 2
#skippy parameters
drop_out = ContinuousSpace([1e-5, .9], 'dropout') * 10 # drop_out rate
lr_rate = ContinuousSpace([1e-4, 1.0e-0], 'lr') # learning rate
l2_regularizer = ContinuousSpace([1e-5, 1e-2], 'l2') # l2_regularizer
search_space = stack_sizes * strides * filters * kernel_size * activation * activation_dense * drop_out * lr_rate * l2_regularizer * step * global_pooling * skstart * skstep * max_pooling * dense_size
n_init_sample = 1
samples = search_space.sampling(n_init_sample)
print(samples)
var_names = search_space.var_name.tolist()
print(var_names)
#a sample
#samples = [[1, 1, 1, 1, 2, 3, 10, 10, 5, 10, 10, 10, 10, 3, 4, 2, 1, 3, 1, 3, 'relu', 'softmax', 0.7105013348601977, 0.24225495530708516, 0.5278997344637044, 0.7264822991098491, 0.0072338759099408985, 0.00010867041652507452, False, True]]
#test parameters
#original parameters
#RESnet-34-like
stack_0 = 1
stack_1 = 6
stack_2 = 4
stack_3 = 4
stack_4 = 6
stack_5 = 6
stack_6 = 6
s_0 = 2 #1#2
s_1 = 2
s_2 = 1 #1
s_3 = 2
s_4 = 1
s_5 = 2
s_6 = 1
filters_0 = 64
filters_1 = 64
filters_2 = 64
filters_3 = 64
filters_4 = 128
filters_5 = 128
filters_6 = 128
filters_7 = 128
filters_8 = 256
filters_9 = 256
filters_10 = 256
filters_11 = 256
filters_12 = 512
filters_13 = 512
k_0 = 7
k_1 = 1
k_2 = 3
k_3 = 1
k_4 = 3
k_5 = 1
k_6 = 3
k_7 = 1
k_8 = 3
k_9 = 1
k_10 = 3
k_11 = 1
k_12 = 3
k_13 = 1
activation = 'relu'
activ_dense = 'softmax'
dropout_0 = 0.001
dropout_1 = 0.001
dropout_2 = 0.001
dropout_3 = 0.001
dropout_4 = 0.001
dropout_5 = 0.001
dropout_6 = 0.001
dropout_7 = 0.001
dropout_8 = 0.001
dropout_9 = 0.001
lr = 0.01
l2 = 0.0001
step = False #True
global_pooling = True
#skippy parameters
om_en_om = 1
ranges = [stack_6, stack_5, stack_4, stack_3, stack_2, stack_1, stack_0]
for w in range(len(ranges)): #TODO testcode: remove
om_en_om = om_en_om << 1
for z in range(ranges[w] // 2):
om_en_om = om_en_om << 2
om_en_om += 1
om_en_om = om_en_om << 1
skstart_0 = 1 #inv_gray(om_en_om)#3826103921638#2**30-1
skstart_1 = 1 #19283461627361826#2**30-1
skstart_2 = 1 #473829102637452916#2**30-1
skstart_3 = 1 #473829102637452916#2**30-1
skstart_4 = 1 #473829102637452916#2**30-1
skstep_0 = 2
skstep_1 = 1
skstep_2 = 1
skstep_3 = 1
skstep_4 = 1
max_pooling = True
dense_size_0 = 1000
dense_size_1 = 0
#skippy parameters
#assembling parameters
samples = [[
stack_0, stack_1, stack_2, stack_3, stack_4, stack_5, stack_6, s_0,
s_1, s_2, s_3, s_4, s_5, s_6, filters_0, filters_1, filters_2,
filters_3, filters_4, filters_5, filters_6, filters_7, filters_8,
filters_9, filters_10, filters_11, filters_12, filters_13, k_0, k_1,
k_2, k_3, k_4, k_5, k_6, k_7, k_8, k_9, k_10, k_11, k_12, k_13,
activation, activ_dense, dropout_0, dropout_1, dropout_2, dropout_3,
dropout_4, dropout_5, dropout_6, dropout_7, dropout_8, dropout_9, lr,
l2, step, global_pooling, skstart_0, skstart_1, skstart_2, skstart_3,
skstart_4, skstep_0, skstep_1, skstep_2, skstep_3, skstep_4,
max_pooling, dense_size_0, dense_size_1
]]
#var_names
#['stack_0', 'stack_1', 'stack_2', 's_0', 's_1', 's_2', 'filters_0', 'filters_1', 'filters_2', 'filters_3', 'filters_4', 'filters_5', 'filters_6', 'k_0', 'k_1', 'k_2', 'k_3', 'k_4', 'k_5', 'k_6', 'activation', 'activ_dense', 'dropout_0', 'dropout_1', 'dropout_2', 'dropout_3', 'lr', 'l2', 'step', 'global_pooling']
dropout_mult = 1.0
lr_mult = 1.0
X = [
Solution(s, index=k, var_name=var_names) for k, s in enumerate(samples)
]
vla = {
's_4': 3,
'k_12': 1,
'k_13': 13,
'k_4': 3,
'filters_9': 273,
'stack_2': 5,
'skstep_1': 8,
'stack_4': 2,
's_2': 7,
'filters_8': 463,
's_6': 5,
'dropout_7': 0.14258839346689015 * dropout_mult,
's_5': 8,
'dropout_3': 0.4887239563686235 * dropout_mult,
'k_0': 3,
'filters_13': 506,
'dropout_1': 0.02305687664777915 * dropout_mult,
'stack_5': 6,
'skstart_4': 5,
'dropout_4': 0.2198815770696341 * dropout_mult,
'filters_12': 368,
'k_9': 13,
'dense_size_0': 915,
'max_pooling': True,
'k_8': 1,
'skstart_1': 4,
'k_1': 3,
's_1': 6,
'filters_6': 476,
'dropout_9': 0.237736517209488 * dropout_mult,
'k_3': 2,
'skstart_2': 0,
's_3': 4,
'step': True,
'filters_1': 251,
'stack_3': 7,
'dropout_6': 0.009317366697570491 * dropout_mult,
'filters_5': 199,
'k_10': 10,
'skstart_0': 2,
'filters_4': 239,
'filters_0': 266,
'dense_size_1': 2114,
'lr': 0.0097450688503161 * lr_mult,
'skstep_4': 9,
'dropout_8': 0.06911053842571835 * dropout_mult,
'filters_2': 397,
'filters_3': 341,
'filters_10': 409,
's_0': 3,
'activation': 'elu',
'k_7': 9,
'stack_6': 2,
'skstart_3': 4,
'stack_0': 4,
'k_11': 6,
'k_2': 3,
'l2': 0.0005256770455060354,
'skstep_0': 7,
'skstep_2': 6,
'dropout_2': 0.12188479132476926 * dropout_mult,
'k_5': 13,
'global_pooling': True,
'skstep_3': 2,
'filters_11': 59,
'dropout_0': 0.0010461409934142763,
'k_6': 4,
'stack_1': 0,
'filters_7': 394,
'dropout_5': 0.3355844862089496 * dropout_mult,
'activ_dense': 'softmax'
} #'droput_0': 0.0010461409934142763
print(X)
print(X[0].to_dict())
#cfg = [Solution(x, index=len(self.data) + i, var_name=self.var_names) for i, x in enumerate(X)]
test = True
if test:
model = CNN_conf(X[0].to_dict(), test=test)
#model = CNN_conf(vla,test=test)
plot_model(model,
to_file='model_skippy_test.png',
show_shapes=True,
show_layer_names=True)
model.summary()
print(model.count_params())
print(str(model.count_params() * 4 * 2 / 1024 / 1024 / 1024) + ' Gb')
else:
#timer, loss = CNN_conf(X[0].to_dict(),test=test,epochs= 2000,verbose=1)
timer, loss = CNN_conf(vla, test=test, epochs=200, verbose=1)
#timer, loss = CNN_conf(vla,test=test,epochs= 200,verbose=1,data_augmentation=True,use_validation=True) #TODO use this for data augmentation and make sure the val set is used for val accuracy, not the test set
print('timer, loss:')
print(timer, loss)
def test_skippy():
from mipego.mipego import Solution #TODO remove this, only for testing
from mipego.SearchSpace import ContinuousSpace, NominalSpace, OrdinalSpace
from keras.utils import plot_model
#define the search space.
#objective = obj_func('./all-cnn_bi.py')
activation_fun = ["softmax"]
activation_fun_conv = ["elu","relu","tanh","sigmoid","selu"]
filters = OrdinalSpace([10, 100], 'filters') * 14 #TODO [0,100] should be [0,600]
kernel_size = OrdinalSpace([1, 8], 'k') * 14
strides = OrdinalSpace([1, 5], 's') * 7
stack_sizes = OrdinalSpace([0, 4], 'stack') * 7 #TODO [0,4] should be [0,7]
activation = NominalSpace(activation_fun_conv, "activation") # activation function
activation_dense = NominalSpace(activation_fun, "activ_dense") # activation function for dense layer
step = NominalSpace([True, False], "step") # step
global_pooling = NominalSpace([True, False], "global_pooling") # global_pooling
#skippy parameters
skstart = OrdinalSpace([0, 50], 'skstart') * 5
skstep = OrdinalSpace([1, 50], 'skstep') * 5
max_pooling = NominalSpace([True, False], "max_pooling")
dense_size = OrdinalSpace([0,2000],'dense_size')*2
#skippy parameters
drop_out = ContinuousSpace([1e-5, .9], 'dropout') * 10 # drop_out rate
lr_rate = ContinuousSpace([1e-4, 1.0e-0], 'lr') # learning rate
l2_regularizer = ContinuousSpace([1e-5, 1e-2], 'l2')# l2_regularizer
search_space = stack_sizes * strides * filters * kernel_size * activation * activation_dense * drop_out * lr_rate * l2_regularizer * step * global_pooling * skstart * skstep * max_pooling * dense_size
n_init_sample = 1
samples = search_space.sampling(n_init_sample)
print(samples)
var_names = search_space.var_name.tolist()
print(var_names)
#a sample
#samples = [[1, 1, 1, 1, 2, 3, 10, 10, 5, 10, 10, 10, 10, 3, 4, 2, 1, 3, 1, 3, 'relu', 'softmax', 0.7105013348601977, 0.24225495530708516, 0.5278997344637044, 0.7264822991098491, 0.0072338759099408985, 0.00010867041652507452, False, True]]
#test parameters
#original parameters
#RESnet-34-like
stack_0 = 1
stack_1 = 6
stack_2 = 4
stack_3 = 4
stack_4 = 6
stack_5 = 6
stack_6 = 6
s_0=2#1#2
s_1=2
s_2=1#1
s_3=2
s_4=1
s_5=2
s_6=1
filters_0=64
filters_1=64
filters_2=64
filters_3=64
filters_4=128
filters_5=128
filters_6=128
filters_7=128
filters_8=256
filters_9=256
filters_10=256
filters_11=256
filters_12=512
filters_13=512
k_0=7
k_1=1
k_2=3
k_3=1
k_4=3
k_5=1
k_6=3
k_7=1
k_8=3
k_9=1
k_10=3
k_11=1
k_12=3
k_13=1
activation='relu'
activ_dense='softmax'
dropout_0=0.001
dropout_1=0.001
dropout_2=0.001
dropout_3=0.001
dropout_4=0.001
dropout_5=0.001
dropout_6=0.001
dropout_7=0.001
dropout_8=0.001
dropout_9=0.001
lr=0.01
l2=0.0001
step=False#True
global_pooling=True
#skippy parameters
om_en_om = 1
ranges = [stack_6,stack_5,stack_4,stack_3,stack_2,stack_1,stack_0]
for w in range(len(ranges)):#TODO testcode: remove
om_en_om = om_en_om << 1
for z in range(ranges[w]//2):
om_en_om = om_en_om << 2
om_en_om += 1
om_en_om = om_en_om << 1
skstart_0 = 1#inv_gray(om_en_om)#3826103921638#2**30-1
skstart_1 = 1#19283461627361826#2**30-1
skstart_2 = 1#473829102637452916#2**30-1
skstart_3 = 1#473829102637452916#2**30-1
skstart_4 = 1#473829102637452916#2**30-1
skstep_0 = 2
skstep_1 = 1
skstep_2 = 1
skstep_3 = 1
skstep_4 = 1
max_pooling = True
dense_size_0 = 1000
dense_size_1 = 0
#skippy parameters
#assembling parameters
samples = [[stack_0, stack_1, stack_2, stack_3, stack_4, stack_5, stack_6, s_0, s_1, s_2, s_3, s_4, s_5, s_6, filters_0, filters_1, filters_2, filters_3, filters_4, filters_5, filters_6, filters_7, filters_8, filters_9, filters_10, filters_11, filters_12, filters_13,k_0, k_1, k_2, k_3, k_4, k_5, k_6, k_7, k_8, k_9, k_10, k_11, k_12, k_13, activation, activ_dense, dropout_0, dropout_1, dropout_2, dropout_3, dropout_4, dropout_5, dropout_6, dropout_7, dropout_8, dropout_9, lr, l2, step, global_pooling, skstart_0, skstart_1, skstart_2, skstart_3, skstart_4, skstep_0, skstep_1, skstep_2, skstep_3, skstep_4, max_pooling, dense_size_0, dense_size_1]]
#var_names
#['stack_0', 'stack_1', 'stack_2', 's_0', 's_1', 's_2', 'filters_0', 'filters_1', 'filters_2', 'filters_3', 'filters_4', 'filters_5', 'filters_6', 'k_0', 'k_1', 'k_2', 'k_3', 'k_4', 'k_5', 'k_6', 'activation', 'activ_dense', 'dropout_0', 'dropout_1', 'dropout_2', 'dropout_3', 'lr', 'l2', 'step', 'global_pooling']
X = [Solution(s, index=k, var_name=var_names) for k, s in enumerate(samples)]
vla = {'s_2': 7, 'lr': 0.005478541674651396, 'skstep_2': 4, 'dropout_8': 0.5440199827441856, 'k_12': 15, 'activ_dense': 'softmax', 'stack_4': 3, 'k_5': 2, 'dropout_4': 0.24617655948523018, 's_3': 6, 'k_11': 13, 'filters_10': 84, 'dropout_0': 0.0639815161048702, 'k_7': 13, 'filters_9': 178, 'k_1': 13, 'dropout_6': 0.1752239013431692, 'filters_7': 353, 'skstep_4': 6, 'skstart_2': 0, 'stack_0': 0, 'stack_5': 1, 's_5': 2, 'k_13': 6, 'filters_2': 110, 'filters_0': 248, 'skstart_1': 5, 'filters_6': 341, 'filters_8': 165, 'skstart_4': 2, 'l2': 0.0012874308061650037, 's_0': 9, 'global_pooling': False, 'stack_6': 1, 's_1': 2, 'skstep_0': 4, 'dropout_3': 0.495646008202597, 'skstart_0': 3, 'k_6': 2, 'filters_1': 61, 'dropout_2': 0.028121315386701783, 'stack_3': 2, 'filters_3': 299, 'stack_1': 3, 'max_pooling': True, 'filters_4': 259, 'filters_11': 207, 'k_3': 15, 'k_0': 15, 'dense_size_0': 1400, 'k_4': 10, 's_6': 5, 'dropout_9': 0.004273458743956573, 'skstep_3': 6, 'filters_5': 16, 's_4': 2, 'dropout_1': 0.42526328646019135, 'dense_size_1': 2990, 'k_10': 9, 'k_2': 4, 'skstep_1': 6, 'dropout_5': 0.3927105783290164, 'filters_12': 283, 'dropout_7': 0.01357058138235737, 'activation': 'selu', 'filters_13': 228, 'step': False, 'k_8': 2, 'k_9': 2, 'skstart_3': 1, 'stack_2': 3}
print(X)
print(X[0].to_dict())
#cfg = [Solution(x, index=len(self.data) + i, var_name=self.var_names) for i, x in enumerate(X)]
test = False
if test:
#model = CNN_conf(X[0].to_dict(),test=test)
model = CNN_conf(vla,test=test)
plot_model(model, to_file='model_skippy_test.png',show_shapes=True,show_layer_names=True)
model.summary()
print(model.count_params())
print(str(model.count_params() * 4 * 2 / 1024/1024/1024) + ' Gb')
else:
#timer, loss = CNN_conf(X[0].to_dict(),test=test,epochs= 2000,verbose=1)
timer, loss = CNN_conf(vla,test=test,epochs= 2000,verbose=1)
print('timer, loss:')
print(timer, loss)
tuple_str1 = '800000' #CHRIS 2 times timeout value
tuple_str2 = str(
-1 *
math.log(0.05)) #CHRIS half the accuracy of random guessing
tuple = (float(tuple_str1), float(tuple_str2), success)
#return outputval
with open(self.save_name + '_thread_log.json', 'a') as outfile:
outfile.write('thread ' + str(gpu_no) +
': step 3 gpu 3 obj_func 9\n')
return tuple
#define the search space.
save_name = '../../../data/s0315435/data_skippy_cifar10_better_data_augmentation_train_tweak_big_one_restarted1'
objective = obj_func('./all_cnn_bi_skippy_aug_tr_tw.py', save_name=save_name)
lr_rate = ContinuousSpace([1e-4, 2 * 0.003521543292982737], 'lr')
drop = ContinuousSpace([0.1, 0.95], 'drop')
epochs_drop = OrdinalSpace([1, 40], 'epochs_drop')
momentum = ContinuousSpace([0.8, 0.99], 'momentum')
optimizer = NominalSpace(
["SGD", "RMSprop", "Adagrad", "Adadelta", "Adam", "Adamax", "Nadam"],
'optimizer')
rho = ContinuousSpace([0.8, 0.99], 'rho')
search_space = lr_rate * drop * epochs_drop * momentum * optimizer * rho
print('starting program...')
#available_gpus = gp.getAvailable(limit=2)
gpu_limit = 16
available_gpus = gp.getAvailable(limit=gpu_limit)
activation_dense = NominalSpace(
activation_fun, "activ_dense") # activation function for dense layer
step = NominalSpace([True, False], "step") # step
global_pooling = NominalSpace(
[True, False], "global_pooling") # global_pooling#CHRIS TODO removed False
#skippy parameters
skstart = OrdinalSpace([0, 7], 'skstart') * 5
skstep = OrdinalSpace(
[1, 10], 'skstep'
) * 5 #CHRIS a skip step of 1 means no skip connection#OrdinalSpace([1, 10], 'skst') * 3
max_pooling = NominalSpace([True, False], "max_pooling")
dense_size = OrdinalSpace([0, 4000], 'dense_size') * 2 #CHRIS tweaked
#skippy parameters
drop_out = ContinuousSpace([1e-5, .9], 'dropout') * 10 # drop_out rate
lr_rate = ContinuousSpace([1e-4, 1.0e-2], 'lr') # learning rate#CHRIS tweaked
l2_regularizer = ContinuousSpace([1e-5, 1e-2], 'l2') # l2_regularizer
search_space = stack_sizes * strides * filters * kernel_size * activation * activation_dense * drop_out * lr_rate * l2_regularizer * step * global_pooling * skstart * skstep * max_pooling * dense_size
print('starting program...')
#available_gpus = gp.getAvailable(limit=2)
gpu_limit = 16
available_gpus = gp.getAvailable(limit=gpu_limit)
ignore_gpu = []
if len(sys.argv) > 1:
for i in range(1, int(len(sys.argv))):
print(int(sys.argv[i]))
ignore_gpu.append(int(sys.argv[i]))
#import our package, the surrogate model and the search space classes
from mipego import ParallelBO
from mipego.Surrogate import RandomForest
from mipego.SearchSpace import ContinuousSpace, NominalSpace, OrdinalSpace
# Load the dataset
iris = load_iris()
X_iris = iris.data
y_iris = iris.target
# First we need to define the Search Space
# the search space consists of one continues variable
# one ordinal (integer) variable
# and two categorical (nominal) variables.
Cvar = ContinuousSpace([1.0, 20.0], 'C') # one integer variable with label C
degree = OrdinalSpace([2, 6], 'degree')
gamma = NominalSpace(['scale', 'auto'], 'gamma')
kernel = NominalSpace(['linear', 'poly', 'rbf', 'sigmoid'], 'kernel')
#the complete search space is just the sum of the parameter spaces
search_space = Cvar + gamma + degree + kernel
#now we define the objective function (the model optimization)
def train_model(c):
#define the model
# We will use a Support Vector Classifier
svm = SVC(kernel=c['kernel'],
gamma=c['gamma'],
C=c['C'],
def test_skippy():
from mipego.mipego import Solution #TODO remove this, only for testing
from mipego.SearchSpace import ContinuousSpace, NominalSpace, OrdinalSpace
from keras.utils import plot_model
with open('skippy_test_train_hist' + '_eval_train_hist.json', 'w') as f:
f.write('')
#define the search space.
#objective = obj_func('./all-cnn_bi.py')
activation_fun = ["softmax"]
activation_fun_conv = ["elu","relu","tanh","sigmoid","selu"]
filters = OrdinalSpace([10, 100], 'filters') * 14 #TODO [0,100] should be [0,600]
kernel_size = OrdinalSpace([1, 8], 'k') * 14
strides = OrdinalSpace([1, 5], 's') * 7
stack_sizes = OrdinalSpace([0, 4], 'stack') * 7 #TODO [0,4] should be [0,7]
activation = NominalSpace(activation_fun_conv, "activation") # activation function
activation_dense = NominalSpace(activation_fun, "activ_dense") # activation function for dense layer
step = NominalSpace([True, False], "step") # step
global_pooling = NominalSpace([True, False], "global_pooling") # global_pooling
#skippy parameters
skstart = OrdinalSpace([0, 50], 'skstart') * 5
skstep = OrdinalSpace([1, 50], 'skstep') * 5
max_pooling = NominalSpace([True, False], "max_pooling")
dense_size = OrdinalSpace([0,2000],'dense_size')*2
#skippy parameters
drop_out = ContinuousSpace([1e-5, .9], 'dropout') * 10 # drop_out rate
lr_rate = ContinuousSpace([1e-4, 1.0e-0], 'lr') # learning rate
l2_regularizer = ContinuousSpace([1e-5, 1e-2], 'l2')# l2_regularizer
search_space = stack_sizes * strides * filters * kernel_size * activation * activation_dense * drop_out * lr_rate * l2_regularizer * step * global_pooling * skstart * skstep * max_pooling * dense_size
n_init_sample = 1
samples = search_space.sampling(n_init_sample)
print(samples)
var_names = search_space.var_name.tolist()
print(var_names)
#a sample
#samples = [[1, 1, 1, 1, 2, 3, 10, 10, 5, 10, 10, 10, 10, 3, 4, 2, 1, 3, 1, 3, 'relu', 'softmax', 0.7105013348601977, 0.24225495530708516, 0.5278997344637044, 0.7264822991098491, 0.0072338759099408985, 0.00010867041652507452, False, True]]
#test parameters
#original parameters
#RESnet-34-like
stack_0 = 1
stack_1 = 6
stack_2 = 4
stack_3 = 4
stack_4 = 6
stack_5 = 6
stack_6 = 6
s_0=2#1#2
s_1=2
s_2=1#1
s_3=2
s_4=1
s_5=2
s_6=1
filters_0=64
filters_1=64
filters_2=64
filters_3=64
filters_4=128
filters_5=128
filters_6=128
filters_7=128
filters_8=256
filters_9=256
filters_10=256
filters_11=256
filters_12=512
filters_13=512
k_0=7
k_1=1
k_2=3
k_3=1
k_4=3
k_5=1
k_6=3
k_7=1
k_8=3
k_9=1
k_10=3
k_11=1
k_12=3
k_13=1
activation='relu'
activ_dense='softmax'
dropout_0=0.001
dropout_1=0.001
dropout_2=0.001
dropout_3=0.001
dropout_4=0.001
dropout_5=0.001
dropout_6=0.001
dropout_7=0.001
dropout_8=0.001
dropout_9=0.001
lr=0.0097450688503161#0.01
l2=0.0005256770455060354#0.0001
step=True
global_pooling=True
#skippy parameters
om_en_om = 1
ranges = [stack_6,stack_5,stack_4,stack_3,stack_2,stack_1,stack_0]
for w in range(len(ranges)):#TODO testcode: remove
om_en_om = om_en_om << 1
for z in range(ranges[w]//2):
om_en_om = om_en_om << 2
om_en_om += 1
om_en_om = om_en_om << 1
skstart_0 = 1#inv_gray(om_en_om)#3826103921638#2**30-1
skstart_1 = 1#19283461627361826#2**30-1
skstart_2 = 1#473829102637452916#2**30-1
skstart_3 = 1#473829102637452916#2**30-1
skstart_4 = 1#473829102637452916#2**30-1
skstep_0 = 2
skstep_1 = 1
skstep_2 = 1
skstep_3 = 1
skstep_4 = 1
max_pooling = True
dense_size_0 = 1000
dense_size_1 = 0
#skippy parameters
#assembling parameters
samples = [[stack_0, stack_1, stack_2, stack_3, stack_4, stack_5, stack_6, s_0, s_1, s_2, s_3, s_4, s_5, s_6, filters_0, filters_1, filters_2, filters_3, filters_4, filters_5, filters_6, filters_7, filters_8, filters_9, filters_10, filters_11, filters_12, filters_13,k_0, k_1, k_2, k_3, k_4, k_5, k_6, k_7, k_8, k_9, k_10, k_11, k_12, k_13, activation, activ_dense, dropout_0, dropout_1, dropout_2, dropout_3, dropout_4, dropout_5, dropout_6, dropout_7, dropout_8, dropout_9, lr, l2, step, global_pooling, skstart_0, skstart_1, skstart_2, skstart_3, skstart_4, skstep_0, skstep_1, skstep_2, skstep_3, skstep_4, max_pooling, dense_size_0, dense_size_1]]
#var_names
#['stack_0', 'stack_1', 'stack_2', 's_0', 's_1', 's_2', 'filters_0', 'filters_1', 'filters_2', 'filters_3', 'filters_4', 'filters_5', 'filters_6', 'k_0', 'k_1', 'k_2', 'k_3', 'k_4', 'k_5', 'k_6', 'activation', 'activ_dense', 'dropout_0', 'dropout_1', 'dropout_2', 'dropout_3', 'lr', 'l2', 'step', 'global_pooling']
X = [Solution(s, index=k, var_name=var_names) for k, s in enumerate(samples)]
vla = {'stack_4': 5, 'k_9': 7, 'filters_8': 271, 'lr': 0.004160465185980011, 'filters_11': 502, 'skstep_2': 7, 'filters_4': 137, 'dropout_6': 0.8661297289998243, 'dropout_1': 0.4391661461074476, 'k_13': 1, 'stack_5': 0, 'dense_size_0': 447, 'filters_5': 595, 'k_2': 2, 's_4': 3, 'dropout_0': 0.2844338183563386, 'filters_0': 80, 'filters_7': 573, 's_2': 3, 'dropout_3': 0.02013305610010194, 'k_3': 6, 'dropout_5': 0.5555388378722113, 'k_5': 4, 'skstart_2': 6, 'dropout_4': 0.43662540748596945, 'dropout_2': 0.11810545604797541, 'stack_3': 5, 'skstep_1': 1, 'dense_size_1': 1411, 'k_11': 6, 'filters_1': 583, 'dropout_9': 0.651891663204505, 'skstep_4': 2, 'dropout_7': 0.25151371341217515, 'skstep_0': 3, 'skstart_0': 6, 'k_8': 1, 's_5': 4, 'filters_3': 452, 'skstep_3': 3, 'max_pooling': False, 'filters_2': 11, 'global_pooling': False, 'l2': 0.00036819381287011194, 'dropout_8': 0.4454841154679102, 'skstart_3': 4, 'skstart_4': 6, 'stack_6': 0, 'k_1': 2, 'stack_0': 2, 'skstart_1': 6, 's_6': 3, 'k_4': 4, 'filters_6': 240, 'filters_12': 43, 'filters_10': 220, 'step': True, 'stack_2': 3, 'k_6': 5, 'k_7': 2, 'stack_1': 3, 's_0': 3, 'k_12': 1, 'filters_13': 383, 'k_10': 2, 's_3': 2, 'k_0': 4, 's_1': 4, 'activ_dense': 'softmax', 'filters_9': 373, 'activation': 'tanh'}
print(X)
print(X[0].to_dict())
#cfg = [Solution(x, index=len(self.data) + i, var_name=self.var_names) for i, x in enumerate(X)]
test = True
if test:
model = CNN_conf(X[0].to_dict(),test=test)
#model = CNN_conf(vla,test=test)
plot_model(model, to_file='model_skippy_test.png',show_shapes=True,show_layer_names=True)
model.summary()
print(model.count_params())
print(str(model.count_params() * 4 * 2 / 1024/1024/1024) + ' Gb')
else:
timer, loss = CNN_conf(X[0].to_dict(),test=test,epochs= 200,verbose=1)
#timer, loss = CNN_conf(vla,test=test,epochs= 2000,verbose=1)
print('timer, loss:')
print(timer, loss)
activation_fun, "activ_dense") # activation function for dense layer
step = NominalSpace([True, False], "step") # step
global_pooling = NominalSpace(
[True, False], "global_pooling") # global_pooling#CHRIS TODO removed False
#skippy parameters
skstart = OrdinalSpace([0, 7], 'skstart') * 5
skstep = OrdinalSpace(
[1, 10], 'skstep'
) * 5 #CHRIS a skip step of 1 means no skip connection#OrdinalSpace([1, 10], 'skst') * 3
max_pooling = NominalSpace([True, False], "max_pooling")
dense_size = OrdinalSpace([0, 4000], 'dense_size') * 2 #CHRIS tweaked
#skippy parameters
drop_out = ContinuousSpace(
[0.0, .42], 'dropout'
) * 10 # drop_out rate #tweaked again, min used to be 1e-5 max used to be 0.9
lr_rate = ContinuousSpace(
[1e-4, 1.0e-2], 'lr'
) # learning rate#CHRIS tweaked #CHRIS tweaked again for data augmentation (max used to be 1.0e-2)
l2_regularizer = ContinuousSpace([1e-5, 1e-2], 'l2') # l2_regularizer
batch_size_sp = OrdinalSpace(
[50, 200], 'batch_size_sp') #CHRIS tweaked again: added to search space
#augmented parameters
featurewise_center = NominalSpace([False], "featurewise_center")
samplewise_center = NominalSpace([False], "samplewise_center")
featurewise_std_normalization = NominalSpace([False],
"featurewise_std_normalization")
samplewise_std_normalization = NominalSpace([False],
def test_skippy():
from mipego.mipego import Solution #TODO remove this, only for testing
from mipego.SearchSpace import ContinuousSpace, NominalSpace, OrdinalSpace
from keras.utils import plot_model
with open('skippy_test_train_hist_aug' + '_eval_train_hist.json',
'w') as f:
f.write('')
#define the search space.
#objective = obj_func('./all-cnn_bi.py')
activation_fun = ["softmax"]
activation_fun_conv = ["elu", "relu", "tanh", "sigmoid", "selu"]
filters = OrdinalSpace([10, 100],
'filters') * 14 #TODO [0,100] should be [0,600]
kernel_size = OrdinalSpace([1, 8], 'k') * 14
strides = OrdinalSpace([1, 5], 's') * 7
stack_sizes = OrdinalSpace([0, 4],
'stack') * 7 #TODO [0,4] should be [0,7]
activation = NominalSpace(activation_fun_conv,
"activation") # activation function
activation_dense = NominalSpace(
activation_fun, "activ_dense") # activation function for dense layer
step = NominalSpace([True, False], "step") # step
global_pooling = NominalSpace([True, False],
"global_pooling") # global_pooling
#skippy parameters
skstart = OrdinalSpace([0, 50], 'skstart') * 5
skstep = OrdinalSpace([1, 50], 'skstep') * 5
max_pooling = NominalSpace([True, False], "max_pooling")
dense_size = OrdinalSpace([0, 2000], 'dense_size') * 2
#skippy parameters
drop_out = ContinuousSpace([1e-5, .9], 'dropout') * 10 # drop_out rate
lr_rate = ContinuousSpace([1e-4, 1.0e-0], 'lr') # learning rate
l2_regularizer = ContinuousSpace([1e-5, 1e-2], 'l2') # l2_regularizer
search_space = stack_sizes * strides * filters * kernel_size * activation * activation_dense * drop_out * lr_rate * l2_regularizer * step * global_pooling * skstart * skstep * max_pooling * dense_size
n_init_sample = 1
samples = search_space.sampling(n_init_sample)
print(samples)
var_names = search_space.var_name.tolist()
print(var_names)
#a sample
#samples = [[1, 1, 1, 1, 2, 3, 10, 10, 5, 10, 10, 10, 10, 3, 4, 2, 1, 3, 1, 3, 'relu', 'softmax', 0.7105013348601977, 0.24225495530708516, 0.5278997344637044, 0.7264822991098491, 0.0072338759099408985, 0.00010867041652507452, False, True]]
#test parameters
#original parameters
#RESnet-34-like
stack_0 = 1
stack_1 = 6
stack_2 = 4
stack_3 = 4
stack_4 = 6
stack_5 = 6
stack_6 = 6
s_0 = 2 #1#2
s_1 = 2
s_2 = 1 #1
s_3 = 2
s_4 = 1
s_5 = 2
s_6 = 1
filters_0 = 64
filters_1 = 64
filters_2 = 64
filters_3 = 64
filters_4 = 128
filters_5 = 128
filters_6 = 128
filters_7 = 128
filters_8 = 256
filters_9 = 256
filters_10 = 256
filters_11 = 256
filters_12 = 512
filters_13 = 512
k_0 = 7
k_1 = 1
k_2 = 3
k_3 = 1
k_4 = 3
k_5 = 1
k_6 = 3
k_7 = 1
k_8 = 3
k_9 = 1
k_10 = 3
k_11 = 1
k_12 = 3
k_13 = 1
activation = 'relu'
activ_dense = 'softmax'
dropout_0 = 0.001
dropout_1 = 0.001
dropout_2 = 0.001
dropout_3 = 0.001
dropout_4 = 0.001
dropout_5 = 0.001
dropout_6 = 0.001
dropout_7 = 0.001
dropout_8 = 0.001
dropout_9 = 0.001
lr = 0.01
l2 = 0.0001
step = False #True
global_pooling = True
#skippy parameters
om_en_om = 1
ranges = [stack_6, stack_5, stack_4, stack_3, stack_2, stack_1, stack_0]
for w in range(len(ranges)): #TODO testcode: remove
om_en_om = om_en_om << 1
for z in range(ranges[w] // 2):
om_en_om = om_en_om << 2
om_en_om += 1
om_en_om = om_en_om << 1
skstart_0 = 1 #inv_gray(om_en_om)#3826103921638#2**30-1
skstart_1 = 1 #19283461627361826#2**30-1
skstart_2 = 1 #473829102637452916#2**30-1
skstart_3 = 1 #473829102637452916#2**30-1
skstart_4 = 1 #473829102637452916#2**30-1
skstep_0 = 2
skstep_1 = 1
skstep_2 = 1
skstep_3 = 1
skstep_4 = 1
max_pooling = True
dense_size_0 = 1000
dense_size_1 = 0
#skippy parameters
#assembling parameters
samples = [[
stack_0, stack_1, stack_2, stack_3, stack_4, stack_5, stack_6, s_0,
s_1, s_2, s_3, s_4, s_5, s_6, filters_0, filters_1, filters_2,
filters_3, filters_4, filters_5, filters_6, filters_7, filters_8,
filters_9, filters_10, filters_11, filters_12, filters_13, k_0, k_1,
k_2, k_3, k_4, k_5, k_6, k_7, k_8, k_9, k_10, k_11, k_12, k_13,
activation, activ_dense, dropout_0, dropout_1, dropout_2, dropout_3,
dropout_4, dropout_5, dropout_6, dropout_7, dropout_8, dropout_9, lr,
l2, step, global_pooling, skstart_0, skstart_1, skstart_2, skstart_3,
skstart_4, skstep_0, skstep_1, skstep_2, skstep_3, skstep_4,
max_pooling, dense_size_0, dense_size_1
]]
#var_names
#['stack_0', 'stack_1', 'stack_2', 's_0', 's_1', 's_2', 'filters_0', 'filters_1', 'filters_2', 'filters_3', 'filters_4', 'filters_5', 'filters_6', 'k_0', 'k_1', 'k_2', 'k_3', 'k_4', 'k_5', 'k_6', 'activation', 'activ_dense', 'dropout_0', 'dropout_1', 'dropout_2', 'dropout_3', 'lr', 'l2', 'step', 'global_pooling']
dropout_mult = 1.0
lr_mult = 1.0
X = [
Solution(s, index=k, var_name=var_names) for k, s in enumerate(samples)
]
vla = {
'k_6': 8,
'dropout_3': 0.11567654065541401,
'stack_2': 2,
'skstep_3': 2,
'skstart_2': 6,
'k_11': 9,
'zca_whitening': False,
'k_1': 11,
's_4': 3,
'filters_4': 191,
'k_7': 9,
'dropout_6': 0.19656398582242512,
'fill_mode': 'nearest',
'filters_0': 246,
'lr': 0.003521543292982737,
'skstart_4': 1,
's_3': 3,
'height_shift_range': 0.5512549395731117,
'dropout_9': 0.282536761776477,
'dense_size_0': 1344,
'filters_11': 507,
's_0': 2,
'dropout_4': 0.025329970168830974,
'filters_10': 305,
'filters_12': 474,
'dropout_8': 0.10220859898802954,
'samplewise_std_normalization': False,
'cval': 0.24779638415638786,
'step': False,
'skstep_0': 2,
'skstart_3': 3,
'featurewise_std_normalization': False,
's_5': 3,
'skstep_1': 8,
'k_4': 14,
'stack_0': 3,
'max_pooling': True,
'dropout_0': 0.005004155479145995,
'batch_size_sp': 75,
'skstart_1': 0,
'skstep_2': 2,
'filters_6': 535,
'k_12': 4,
'stack_5': 0,
'horizontal_flip': True,
'filters_2': 397,
'stack_4': 5,
'l2': 0.00043366714416766863,
'skstart_0': 6,
'filters_7': 202,
'filters_13': 350,
'k_2': 4,
'k_3': 4,
's_2': 3,
's_6': 3,
'rotation_range': 31,
'shear_range': 4.413108635288765,
'filters_5': 109,
's_1': 1,
'k_8': 9,
'k_9': 5,
'channel_shift_range': 0.002134671459292783,
'samplewise_center': False,
'k_0': 2,
'dropout_5': 0.09773198911653828,
'vertical_flip': False,
'k_5': 10,
'zoom_range': 0.02446592218470434,
'width_shift_range': 0.11326574574565945,
'stack_6': 0,
'k_10': 10,
'dropout_2': 0.3496803660826153,
'activation': 'selu',
'stack_3': 1,
'k_13': 4,
'zca_epsilon': 1.2393513955305375e-06,
'filters_3': 473,
'dense_size_1': 1216,
'stack_1': 1,
'dropout_1': 0.16597601970646955,
'filters_8': 353,
'dropout_7': 0.2567508735733037,
'featurewise_center': False,
'filters_9': 339,
'global_pooling': True,
'skstep_4': 1,
'activ_dense': 'softmax',
'filters_1': 120
}
print(X)
print(X[0].to_dict())
#cfg = [Solution(x, index=len(self.data) + i, var_name=self.var_names) for i, x in enumerate(X)]
test = False
if test:
#model = CNN_conf(X[0].to_dict(),test=test)
model = CNN_conf(vla, test=test)
plot_model(model,
to_file='model_skippy_test.png',
show_shapes=True,
show_layer_names=True)
model.summary()
print(model.count_params())
print(str(model.count_params() * 4 * 2 / 1024 / 1024 / 1024) + ' Gb')
else:
#timer, loss = CNN_conf(X[0].to_dict(),test=test,epochs= 2000,verbose=1)
#timer, loss = CNN_conf(vla,test=test,epochs= 200,verbose=1)
timer, loss = CNN_conf(
vla,
test=test,
epochs=200,
verbose=1,
data_augmentation=True,
use_validation=True,
test_on_validation=True
) #TODO use this for data augmentation and make sure the val set is used for val accuracy, not the test set
print('timer, loss:')
print(timer, loss)