def base_minimizer(model,
data,
functions,
algo,
max_evals,
trials,
rseed=1337,
full_model_string=None,
notebook_name=None,
verbose=True,
stack=3,
keep_temp=False,
data_args=None):
if full_model_string is not None:
model_str = full_model_string
else:
model_str = get_hyperopt_model_string(model,
data,
functions,
notebook_name,
verbose,
stack,
data_args=data_args)
temp_file = './temp_model.py'
write_temp_files(model_str, temp_file)
if 'temp_model' in sys.modules:
del sys.modules["temp_model"]
try:
from temp_model import keras_fmin_fnct, get_space
except:
print("Unexpected error: {}".format(sys.exc_info()[0]))
raise
try:
if not keep_temp:
os.remove(temp_file)
os.remove(temp_file + 'c')
except OSError:
pass
try:
# for backward compatibility.
return (fmin(keras_fmin_fnct,
space=get_space(),
algo=algo,
max_evals=max_evals,
trials=trials,
rseed=rseed,
return_argmin=True), get_space())
except TypeError:
pass
return (fmin(keras_fmin_fnct,
space=get_space(),
algo=algo,
max_evals=max_evals,
trials=trials,
rstate=np.random.RandomState(rseed),
return_argmin=True), get_space())
def base_minimizer(model, data, algo, max_evals, trials, rseed=1337, full_model_string=None):
if full_model_string is not None:
model_str = full_model_string
else:
model_str = get_hyperopt_model_string(model, data)
write_temp_files(model_str)
try:
from temp_model import keras_fmin_fnct, get_space
except:
print("Unexpected error: {}".format(sys.exc_info()[0]))
raise
try:
os.remove('./temp_model.py')
os.remove('./temp_model.pyc')
except OSError:
pass
try: # for backward compatibility.
best_run = fmin(keras_fmin_fnct,
space=get_space(),
algo=algo,
max_evals=max_evals,
trials=trials,
rseed=rseed)
except TypeError:
best_run = fmin(keras_fmin_fnct,
space=get_space(),
algo=algo,
max_evals=max_evals,
trials=trials,
rstate=np.random.RandomState(rseed))
return best_run
def base_minimizer(model, data, functions, algo, max_evals, trials,
rseed=1337, full_model_string=None, notebook_name=None,
verbose=True, stack=3, keep_temp=False):
if full_model_string is not None:
model_str = full_model_string
else:
model_str = get_hyperopt_model_string(model, data, functions, notebook_name, verbose, stack)
temp_file = './temp_model.py'
write_temp_files(model_str, temp_file)
if 'temp_model' in sys.modules:
del sys.modules["temp_model"]
try:
from temp_model import keras_fmin_fnct, get_space
except:
print("Unexpected error: {}".format(sys.exc_info()[0]))
raise
try:
if not keep_temp:
os.remove(temp_file)
os.remove(temp_file + 'c')
except OSError:
pass
try:
# for backward compatibility.
return (
fmin(keras_fmin_fnct,
space=get_space(),
algo=algo,
max_evals=max_evals,
trials=trials,
rseed=rseed,
return_argmin=True),
get_space()
)
except TypeError:
pass
return (
fmin(keras_fmin_fnct,
space=get_space(),
algo=algo,
max_evals=max_evals,
trials=trials,
rstate=np.random.RandomState(rseed),
return_argmin=True),
get_space()
)
def base_minimizer(model,
data,
algo,
max_evals,
trials,
rseed=1337,
full_model_string=None):
if full_model_string is not None:
model_str = full_model_string
else:
model_str = get_hyperopt_model_string(model, data)
write_temp_files(model_str)
try:
from temp_model import keras_fmin_fnct, get_space
except:
print("Unexpected error: {}".format(sys.exc_info()[0]))
raise
try:
os.remove('./temp_model.py')
os.remove('./temp_model.pyc')
except OSError:
pass
best_run = fmin(keras_fmin_fnct,
space=get_space(),
algo=algo,
max_evals=max_evals,
trials=trials,
rseed=rseed)
return best_run
def base_minimizer(model,
data,
algo,
max_evals,
trials,
rseed=1337,
full_model_string=None,
notebook_name=None,
verbose=True,
stack=3):
if full_model_string is not None:
model_str = full_model_string
else:
model_str = get_hyperopt_model_string(model, data, notebook_name,
verbose, stack)
temp_file = './temp_model.py'
write_temp_files(model_str, temp_file)
try:
from temp_model import keras_fmin_fnct, get_space
except:
print("Unexpected error: {}".format(sys.exc_info()[0]))
raise
try:
os.remove(temp_file)
os.remove(temp_file + 'c')
except OSError:
pass
try: # for backward compatibility.
return fmin(keras_fmin_fnct,
space=get_space(),
algo=algo,
max_evals=max_evals,
trials=trials,
rseed=rseed)
except TypeError:
pass
return fmin(keras_fmin_fnct,
space=get_space(),
algo=algo,
max_evals=max_evals,
trials=trials,
rstate=np.random.RandomState(rseed))
def minimize(model, data, algo, max_evals, trials):
import inspect
import re
model_string = inspect.getsource(model)
data_string = inspect.getsource(data)
parts = re.findall(r"(\w+(?=\s*[\=\(]\s*\{\{[^}]+}\}))", model_string)
part_dict = {}
for i, part in enumerate(parts):
if part in part_dict.keys():
part_dict[part] += 1
parts[i] = part + "_" + str(part_dict[part])
else:
part_dict[part] = 0
aug_parts = []
for (i, part) in enumerate(parts):
aug_parts.append("space['" + part + "']")
hyperopt_params = re.findall(r"(\{\{[^}]+}\})", model_string)
for (i, param) in enumerate(hyperopt_params):
hyperopt_params[i] = re.sub(r"[\{\}]", '', param)
space = "def get_space(): \n return {\n"
for name, param in zip(parts, hyperopt_params):
param = re.sub(r"\(", "('" + name + "', ", param)
space += "'" + name + "': hp." + param + ","
space = space[:-1]
space += "\n}"
print('>>> Hyperas search space:\n')
print(space)
first_line = model_string.split("\n")[0]
model_string = model_string.replace(first_line,
"def keras_fmin_fnct(space):\n")
# model_string = re.sub(r"def \s*\w*\s*\(", "def keras_fmin_fnct(space, ", model_string)
result = re.sub(r"(\{\{[^}]+}\})",
lambda match: aug_parts.pop(0),
model_string,
count=len(parts))
print('>>> Resulting replaced keras model:\n')
print(result)
first_line = data_string.split("\n")[0]
data_string = data_string.replace(first_line, "")
data_string = re.sub(r"return.*", "", data_string)
split_data = data_string.split("\n")
for i, line in enumerate(split_data):
split_data[i] = line.strip() + "\n"
data_string = ''.join(split_data)
print(">>> Data")
print(data_string)
with open('./temp_model.py', 'w') as f:
f.write("from hyperopt import fmin, tpe, hp, STATUS_OK, Trials\n")
f.write(data_string)
f.write("\n\n")
f.write(result)
f.write("\n\n")
f.write(space)
f.close()
import sys
sys.path.append(".")
from temp_model import keras_fmin_fnct, get_space
try:
os.remove('./temp_model.py')
os.remove('./temp_model.pyc')
except OSError:
pass
best = fmin(keras_fmin_fnct,
space=get_space(),
algo=algo,
max_evals=max_evals,
trials=trials)
return best