for testing and deploying your application
for finding and fixing issues
for empowering human code reviews
# Author: Simon Blanke
# Email: [email protected]
# License: MIT License
import dill
import random
import numpy as np
class SearchSpace:
def __init__(self, _core_, model_nr):
self.search_space = _core_.search_config[list(_core_.search_config)[model_nr]]
self.pos_space_limit()
self.init_type = None
self.para_names = list(self.search_space.keys())
if _core_.init_config:
self.init_para = _core_.init_config[list(_core_.init_config)[model_nr]]
if list(self.init_para.keys())[0] == list(self.search_space.keys())[0]:
self.init_type = "warm_start"
elif list(self.init_para.keys())[0] == "scatter_init":
self.init_type = "scatter_init"
def pos_space_limit(self):
dim = []
for pos_key in self.search_space:
dim.append(len(self.search_space[pos_key]) - 1)
self.dim = np.array(dim)
def get_random_pos(self):
pos_new = np.random.uniform(np.zeros(self.dim.shape), self.dim, self.dim.shape)
pos = np.rint(pos_new).astype(int)
return pos
def get_random_pos_scalar(self, hyperpara_name):
n_para_values = len(self.search_space[hyperpara_name])
pos = random.randint(0, n_para_values - 1)
def pos2para(self, pos):
if len(self.search_space.keys()) == pos.size:
values_dict = {}
for i, key in enumerate(self.search_space.keys()):
pos_ = int(pos[i])
values_dict[key] = list(self.search_space[key])[pos_]
return values_dict
SimonBlanke /
Hyperactive
