gradient_free_optimizers.search.Search._init_memory() - Code Metrics - Inspection of "small update in tree based opt." - SimonBlanke/Gradient-Free-Optimizers - Measure and Improve Code Quality continuously with Scrutinizer

Passed

Push — master ( b0b7f3...e3d174 )

by Simon

created 2020-11-05 15:37 UTC

Search._init_memory() A

↳ Parent: gradient_free_optimizers.search

Complexity

Conditions

Size

Total Lines	28
Code Lines	20

Duplication

Lines	0
Ratio	0 %

Importance

Changes

Metric	Value
cc	3
eloc	20
nop	2
dl	0
loc	28
rs	9.4
c	0
b	0
f	0

# Author: Simon Blanke
# Email: [email protected]
# License: MIT License

import time
import random

import numpy as np
import pandas as pd

from .init_positions import Initializer
from .progress_bar import ProgressBarLVL0, ProgressBarLVL1
from .times_tracker import TimesTracker
from .results_manager import ResultsManager
from .memory import Memory
from .converter import Converter
from .print_info import print_info

p_bar_dict = {
    False: ProgressBarLVL0,
    True: ProgressBarLVL1,
}


def time_exceeded(start_time, max_time):
    run_time = time.time() - start_time
    return max_time and run_time > max_time


def score_exceeded(score_best, max_score):
    return max_score and score_best >= max_score


def set_random_seed(nth_process, random_state):
    """
    Sets the random seed separately for each thread
    (to avoid getting the same results in each thread)
    """
    if nth_process is None:
        nth_process = 0

    if random_state is None:
        random_state = np.random.randint(0, high=2 ** 32 - 2)

    random.seed(random_state + nth_process)
    np.random.seed(random_state + nth_process)


class Search(TimesTracker):
    def __init__(self):
        super().__init__()

        self.optimizers = []
        self.new_results_list = []
        self.all_results_list = []

    @TimesTracker.eval_time
    def _score(self, pos):
        return self.score(pos)

    def _init_memory(self, memory):
        memory_warm_start = self.memory_warm_start

        self.memory_dict = {}
        self.memory_dict_new = {}

        if isinstance(memory_warm_start, pd.DataFrame):
            parameter = set(self.search_space.keys())
            memory_para = set(memory_warm_start.columns)

            if parameter <= memory_para:
                values_list = list(
                    memory_warm_start[list(self.search_space.keys())].values
                )
                scores = memory_warm_start["score"]

                value_tuple_list = list(map(tuple, values_list))
                self.memory_dict = dict(zip(value_tuple_list, scores))
            else:
                missing = parameter - memory_para

                print(
                    "\nWarning:",
                    '"{}"'.format(*missing),
                    "is in search_space but not in memory dataframe",
                )
                print(
                    "Optimization run will continue "
                    "without memory warm start\n"
                )

    @TimesTracker.iter_time
    def _initialization(self, init_pos):
        self.init_pos(init_pos)

        score_new = self._score(init_pos)
        self.evaluate(score_new)

        self.p_bar.update(score_new, init_pos)

    @TimesTracker.iter_time
    def _iteration(self):
        pos_new = self.iterate()

        score_new = self._score(pos_new)
        self.evaluate(score_new)

        self.p_bar.update(score_new, pos_new)

    def _init_search(self):
        self._init_memory(self.memory)
        self.p_bar = p_bar_dict[self.progress_bar](
            self.nth_process, self.n_iter, self.objective_function
        )
        set_random_seed(self.nth_process, self.random_state)

        if self.warm_start is not None:
            self.initialize["warm_start"] = self.warm_start

        # get init positions
        init = Initializer(self.conv)
        init_positions = init.set_pos(self.initialize)

        return init_positions

    def _early_stop(self):
        if time_exceeded(self.start_time, self.max_time):
            return True
        elif score_exceeded(self.p_bar.score_best, self.max_score):
            return True
        else:
            return False

    def _init_verb_dict(self, verb_dict):
        if verb_dict in [None, False]:
            return {
                "progress_bar": False,
                "print_results": False,
                "print_times": False,
            }

        verb_default = {
            "progress_bar": True,
            "print_results": True,
            "print_times": True,
        }

        for verb_key in verb_default.keys():
            if verb_key not in verb_dict:
                verb_dict[verb_key] = verb_default[verb_key]

        return verb_dict

    def search(
        self,
        objective_function,
        n_iter,
        initialize={"grid": 8, "random": 4, "vertices": 8},
        warm_start=None,
        max_time=None,
        max_score=None,
        memory=True,
        memory_warm_start=None,
        verbosity={
            "progress_bar": True,
            "print_results": True,
            "print_times": True,
        },
        random_state=None,
        nth_process=None,
    ):
        self.start_time = time.time()

        verbosity = self._init_verb_dict(verbosity)

        self.objective_function = objective_function
        self.n_iter = n_iter
        self.initialize = initialize
        self.warm_start = warm_start
        self.max_time = max_time
        self.max_score = max_score
        self.memory = memory
        self.memory_warm_start = memory_warm_start
        self.progress_bar = verbosity["progress_bar"]
        self.random_state = random_state
        self.nth_process = nth_process

        conv = Converter(self.search_space)
        self.conv = conv
        results = ResultsManager(objective_function, conv)
        init_positions = self._init_search()

        if memory is True:
            mem = Memory(memory_warm_start, conv)
            self.score = mem.memory(results.score)
        else:
            self.score = results.score

        # loop to initialize N positions
        for init_pos, nth_iter in zip(init_positions, range(n_iter)):
            if self._early_stop():
                break
            self._initialization(init_pos)

        # loop to do the iterations
        for nth_iter in range(len(init_positions), n_iter):
            if self._early_stop():
                break
            self._iteration()

        self.results = pd.DataFrame(results.results_list)

        self.best_score = self.p_bar.score_best
        self.best_value = conv.position2value(self.p_bar.pos_best)
        self.best_para = conv.value2para(self.best_value)

        eval_time = np.array(self.eval_times).sum()
        iter_time = np.array(self.iter_times).sum()

        self.p_bar.close()

        print_info(
            verbosity,
            self.objective_function,
            self.best_score,
            self.best_para,
            eval_time,
            iter_time,
            self.n_iter,
        )



1			# Author: Simon Blanke
2			# Email: [email protected]
3			# License: MIT License
4
5			import time
6			import random
7
8			import numpy as np
9			import pandas as pd
10
11			from .init_positions import Initializer
12			from .progress_bar import ProgressBarLVL0, ProgressBarLVL1
13			from .times_tracker import TimesTracker
14			from .results_manager import ResultsManager
15			from .memory import Memory
16			from .converter import Converter
17			from .print_info import print_info
18
19			p_bar_dict = {
20			False: ProgressBarLVL0,
21			True: ProgressBarLVL1,
22			}
23
24
25			def time_exceeded(start_time, max_time):
26			run_time = time.time() - start_time
27			return max_time and run_time > max_time
28
29
30			def score_exceeded(score_best, max_score):
31			return max_score and score_best >= max_score
32
33
34			def set_random_seed(nth_process, random_state):
35			"""
36			Sets the random seed separately for each thread
37			(to avoid getting the same results in each thread)
38			"""
39			if nth_process is None:
40			nth_process = 0
41
42			if random_state is None:
43			random_state = np.random.randint(0, high=2 ** 32 - 2)
44
45			random.seed(random_state + nth_process)
46			np.random.seed(random_state + nth_process)
47
48
49			class Search(TimesTracker):
50			def __init__(self):
51			super().__init__()
52
53			self.optimizers = []
54			self.new_results_list = []
55			self.all_results_list = []
56
57			@TimesTracker.eval_time
58			def _score(self, pos):
59			return self.score(pos)
60
61			def _init_memory(self, memory):
62			memory_warm_start = self.memory_warm_start
63
64			self.memory_dict = {}
65			self.memory_dict_new = {}
66
67			if isinstance(memory_warm_start, pd.DataFrame):
68			parameter = set(self.search_space.keys())
69			memory_para = set(memory_warm_start.columns)
70
71			if parameter <= memory_para:
72			values_list = list(
73			memory_warm_start[list(self.search_space.keys())].values
74			)
75			scores = memory_warm_start["score"]
76
77			value_tuple_list = list(map(tuple, values_list))
78			self.memory_dict = dict(zip(value_tuple_list, scores))
79			else:
80			missing = parameter - memory_para
81
82			print(
83			"\nWarning:",
84			'"{}"'.format(*missing),
85			"is in search_space but not in memory dataframe",
86			)
87			print(
88			"Optimization run will continue "
89			"without memory warm start\n"
90			)
91
92			@TimesTracker.iter_time
93			def _initialization(self, init_pos):
94			self.init_pos(init_pos)
95
96			score_new = self._score(init_pos)
97			self.evaluate(score_new)
98
99			self.p_bar.update(score_new, init_pos)
100
101			@TimesTracker.iter_time
102			def _iteration(self):
103			pos_new = self.iterate()
104
105			score_new = self._score(pos_new)
106			self.evaluate(score_new)
107
108			self.p_bar.update(score_new, pos_new)
109
110			def _init_search(self):
111			self._init_memory(self.memory)
112			self.p_bar = p_bar_dict[self.progress_bar](
113			self.nth_process, self.n_iter, self.objective_function
114			)
115			set_random_seed(self.nth_process, self.random_state)
116
117			if self.warm_start is not None:
118			self.initialize["warm_start"] = self.warm_start
119
120			# get init positions
121			init = Initializer(self.conv)
122			init_positions = init.set_pos(self.initialize)
123
124			return init_positions
125
126			def _early_stop(self):
127			if time_exceeded(self.start_time, self.max_time):
128			return True
129			elif score_exceeded(self.p_bar.score_best, self.max_score):
130			return True
131			else:
132			return False
133
134			def _init_verb_dict(self, verb_dict):
135			if verb_dict in [None, False]:
136			return {
137			"progress_bar": False,
138			"print_results": False,
139			"print_times": False,
140			}
141
142			verb_default = {
143			"progress_bar": True,
144			"print_results": True,
145			"print_times": True,
146			}
147
148			for verb_key in verb_default.keys():
149			if verb_key not in verb_dict:
150			verb_dict[verb_key] = verb_default[verb_key]
151
152			return verb_dict
153
154			def search(
155			self,
156			objective_function,
157			n_iter,
158			initialize={"grid": 8, "random": 4, "vertices": 8},
159			warm_start=None,
160			max_time=None,
161			max_score=None,
162			memory=True,
163			memory_warm_start=None,
164			verbosity={
165			"progress_bar": True,
166			"print_results": True,
167			"print_times": True,
168			},
169			random_state=None,
170			nth_process=None,
171			):
172			self.start_time = time.time()
173
174			verbosity = self._init_verb_dict(verbosity)
175
176			self.objective_function = objective_function
177			self.n_iter = n_iter
178			self.initialize = initialize
179			self.warm_start = warm_start
180			self.max_time = max_time
181			self.max_score = max_score
182			self.memory = memory
183			self.memory_warm_start = memory_warm_start
184			self.progress_bar = verbosity["progress_bar"]
185			self.random_state = random_state
186			self.nth_process = nth_process
187
188			conv = Converter(self.search_space)
189			self.conv = conv
190			results = ResultsManager(objective_function, conv)
191			init_positions = self._init_search()
192
193			if memory is True:
194			mem = Memory(memory_warm_start, conv)
195			self.score = mem.memory(results.score)
196			else:
197			self.score = results.score
198
199			# loop to initialize N positions
200			for init_pos, nth_iter in zip(init_positions, range(n_iter)):
201			if self._early_stop():
202			break
203			self._initialization(init_pos)
204
205			# loop to do the iterations
206			for nth_iter in range(len(init_positions), n_iter):
207			if self._early_stop():
208			break
209			self._iteration()
210
211			self.results = pd.DataFrame(results.results_list)
212
213			self.best_score = self.p_bar.score_best
214			self.best_value = conv.position2value(self.p_bar.pos_best)
215			self.best_para = conv.value2para(self.best_value)
216
217			eval_time = np.array(self.eval_times).sum()
218			iter_time = np.array(self.iter_times).sum()
219
220			self.p_bar.close()
221
222			print_info(
223			verbosity,
224			self.objective_function,
225			self.best_score,
226			self.best_para,
227			eval_time,
228			iter_time,
229			self.n_iter,
230			)
231
232

SimonBlanke / Gradient-Free-Optimizers

Push — master ( b0b7f3...e3d174 )

Search._init_memory() A

Complexity

Size

Duplication

Importance

Duplication Side-by-Side

Filter issues like