# SPDX-License-Identifier: MPL-2.0
# Copyright (C) 2020- The University of Tokyo
#
# This Source Code Form is subject to the terms of the Mozilla Public
# License, v. 2.0. If a copy of the MPL was not distributed with this
# file, You can obtain one at https://mozilla.org/MPL/2.0/.
import numpy as np
import copy
from .. import utility
MAX_SEARCH = int(30000)
[docs]
class History:
def __init__(self):
self.num_runs = int(0)
self.total_num_search = int(0)
self.fx = np.zeros(MAX_SEARCH, dtype=float)
self.chosen_actions = np.zeros(MAX_SEARCH, dtype=int)
self.terminal_num_run = np.zeros(MAX_SEARCH, dtype=int)
self.time_total_ = np.zeros(MAX_SEARCH, dtype=float)
self.time_update_predictor_ = np.zeros(MAX_SEARCH, dtype=float)
self.time_get_action_ = np.zeros(MAX_SEARCH, dtype=float)
self.time_run_simulator_ = np.zeros(MAX_SEARCH, dtype=float)
@property
def time_total(self):
return copy.copy(self.time_total_[0 : self.num_runs])
@property
def time_update_predictor(self):
return copy.copy(self.time_update_predictor_[0 : self.num_runs])
@property
def time_get_action(self):
return copy.copy(self.time_get_action_[0 : self.num_runs])
@property
def time_run_simulator(self):
return copy.copy(self.time_run_simulator_[0 : self.num_runs])
@property
def valid_mask(self):
"""
Mask of valid observations (True) vs failed ones (False).
An observation is failed when its objective value is not finite
(NaN or +-Inf). Failed observations are kept in the history but
excluded from the training data and the best-value tracking.
Returns
-------
numpy.ndarray of bool, shape (total_num_search,)
"""
return utility.finite_mask(self.fx[0 : self.total_num_search])
[docs]
def export_valid(self):
"""
Export the valid (successfully evaluated) observations.
Returns
-------
actions: numpy.ndarray
Indexes of the actions of the valid observations.
fx: numpy.ndarray
Objective values of the valid observations.
"""
N = self.total_num_search
mask = self.valid_mask
return self.chosen_actions[0:N][mask], self.fx[0:N][mask]
[docs]
def write(
self,
t,
action,
time_total=None,
time_update_predictor=None,
time_get_action=None,
time_run_simulator=None,
):
"""
Overwrite fx and chosen_actions by t and action.
Parameters
----------
t: numpy.ndarray
N dimensional array. The negative energy of each search candidate (value of the objective function to be optimized).
action: numpy.ndarray
N dimensional array. The indexes of actions of each search candidate.
time_total: numpy.ndarray
N dimenstional array. The total elapsed time in each step.
If None (default), filled by 0.0.
time_update_predictor: numpy.ndarray
N dimenstional array. The elapsed time for updating predictor (e.g., learning hyperparemters) in each step.
If None (default), filled by 0.0.
time_get_action: numpy.ndarray
N dimenstional array. The elapsed time for getting next action in each step.
If None (default), filled by 0.0.
time_run_simulator: numpy.ndarray
N dimenstional array. The elapsed time for running the simulator in each step.
If None (default), filled by 0.0.
Returns
-------
"""
N = utility.length_vector(action)
st = self.total_num_search
en = st + N
t_shape = t.shape
if t_shape[0] != N:
raise ValueError(f"Number of actions and t must be the same: {N} != {t_shape[0]}")
if t_shape[1] != 1:
raise ValueError(f"t.shape[1] must be 1: {t_shape}")
self.terminal_num_run[self.num_runs] = en
self.fx[st:en] = t[:, 0]
self.chosen_actions[st:en] = action
self.num_runs += 1
self.total_num_search += N
if time_total is None:
time_total = np.zeros(N, dtype=float)
self.time_total_[st:en] = time_total
if time_update_predictor is None:
time_update_predictor = np.zeros(N, dtype=float)
self.time_update_predictor_[st:en] = time_update_predictor
if time_get_action is None:
time_get_action = np.zeros(N, dtype=float)
self.time_get_action_[st:en] = time_get_action
if time_run_simulator is None:
time_run_simulator = np.zeros(N, dtype=float)
self.time_run_simulator_[st:en] = time_run_simulator
[docs]
def export_sequence_best_fx(self):
"""
Export fx and actions at each sequence.
(The total number of data is num_runs.)
Returns
-------
best_fx: numpy.ndarray
best_actions: numpy.ndarray
"""
all_best_fx, all_best_actions = self.export_all_sequence_best_fx()
best_fx = np.zeros(self.num_runs, dtype=float)
best_actions = np.zeros(self.num_runs, dtype=int)
for n in range(self.num_runs):
index = self.terminal_num_run[n] - 1
best_fx[n] = all_best_fx[index]
best_actions[n] = all_best_actions[index]
return best_fx, best_actions
[docs]
def export_all_sequence_best_fx(self):
"""
Export all fx and actions at each sequence.
(The total number of data is total_num_research.)
Returns
-------
best_fx: numpy.ndarray
best_actions: numpy.ndarray
"""
# Failed observations (non-finite fx) are skipped. Until the first
# valid observation, best_fx is NaN and best_actions is -1.
best_fx = np.full(self.total_num_search, np.nan, dtype=float)
best_actions = np.full(self.total_num_search, -1, dtype=int)
for n in range(self.total_num_search):
if n > 0:
best_fx[n] = best_fx[n - 1]
best_actions[n] = best_actions[n - 1]
fx = self.fx[n]
if np.isfinite(fx) and (np.isnan(best_fx[n]) or best_fx[n] < fx):
best_fx[n] = fx
best_actions[n] = self.chosen_actions[n]
return best_fx, best_actions
[docs]
def save(self, filename):
"""
Save the information of the history.
Parameters
----------
filename: str
The name of the file which stores the information of the history
Returns
-------
"""
N = self.total_num_search
M = self.num_runs
np.savez_compressed(
filename,
num_runs=M,
total_num_search=N,
fx=self.fx[0:N],
chosen_actions=self.chosen_actions[0:N],
terminal_num_run=self.terminal_num_run[0:M],
)
[docs]
def load(self, filename):
"""
Load the information of the history.
Parameters
----------
filename: str
The name of the file which stores the information of the history
Returns
-------
"""
data = np.load(filename)
M = int(data["num_runs"])
N = int(data["total_num_search"])
self.num_runs = M
self.total_num_search = N
self.fx[0:N] = data["fx"]
self.chosen_actions[0:N] = data["chosen_actions"]
self.terminal_num_run[0:M] = data["terminal_num_run"]
[docs]
def show_search_results(self, N):
n = self.total_num_search
fx = self.fx[0:n]
valid = np.isfinite(fx)
if np.any(valid):
# failed observations are skipped
index = np.argmax(np.where(valid, fx, -np.inf))
best_msg = "current best f(x) = %f (best action=%d)" % (
fx[index],
self.chosen_actions[index],
)
else:
best_msg = "current best f(x) = (no valid observation yet)"
if N == 1:
print(
"%04d-th step: f(x) = %f (action=%d)"
% (n, self.fx[n - 1], self.chosen_actions[n - 1])
)
print(" " + best_msg + " \n")
else:
print(best_msg + " ")
print("list of simulation results")
st = self.total_num_search - N
en = self.total_num_search
for n in range(st, en):
print("f(x)=%f (action = %d)" % (self.fx[n], self.chosen_actions[n]))
print("\n")