import numpy as np
import copy
import pickle as pickle
import itertools
import time
from .results import history
from .. import utility
from .. import score as search_score
from ...gp import predictor as gp_predictor
from ...blm import predictor as blm_predictor
from ...misc import set_config
from physbo.variable import variable
[docs]class policy:
def __init__(self, test_X, config=None, initial_data=None, comm=None):
"""
Parameters
----------
test_X: numpy.ndarray or physbo.variable
The set of candidates. Each row vector represents the feature vector of each search candidate.
config: set_config object (physbo.misc.set_config)
initial_data: tuple[np.ndarray, np.ndarray]
The initial training datasets.
The first elements is the array of actions and the second is the array of value of objective functions
comm: MPI.Comm, optional
MPI Communicator
"""
self.predictor = None
self.training = variable()
self.new_data = None
self.test = self._make_variable_X(test_X)
self.actions = np.arange(0, self.test.X.shape[0])
self.history = history()
if config is None:
self.config = set_config()
else:
self.config = config
if initial_data is not None:
if len(initial_data) != 2:
msg = "ERROR: initial_data should be 2-elements tuple or list (actions and objectives)"
raise RuntimeError(msg)
actions, fs = initial_data
if len(actions) != len(fs):
msg = "ERROR: len(initial_data[0]) != len(initial_data[1])"
raise RuntimeError(msg)
self.write(actions, fs)
self.actions = np.array(sorted(list(set(self.actions) - set(actions))))
if comm is None:
self.mpicomm = None
self.mpisize = 1
self.mpirank = 0
else:
self.mpicomm = comm
self.mpisize = comm.size
self.mpirank = comm.rank
self.actions = np.array_split(self.actions, self.mpisize)[self.mpirank]
self.config.learning.is_disp = (
self.config.learning.is_disp and self.mpirank == 0
)
[docs] def set_seed(self, seed):
"""
Setting a seed parameter for np.random.
Parameters
----------
seed: int
seed number
-------
"""
self.seed = seed
np.random.seed(self.seed)
[docs] def write(
self,
action,
t,
X=None,
time_total=None,
time_update_predictor=None,
time_get_action=None,
time_run_simulator=None,
):
"""
Writing history (update history, not output to a file).
Parameters
----------
action: numpy.ndarray
Indexes of actions.
t: numpy.ndarray
N dimensional array. The negative energy of each search candidate (value of the objective function to be optimized).
X: numpy.ndarray
N x d dimensional matrix. Each row of X denotes the d-dimensional feature vector 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
-------
"""
if X is None:
X = self.test.X[action, :]
Z = self.test.Z[action, :] if self.test.Z is not None else None
else:
Z = self.predictor.get_basis(X) if self.predictor is not None else None
self.history.write(
t,
action,
time_total=time_total,
time_update_predictor=time_update_predictor,
time_get_action=time_get_action,
time_run_simulator=time_run_simulator,
)
self.training.add(X=X, t=t, Z=Z)
local_index = np.searchsorted(self.actions, action)
local_index = local_index[
np.take(self.actions, local_index, mode="clip") == action
]
self.actions = self._delete_actions(local_index)
if self.new_data is None:
self.new_data = variable(X=X, t=t, Z=Z)
else:
self.new_data.add(X=X, t=t, Z=Z)
[docs] def random_search(
self, max_num_probes, num_search_each_probe=1, simulator=None, is_disp=True
):
"""
Performing random search.
Parameters
----------
max_num_probes: int
Maximum number of random search process.
num_search_each_probe: int
Number of search at each random search process.
simulator: callable
Callable (function or object with ``__call__``) from action to t
Here, action is an integer which represents the index of the candidate.
is_disp: bool
If true, process messages are outputted.
Returns
-------
history: history object (physbo.search.discrete.results.history)
"""
if self.mpirank != 0:
is_disp = False
N = int(num_search_each_probe)
if is_disp:
utility.show_interactive_mode(simulator, self.history)
for n in range(0, max_num_probes):
time_total = time.time()
if is_disp and N > 1:
utility.show_start_message_multi_search(self.history.num_runs)
time_get_action = time.time()
action = self._get_random_action(N)
time_get_action = time.time() - time_get_action
N_indeed = len(action)
if N_indeed == 0:
if self.mpirank == 0:
print("WARNING: All actions have already searched.")
return copy.deepcopy(self.history)
if simulator is None:
return action
time_run_simulator = time.time()
t = _run_simulator(simulator, action, self.mpicomm)
time_run_simulator = time.time() - time_run_simulator
time_total = time.time() - time_total
self.write(
action,
t,
time_total=[time_total] * N_indeed,
time_update_predictor=np.zeros(N_indeed, dtype=float),
time_get_action=[time_get_action] * N_indeed,
time_run_simulator=[time_run_simulator] * N_indeed,
)
if is_disp:
utility.show_search_results(self.history, N_indeed)
return copy.deepcopy(self.history)
[docs] def bayes_search(
self,
training=None,
max_num_probes=None,
num_search_each_probe=1,
predictor=None,
is_disp=True,
simulator=None,
score="TS",
interval=0,
num_rand_basis=0,
):
"""
Performing Bayesian optimization.
Parameters
----------
training: physbo.variable
Training dataset.
max_num_probes: int
Maximum number of searching process by Bayesian optimization.
num_search_each_probe: int
Number of searching by Bayesian optimization at each process.
predictor: predictor object
Base class is defined in physbo.predictor.
If None, blm_predictor is defined.
is_disp: bool
If true, process messages are outputted.
simulator: callable
Callable (function or object with ``__call__``)
Here, action is an integer which represents the index of the candidate.
score: str
The type of aquision funciton.
TS (Thompson Sampling), EI (Expected Improvement) and PI (Probability of Improvement) are available.
interval: int
The interval number of learning the hyper parameter.
If you set the negative value to interval, the hyper parameter learning is not performed.
If you set zero to interval, the hyper parameter learning is performed only at the first step.
num_rand_basis: int
The number of basis function. If you choose 0, ordinary Gaussian process run.
Returns
-------
history: history object (physbo.search.discrete.results.history)
"""
if self.mpirank != 0:
is_disp = False
old_disp = self.config.learning.is_disp
self.config.learning.is_disp = is_disp
if max_num_probes is None:
max_num_probes = 1
simulator = None
is_rand_expans = num_rand_basis != 0
if training is not None:
self.training = training
if predictor is not None:
self.predictor = predictor
elif self.predictor is None:
self._init_predictor(is_rand_expans)
if max_num_probes == 0 and interval >= 0:
self._learn_hyperparameter(num_rand_basis)
N = int(num_search_each_probe)
for n in range(max_num_probes):
time_total = time.time()
time_update_predictor = time.time()
if utility.is_learning(n, interval):
self._learn_hyperparameter(num_rand_basis)
else:
self._update_predictor()
time_update_predictor = time.time() - time_update_predictor
if num_search_each_probe != 1:
utility.show_start_message_multi_search(self.history.num_runs, score)
time_get_action = time.time()
K = self.config.search.multi_probe_num_sampling
alpha = self.config.search.alpha
action = self._get_actions(score, N, K, alpha)
time_get_action = time.time() - time_get_action
N_indeed = len(action)
if N_indeed == 0:
if self.mpirank == 0:
print("WARNING: All actions have already searched.")
break
if simulator is None:
self.config.learning.is_disp = old_disp
return action
time_run_simulator = time.time()
t = _run_simulator(simulator, action, self.mpicomm)
time_run_simulator = time.time() - time_run_simulator
time_total = time.time() - time_total
self.write(
action,
t,
time_total=[time_total] * N_indeed,
time_update_predictor=[time_update_predictor] * N_indeed,
time_get_action=[time_get_action] * N_indeed,
time_run_simulator=[time_run_simulator] * N_indeed,
)
if is_disp:
utility.show_search_results(self.history, N_indeed)
self._update_predictor()
self.config.learning.is_disp = old_disp
return copy.deepcopy(self.history)
@staticmethod
def _warn_no_predictor(method_name):
print("Warning: Since policy.predictor is not yet set,")
print(" a GP predictor (num_rand_basis=0) is used for predicting")
print(" If you want to use a BLM predictor (num_rand_basis>0),")
print(" call bayes_search(max_num_probes=0, num_rand_basis=nrb)")
print(" before calling {}.".format(method_name))
[docs] def get_post_fmean(self, xs):
"""Calculate mean value of predictor (post distribution)"""
X = self._make_variable_X(xs)
if self.predictor is None:
self._warn_no_predictor("get_post_fmean()")
predictor = gp_predictor(self.config)
predictor.fit(self.training, 0)
predictor.prepare(self.training)
return predictor.get_post_fmean(self.training, X)
else:
self._update_predictor()
return self.predictor.get_post_fmean(self.training, X)
[docs] def get_post_fcov(self, xs):
"""Calculate covariance of predictor (post distribution)"""
X = self._make_variable_X(xs)
if self.predictor is None:
self._warn_no_predictor("get_post_fcov()")
predictor = gp_predictor(self.config)
predictor.fit(self.training, 0)
predictor.prepare(self.training)
return predictor.get_post_fcov(self.training, X)
else:
self._update_predictor()
return self.predictor.get_post_fcov(self.training, X)
[docs] def get_score(
self,
mode,
*,
actions=None,
xs=None,
predictor=None,
training=None,
parallel=True,
alpha=1
):
"""
Calcualte score (acquisition function)
Parameters
----------
mode: str
The type of aquisition funciton. TS, EI and PI are available.
These functions are defined in score.py.
actions: array of int
actions to calculate score
xs: physbo.variable or np.ndarray
input parameters to calculate score
predictor: predictor object
predictor used to calculate score.
If not given, self.predictor will be used.
training:physbo.variable
Training dataset.
If not given, self.training will be used.
parallel: bool
Calculate scores in parallel by MPI (default: True)
alpha: float
Tuning parameter which is used if mode = TS.
In TS, multi variation is tuned as np.random.multivariate_normal(mean, cov*alpha**2, size).
Returns
-------
f: float or list of float
Score defined in each mode.
Raises
------
RuntimeError
If both *actions* and *xs* are given
Notes
-----
When neither *actions* nor *xs* are given, scores for actions not yet searched will be calculated.
When *parallel* is True, it is assumed that the function receives the same input (*actions* or *xs*) for all the ranks.
If you want to split the input array itself, set *parallel* be False and merge results by yourself.
"""
if training is None:
training = self.training
if training.X is None or training.X.shape[0] == 0:
msg = "ERROR: No training data is registered."
raise RuntimeError(msg)
if predictor is None:
if self.predictor is None:
self._warn_no_predictor("get_score()")
predictor = gp_predictor(self.config)
predictor.fit(training, 0)
predictor.prepare(training)
else:
self._update_predictor()
predictor = self.predictor
if xs is not None:
if actions is not None:
raise RuntimeError("ERROR: both actions and xs are given")
test = self._make_variable_X(xs)
if parallel and self.mpisize > 1:
actions = np.array_split(np.arange(test.X.shape[0]), self.mpisize)
test = test.get_subset(actions[self.mpirank])
else:
if actions is None:
actions = self.actions
else:
if isinstance(actions, int):
actions = [actions]
if parallel and self.mpisize > 1:
actions = np.array_split(actions, self.mpisize)[self.mpirank]
test = self.test.get_subset(actions)
f = search_score.score(
mode, predictor=predictor, training=training, test=test, alpha=alpha
)
if parallel and self.mpisize > 1:
fs = self.mpicomm.allgather(f)
f = np.hstack(fs)
return f
def _get_marginal_score(self, mode, chosen_actions, K, alpha):
"""
Getting marginal scores.
Parameters
----------
mode: str
The type of aquision funciton.
TS (Thompson Sampling), EI (Expected Improvement) and PI (Probability of Improvement) are available.
These functions are defined in score.py.
chosen_actions: numpy.ndarray
Array of selected actions.
K: int
The number of samples for evaluating score.
alpha: float
not used.
Returns
-------
f: list
N dimensional scores (score is defined in each mode)
"""
f = np.zeros((K, len(self.actions)), dtype=float)
# draw K samples of the values of objective function of chosen actions
new_test_local = self.test.get_subset(chosen_actions)
virtual_t_local = self.predictor.get_predict_samples(
self.training, new_test_local, K
)
if self.mpisize == 1:
new_test = new_test_local
virtual_t = virtual_t_local
else:
new_test = variable()
for nt in self.mpicomm.allgather(new_test_local):
new_test.add(X=nt.X, t=nt.t, Z=nt.Z)
virtual_t = np.concatenate(self.mpicomm.allgather(virtual_t_local), axis=1)
# virtual_t = self.predictor.get_predict_samples(self.training, new_test, K)
for k in range(K):
predictor = copy.deepcopy(self.predictor)
train = copy.deepcopy(self.training)
virtual_train = new_test
virtual_train.t = virtual_t[k, :]
if virtual_train.Z is None:
train.add(virtual_train.X, virtual_train.t)
else:
train.add(virtual_train.X, virtual_train.t, virtual_train.Z)
predictor.update(train, virtual_train)
f[k, :] = self.get_score(
mode, predictor=predictor, training=train, parallel=False
)
return np.mean(f, axis=0)
def _get_actions(self, mode, N, K, alpha):
"""
Getting next candidates
Parameters
----------
mode: str
The type of aquisition funciton.
TS (Thompson Sampling), EI (Expected Improvement) and PI (Probability of Improvement) are available.
These functions are defined in score.py.
N: int
The total number of actions to return.
K: int
The total number of samples to evaluate marginal score
alpha: float
Tuning parameter which is used if mode = TS.
In TS, multi variation is tuned as np.random.multivariate_normal(mean, cov*alpha**2, size).
Returns
-------
chosen_actions: numpy.ndarray
An N-dimensional array of actions selected in each search process.
"""
f = self.get_score(
mode,
predictor=self.predictor,
training=self.training,
alpha=alpha,
parallel=False,
)
champion, local_champion, local_index = self._find_champion(f)
if champion == -1:
return np.zeros(0, dtype=int)
if champion == local_champion:
self.actions = self._delete_actions(local_index)
chosen_actions = [champion]
for n in range(1, N):
f = self._get_marginal_score(mode, chosen_actions[0:n], K, alpha)
champion, local_champion, local_index = self._find_champion(f)
if champion == -1:
break
if champion == local_champion:
self.actions = self._delete_actions(local_index)
chosen_actions.append(champion)
return np.array(chosen_actions)
def _find_champion(self, f):
if len(f) == 0:
local_fmax = -float("inf")
local_index = -1
local_champion = -1
else:
local_fmax = np.max(f)
local_index = np.argmax(f)
local_champion = self.actions[local_index]
if self.mpisize == 1:
champion = local_champion
else:
local_champions = self.mpicomm.allgather(local_champion)
local_fs = self.mpicomm.allgather(local_fmax)
champion_rank = np.argmax(local_fs)
champion = local_champions[champion_rank]
return champion, local_champion, local_index
def _get_random_action(self, N):
"""
Getting indexes of actions randomly.
Parameters
----------
N: int
Total number of search candidates.
Returns
-------
action: numpy.ndarray
Indexes of actions selected randomly from search candidates.
"""
if self.mpisize == 1:
n = len(self.actions)
if n <= N:
index = np.arange(0, n)
else:
index = np.random.choice(len(self.actions), N, replace=False)
action = self.actions[index]
self.actions = self._delete_actions(index)
else:
nactions = self.mpicomm.gather(len(self.actions), root=0)
local_indices = [[] for _ in range(self.mpisize)]
if self.mpirank == 0:
hi = np.add.accumulate(nactions)
lo = np.roll(hi, 1)
lo[0] = 0
if hi[-1] <= N:
index = np.arange(0, hi[-1])
else:
index = np.random.choice(hi[-1], N, replace=False)
ranks = np.searchsorted(hi, index, side="right")
for r, i in zip(ranks, index):
local_indices[r].append(i - lo[r])
local_indices = self.mpicomm.scatter(local_indices, root=0)
local_actions = self.actions[local_indices]
self.actions = self._delete_actions(local_indices)
action = self.mpicomm.allgather(local_actions)
action = itertools.chain.from_iterable(action)
action = np.array(list(action))
return action
[docs] def save(self, file_history, file_training=None, file_predictor=None):
"""
Saving history, training and predictor into the corresponding files.
Parameters
----------
file_history: str
The name of the file that stores the information of the history.
file_training: str
The name of the file that stores the training dataset.
file_predictor: str
The name of the file that stores the predictor dataset.
Returns
-------
"""
if self.mpirank == 0:
self.history.save(file_history)
if file_training is not None:
self.training.save(file_training)
if file_predictor is not None:
with open(file_predictor, "wb") as f:
pickle.dump(self.predictor, f)
[docs] def load(self, file_history, file_training=None, file_predictor=None):
"""
Loading files about history, training and predictor.
Parameters
----------
file_history: str
The name of the file that stores the information of the history.
file_training: str
The name of the file that stores the training dataset.
file_predictor: str
The name of the file that stores the predictor dataset.
Returns
-------
"""
self.history.load(file_history)
if file_training is None:
N = self.history.total_num_search
X = self.test.X[self.history.chosen_actions[0:N], :]
t = self.history.fx[0:N]
self.training = variable(X=X, t=t)
else:
self.training = variable()
self.training.load(file_training)
if file_predictor is not None:
with open(file_predictor, "rb") as f:
self.predictor = pickle.load(f)
N = self.history.total_num_search
self.actions = self._delete_actions(self.history.chosen_actions[:N])
[docs] def export_predictor(self):
"""
Returning the predictor dataset
Returns
-------
"""
return self.predictor
[docs] def export_training(self):
"""
Returning the training dataset
Returns
-------
"""
return self.training
[docs] def export_history(self):
"""
Returning the information of the history.
Returns
-------
"""
return self.history
def _init_predictor(self, is_rand_expans):
"""
Initialize predictor.
Parameters
----------
is_rand_expans: bool
If true, physbo.blm.predictor is selected.
If false, physbo.gp.predictor is selected.
"""
if is_rand_expans:
self.predictor = blm_predictor(self.config)
else:
self.predictor = gp_predictor(self.config)
def _learn_hyperparameter(self, num_rand_basis):
self.predictor.fit(self.training, num_rand_basis)
self.test.Z = self.predictor.get_basis(self.test.X)
self.training.Z = self.predictor.get_basis(self.training.X)
self.predictor.prepare(self.training)
self.new_data = None
def _update_predictor(self):
if self.new_data is not None:
self.predictor.update(self.training, self.new_data)
self.new_data = None
def _make_variable_X(self, test_X):
"""
Make a new *variable* with X=test_X
Parameters
----------
test_X: numpy.ndarray or physbo.variable
The set of candidates. Each row vector represents the feature vector of each search candidate.
Returns
-------
test_X: numpy.ndarray or physbo.variable
The set of candidates. Each row vector represents the feature vector of each search candidate.
"""
if isinstance(test_X, np.ndarray):
test = variable(X=test_X)
elif isinstance(test_X, variable):
test = test_X
else:
raise TypeError("The type of test_X must be ndarray or physbo.variable")
return test
def _delete_actions(self, index, actions=None):
"""
Returns remaining actions
Notes
-----
This method itself does not modify *self*
Parameters
----------
index: int
Index of an action to be deleted.
actions: numpy.ndarray
Array of actions.
Returns
-------
actions: numpy.ndarray
Array of actions which does not include action specified by index.
"""
if actions is None:
actions = self.actions
return np.delete(actions, index)
def _run_simulator(simulator, action, comm=None):
if comm is None:
return simulator(action)
if comm.rank == 0:
t = simulator(action)
else:
t = 0.0
return comm.bcast(t, root=0)