def main():
instrum = ng.Instrumentation(alpha1=ng.var.Array(1).asscalar(),
alpha2=ng.var.Array(1).asscalar(),
beta1=ng.var.Array(1).asscalar(),
beta2=ng.var.Array(1).asscalar())
optimizer = ng.optimizers.CMA(instrumentation=instrum, budget=100)
for i in range(optimizer.budget):
try:
x = optimizer.ask()
value = score_ngca_algorithm_on_clover_data_by_kmeans(
*x.args, **x.kwargs)
print('{} out of {} - value {}'.format(i, optimizer.budget, value))
optimizer.tell(x, value)
except:
print('{} out of {} - error'.format(i, optimizer.budget))
recommendation = optimizer.provide_recommendation()
print('Optimal params:')
print(recommendation.kwargs)
print('Optimal Score on optimal params:')
# score result
score = score_ngca_algorithm_on_clover_data_by_kmeans(
recommendation.kwargs['alpha1'], recommendation.kwargs['alpha2'],
recommendation.kwargs['beta1'], recommendation.kwargs['beta2'])
utilities.print_score_fixed(score)
evaluate_test_data_by_kmeans(recommendation.kwargs)
def __init__(self, offset):
super().__init__(self.oracle_call, ng.Instrumentation(ng.var.Scalar()))
self.register_initialization(
offset=offset) # to create equivalent instances through "copy"
self.offset = offset
# add your own function descriptors (from base class, we already get "dimension" etc...)
# those will be recorded during benchmarks
self._descriptors.update(offset=offset)
def hparam_opt(config: dict, hparam_space: dict, backend: str, train_dataset=None, test_dataset=None, global_config: dict=global_config, optimizer_name: str="OnePlusOne", budget: int=20, mode="max"):
"""Conduct a hypeparameter space search"""
if backend == "lightning":
training_function = LightningTrainingContainer(config)
else:
training_function = TrainingContainer(config, train_dataset, test_dataset, global_config, backend, mode=mode)
hparam_vars = {key: ng.instrumentation.var.OrderedDiscrete(value) for key, value in hparam_space.items()}
instrumentation = ng.Instrumentation(**hparam_vars)
print(instrumentation)
optimizer = ng.optimizers.registry[optimizer_name](instrumentation=instrumentation, budget=budget)
return optimizer.minimize(training_function)
def _assemble_instrumentation(self, parameters=None):
""" Assemble nevergrad.Instrumentation object from `parameters` list.
Parameters
----------
parameters: List(BaseMCOParameter)
parameter objects containing lower and upper numerical bounds
Returns
----------
instrumentation: ng.Instrumentation
"""
if parameters is None:
parameters = self.parameters
instrumentation = [
self._create_instrumentation_variable(p) for p in parameters
]
return ng.Instrumentation(*instrumentation)
def find_detection_params_for_image(v):
cache = {}
def detection_loss(gauss, d_diam, e_diam, thresh):
params = gauss, d_diam, e_diam, thresh
if params in cache.keys():
return cache[params]
image = do_detection(v, gauss, d_diam, e_diam, thresh)
prop = regionprops(label(image))
c1 = np.array([p.area for p in prop], dtype=np.float32)
if len(c1) < 3:
return np.inf
# Detection loss
ratio1 = -np.log(np.sum(v[image > 0.]) / np.sum(v))
ratio2 = -np.log(
np.sum(v[image > 0.]) / float(np.count_nonzero(image)))
image_loss = ratio1 + ratio2
cache[params] = image_loss
return image_loss
bounds = ng.Instrumentation(
ng.var.Scalar(float).bounded(0., 10.),
ng.var.Scalar(int).bounded(0, 10),
ng.var.Scalar(int).bounded(0, 10),
ng.var.Scalar(float).bounded(0., 1.))
optimizer = ng.optimizers.PortfolioDiscreteOnePlusOne(
instrumentation=bounds, budget=4000)
recommendation = optimizer.minimize(detection_loss)
opt_params = recommendation.args
return opt_params
def test_log_parameters(tmp_path: Path) -> None:
filepath = tmp_path / "logs.txt"
cases = [0, np.int(1), np.float(2.0), np.nan, float("inf"), np.inf]
instrum = ng.Instrumentation(ng.var.Array(1),
ng.var.Scalar(),
blublu=ng.var.SoftmaxCategorical(cases),
array=ng.var.Array(3, 2))
optimizer = optimizerlib.OnePlusOne(instrumentation=instrum, budget=32)
optimizer.register_callback("tell", callbacks.ParametersLogger(filepath, delete_existing_file=True))
optimizer.minimize(_func, verbosity=2)
# pickling
logger = callbacks.ParametersLogger(filepath)
logs = logger.load_flattened()
assert len(logs) == 32
assert isinstance(logs[-1]["#arg1"], float)
assert len(logs[-1]) == 16
logs = logger.load_flattened(max_list_elements=2)
assert len(logs[-1]) == 12
# deletion
logger = callbacks.ParametersLogger(filepath, delete_existing_file=True)
assert not logger.load()
def main():
# Optimize params on test and validation datasets
instrum = ng.Instrumentation(
alpha1=ng.var.Array(1).asscalar(),
alpha2=ng.var.Array(1).asscalar(),
beta1=ng.var.Array(1).bounded(a_min=0, a_max=constant.MAX_BETA_VALUE),
beta2=ng.var.Array(1).bounded(a_min=0, a_max=constant.MAX_BETA_VALUE))
optimizer = ng.optimizers.OnePlusOne(instrumentation=instrum, budget=100)
# recommendation = optimizer.minimize(score_ngca_on_oil_data_by_kmeans)
# ask and tell
for i in range(optimizer.budget):
try:
print('{} out of {}'.format(i, optimizer.budget))
x = optimizer.ask()
value = score_ngca_on_oil_data_by_kmeans(*x.args, **x.kwargs)
optimizer.tell(x, value)
except:
print('Error')
recommendation = optimizer.provide_recommendation()
print('Optimal params:')
print(recommendation.kwargs)
print('Optimal Score on train and validation data:')
# score result
score = score_ngca_on_oil_data_by_kmeans(recommendation.kwargs['alpha1'],
recommendation.kwargs['alpha2'],
recommendation.kwargs['beta1'],
recommendation.kwargs['beta2'])
utilities.print_score_fixed(score)
# Run algorithm with optimal params on test data and evaluate score
# Plot projected test data on the result NG subspace
evaluate_test_data_by_kmeans(recommendation.kwargs)
return
def __init__(self, environment, random_seed, policy_factory,
create_rollout, optimizer, budget, low, high):
self._environment = environment
self._random_seed = random_seed
self._policy_factory = policy_factory
self._create_rollout = create_rollout
self._budget = budget
self._low = low
self._high = high
self._deterministic_update_policy = True
self._observation_space = environment.observation_space
self._action_space = environment.action_space
self._policy = self._create_policy()
self._policy_return = -np.inf
self._policy_steps = -np.inf
self._shape = self._policy.get_parameters()['model'].shape
self._dims = np.prod(self._shape)
instrumentation = ng.Instrumentation(
ng.var.Array(self._dims).bounded(low, high))
self._optimizer = optimizer(instrumentation=instrumentation,
budget=budget)
def main():
cppn = get_model()
cppn.summary()
print("nevergrad optimizers:")
print(list(sorted(ng.optimizers.registry.keys())))
print("cppn weight shapes:")
[print(weight.shape) for weight in cppn.trainable_weights]
vars = [ng.var.Array(*tuple(weight.shape)).bounded(-10, 10) for weight in cppn.trainable_weights]
optimizer = ng.optimizers.registry["RotationInvariantDE"](instrumentation=ng.Instrumentation(*vars))
model_number = 0
best_loss = 123456789.
mnist = tfds.load(name="mnist", split=tfds.Split.TRAIN).repeat().shuffle(128).batch(BATCH_SIZE)
max_batches = BATCHES_PER_TEST - 2
while True:
weights = optimizer.ask()
cppn.set_weights([tf.convert_to_tensor(arg, dtype=tf.float32) for arg in weights.args])
total_loss = test(cppn, mnist, max_batches).numpy().item()
optimizer.tell(weights, total_loss)
if total_loss < best_loss:
best_loss = total_loss
cppn.save("./best.h5")
print(model_number, "loss:", total_loss, "best:", best_loss)
model_number += 1
def get_instrumentation_from_config(cfg):
var_type = ng.instrumentation.utils.Variable
var_keys = cfg.nested_get_key_fn(lambda x: isinstance(x, var_type))
search_vars = {key: cfg.get_with_dot_str(key) for key in var_keys}
return ng.Instrumentation(**search_vars)
count = 0
target_density = np.linspace(0, 1, N_ADSORP+1)
def fitness(*grid):
global count
count += 1
print(count)
grid = np.array(grid)
density = (run_dft_fast(grid))[:41].reshape(41,)
return np.sum(np.abs(density-target_density))
iargs = []
for _ in range(400):
arg1 = ng.var.OrderedDiscrete([0,1])
iargs.append(arg1)
instrum = ng.Instrumentation(*iargs)
# optimizer = ng.optimizers.PSO(instrumentation=instrum, budget=300)
# recommendation = optimizer.minimize(fitness)
# print(recommendation)
# grid = recommendation.data
# tp = (0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 0, 1, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 0, 1, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 1, 0, 1, 0, 1, 0, 1, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 1, 1, 1, 1, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 1, 1, 1, 0, 0, 1, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1, 0, 0, 0, 1, 0, 1, 1, 0, 0, 0, 1, 0, 1, 0, 1, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 1, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 0, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 0, 0, 0, 1, 0, 1, 0, 0, 1, 0, 0, 0, 0, 1, 0, 1, 0, 0, 1, 1, 0, 0, 1, 0, 1, 0, 0, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 0, 1, 1, 0, 0, 0, 1, 1, 1, 1, 0, 1, 0, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 0, 1, 0, 1, 0, 0, 0, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 0, 1, 1, 1, 1, 0, 0, 1, 1, 1, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 0, 1, 0, 0, 0, 1, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 0, 1, 1, 0, 1)
# grid = np.array(tp)
grid = np.array([1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, 1,1,1,1,1,1,1,1,1,1,0,0,1,1,1,1,1,1,1,1, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1, 1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,1,1,1,1,1, 1,1,1,1,1,1,1,1,0,0,0,0,0,1,1,1,1,1,1,1, 1,1,1,1,1,1,1,1,0,0,0,0,1,1,1,1,1,1,1,1, 1,1,1,1,1,1,1,1,0,0,0,1,1,1,1,1,1,1,1,1, 1,1,1,1,1,1,1,1,0,0,1,1,1,1,1,1,1,1,1,1, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,1,1, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, 0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, 0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1])
density = (run_dft_fast(grid))[:41].reshape(41,)
plt.plot(np.linspace(0, 1, 41), density)
plt.show()
user="******",
passwd=os.environ['MYSQL_PASS'],
db="dqn_results")
cur = db.cursor()
experimentName = "soft-actor-critic-fixed-entropy"
entropyCoefficientArg = ng.instrumentation.variables.Gaussian(mean=-2, std=2)
learningRateArg = ng.instrumentation.variables.Gaussian(mean=-3, std=2)
rewardScalingArg = ng.instrumentation.variables.Gaussian(mean=-3, std=2)
actionScalingArg = ng.instrumentation.variables.Gaussian(mean=-1, std=1)
weightRegularizationConstantArg = ng.instrumentation.variables.Gaussian(
mean=-2, std=1)
instrumentation = ng.Instrumentation(entropyCoefficientArg, learningRateArg,
rewardScalingArg, actionScalingArg,
weightRegularizationConstantArg)
optimizer = ng.optimizers.registry["TBPSA"](instrumentation=instrumentation,
budget=os.environ['BUDGET'])
cur.execute("select x1, x2, x3, x4, x5, y from experiments where label = '" +
experimentName + "'")
result = cur.fetchall()
for row in result:
candidate = optimizer.create_candidate.from_call(np.log10(float(row[0])),
np.log10(float(row[1])),
np.log10(float(row[2])),
np.log10(float(row[3])),
np.log10(float(row[4])))
optimizer.tell(candidate, -float(row[5]))
import nevergrad as ng
import numpy as np
from ray.tune import schedulers
from ray.tune.suggest.nevergrad import NevergradSearch
budget = 400
opt_args = {
'scheduler.target_lrs': ng.var.Log(1e-4, 1e-2),
'scheduler.final_lr': ng.var.Log(1e-7, 1e-4),
'optimizer.weight_decay': ng.var.Log(1e-7, 1e-3),
'scheduler.exp_factor': ng.var.Scalar().bounded(-.5, 2),
}
inst = ng.Instrumentation(**opt_args)
opt = ng.optimizers.registry['CMA'](instrumentation=inst, budget=budget)
search_alg = NevergradSearch(opt,
None,
max_concurrent=4,
reward_attr='valid_accuracy',
mode='max')
exp_args = {
'resume': False,
'search_alg': search_alg,
'scheduler': schedulers.FIFOScheduler(),
'stop': {
'is_finished': True
},
'verbose': True,
'num_samples': budget,
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
#
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.
from pathlib import Path
import nevergrad as ng
from nevergrad.instrumentation import FolderFunction
if __name__ == "__main__":
folder = Path(__file__).parents[1] / "instrumentation" / "examples"
func = FolderFunction(folder, ["python", "examples/script.py"],
clean_copy=True)
instrumentation = ng.Instrumentation(value1=ng.var.Array(1).asscalar(),
value2=12,
string=ng.var.SoftmaxCategorical(
["plop", "blublu", "plouf"]))
opt = ng.optimizers.registry["OnePlusOne"](instrumentation, budget=4)
opt.minimize(func)
ng.families.ParametrizedOnePlusOne()