wassname/ray
1
0
Fork 0
mirror of https://github.com/wassname/ray.git synced 2026-09-09 11:32:43 +08:00
Code Issues Packages Projects Releases Wiki Activity

[tune] Refactor search algorithms (#7037)

Browse Source 
* start refactoring of search algorithms

* format

* needs tests

* fix

* suggestions

* Fix PBT

* lint

* refactoring

* hyperopt_working

* dragonfly

* hyperopt

* change_half_of_algs

* save

* code-removed

* remove_lots_of_unneccessary

* changes

* formatting

* suggest

* reset

* rm

* tests

* search-change

* exception

* refactor-doc

* search

* py

* moredocs

* Update doc/source/tune-searchalg.rst

* concurrency

* max

* tune

* betterwarning

* bohb

* tests

* test-change

Co-authored-by: ujvl <misraujval@gmail.com>
This commit is contained in:
Richard Liaw
2020-04-27 08:51:13 -07:00
committed by GitHub
co-authored by ujvl
parent 1d5bceddf0
commit 87557a00fa
31 changed files with 527 additions and 611 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/ray/tune/suggest/nevergrad.py
+18 -44
View File
@@ -5,15 +5,16 @@ try:
except ImportError:
ng = None
from ray.tune.suggest.suggestion import SuggestionAlgorithm
from ray.tune.suggest import Searcher
logger = logging.getLogger(__name__)
class NevergradSearch(SuggestionAlgorithm):
class NevergradSearch(Searcher):
"""A wrapper around Nevergrad to provide trial suggestions.
Requires Nevergrad to be installed.
Nevergrad is an open source tool from Facebook for derivative free
optimization of parameters and/or hyperparameters. It features a wide
range of optimizers in a standard ask and tell interface. More information
@@ -26,20 +27,20 @@ class NevergradSearch(SuggestionAlgorithm):
the dimension of the optimizer output. Alternatively, set to None
if the optimizer is already instrumented with kwargs
(see nevergrad v0.2.0+).
max_concurrent (int): Number of maximum concurrent trials. Defaults
to 10.
metric (str): The training result objective value attribute.
mode (str): One of {min, max}. Determines whether objective is
minimizing or maximizing the metric attribute.
use_early_stopped_trials (bool): Whether to use early terminated
trial results in the optimization process.
use_early_stopped_trials: Deprecated.
max_concurrent: Deprecated.
Example:
>>> from nevergrad.optimization import optimizerlib
>>> instrumentation = 1
>>> optimizer = optimizerlib.OnePlusOne(instrumentation, budget=100)
>>> algo = NevergradSearch(optimizer, ["lr"], max_concurrent=4,
>>> metric="mean_loss", mode="min")
.. code-block:: python
from nevergrad.optimization import optimizerlib
instrumentation = 1
optimizer = optimizerlib.OnePlusOne(instrumentation, budget=100)
algo = NevergradSearch(
optimizer, ["lr"], metric="mean_loss", mode="min")
Note:
In nevergrad v0.2.0+, optimizers can be instrumented.
@@ -51,34 +52,21 @@ class NevergradSearch(SuggestionAlgorithm):
>>> lr = inst.var.Array(1).bounded(1, 2).asfloat()
>>> instrumentation = inst.Instrumentation(lr=lr)
>>> optimizer = optimizerlib.OnePlusOne(instrumentation, budget=100)
>>> algo = NevergradSearch(optimizer, None, max_concurrent=4,
>>> metric="mean_loss", mode="min")
>>> algo = NevergradSearch(
optimizer, None, metric="mean_loss", mode="min")
"""
def __init__(self,
optimizer,
parameter_names,
max_concurrent=10,
reward_attr=None,
metric="episode_reward_mean",
mode="max",
**kwargs):
assert ng is not None, "Nevergrad must be installed!"
assert type(max_concurrent) is int and max_concurrent > 0
assert mode in ["min", "max"], "`mode` must be 'min' or 'max'!"
if reward_attr is not None:
mode = "max"
metric = reward_attr
logger.warning(
"`reward_attr` is deprecated and will be removed in a future "
"version of Tune. "
"Setting `metric={}` and `mode=max`.".format(reward_attr))
self._max_concurrent = max_concurrent
self._parameters = parameter_names
self._metric = metric
# nevergrad.tell internally minimizes, so "max" => -1
if mode == "max":
self._metric_op = -1.
@@ -110,8 +98,6 @@ class NevergradSearch(SuggestionAlgorithm):
"dimension for non-instrumented optimizers")
def suggest(self, trial_id):
if self._num_live_trials() >= self._max_concurrent:
return None
suggested_config = self._nevergrad_opt.ask()
self._live_trial_mapping[trial_id] = suggested_config
# in v0.2.0+, output of ask() is a Candidate,
@@ -122,14 +108,7 @@ class NevergradSearch(SuggestionAlgorithm):
else:
return suggested_config.kwargs
def on_trial_result(self, trial_id, result):
pass
def on_trial_complete(self,
trial_id,
result=None,
error=False,
early_terminated=False):
def on_trial_complete(self, trial_id, result=None, error=False):
"""Notification for the completion of trial.
The result is internally negated when interacting with Nevergrad
@@ -137,20 +116,15 @@ class NevergradSearch(SuggestionAlgorithm):
as it minimizes on default.
"""
if result:
self._process_result(trial_id, result, early_terminated)
self._process_result(trial_id, result)
self._live_trial_mapping.pop(trial_id)
def _process_result(self, trial_id, result, early_terminated=False):
if early_terminated and self._use_early_stopped is False:
return
def _process_result(self, trial_id, result):
ng_trial_info = self._live_trial_mapping[trial_id]
self._nevergrad_opt.tell(ng_trial_info,
self._metric_op * result[self._metric])
def _num_live_trials(self):
return len(self._live_trial_mapping)
def save(self, checkpoint_dir):
trials_object = (self._nevergrad_opt, self._parameters)
with open(checkpoint_dir, "wb") as outputFile: