yujunjun/autogen
1
0
Fork 0
mirror of https://github.com/microsoft/autogen.git synced 2025-10-31 01:40:58 +00:00
Code Issues Packages Projects Releases Wiki Activity

constraints (#88)

Browse Source 
* pre-training constraints

* metric constraints after training
This commit is contained in:
Chi Wang 2021-05-18 15:57:42 -07:00 committed by GitHub
parent 3083229e40
commit 0925e2b308
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
9 changed files with 137 additions and 87 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
flaml/automl.py
View File

@ -1036,9 +1036,8 @@ class AutoML:
prune_attr=prune_attr,
min_resource=min_resource,
max_resource=max_resource,
resources_per_trial={"cpu": self._state.n_jobs,
"mem": self._mem_thres},
mem_size=learner_class.size)
config_constraints=[(learner_class.size, '<=', self._mem_thres)]
)
else:
algo = SearchAlgo(
metric='val_loss', mode='min', space=search_space,

5
flaml/data.py
View File

@ -237,8 +237,8 @@ class DataTransformer:
SimpleImputer(missing_values=np.nan, strategy='median'),
X_num.columns)])
X[num_columns] = self.transformer.fit_transform(X_num)
self._cat_columns, self._num_columns, self._datetime_columns = cat_columns, \
num_columns, datetime_columns
self._cat_columns, self._num_columns, self._datetime_columns = \
cat_columns, num_columns, datetime_columns
self._drop = drop
if task == 'regression':
@ -275,4 +275,3 @@ class DataTransformer:
X_num.columns = range(X_num.shape[1])
X[num_columns] = self.transformer.transform(X_num)
return X

71
flaml/searcher/blendsearch.py
View File

@ -39,9 +39,11 @@ class BlendSearch(Searcher):
min_resource: Optional[float] = None,
max_resource: Optional[float] = None,
reduction_factor: Optional[float] = None,
resources_per_trial: Optional[dict] = None,
global_search_alg: Optional[Searcher] = None,
mem_size: Callable[[dict], float] = None,
config_constraints: Optional[
List[Tuple[Callable[[dict], float], str, float]]] = None,
metric_constraints: Optional[
List[Tuple[str, str, float]]] = None,
seed: Optional[int] = 20):
'''Constructor
@ -82,14 +84,23 @@ class BlendSearch(Searcher):
prune_attr; only valid if prune_attr is not in space.
reduction_factor: A float of the reduction factor used for
incremental pruning.
resources_per_trial: A dictionary of the resources permitted per
trial, such as 'mem'.
global_search_alg: A Searcher instance as the global search
instance. If omitted, Optuna is used. The following algos have
known issues when used as global_search_alg:
- HyperOptSearch raises exception sometimes
- TuneBOHB has its own scheduler
mem_size: A function to estimate the memory size for a given config.
config_constraints: A list of config constraints to be satisfied.
e.g.,
.. code-block: python
config_constraints = [(mem_size, '<=', 1024**3)]
mem_size is a function which produces a float number for the bytes
needed for a config.
It is used to skip configs which do not fit in memory.
metric_constraints: A list of metric constraints to be satisfied.
e.g., `['precision', '>=', 0.9]`
seed: An integer of the random seed.
'''
self._metric, self._mode = metric, mode
@ -104,10 +115,8 @@ class BlendSearch(Searcher):
self._ls = LocalSearch(
init_config, metric, mode, cat_hp_cost, space,
prune_attr, min_resource, max_resource, reduction_factor, seed)
self._resources_per_trial = resources_per_trial
self._mem_size = mem_size
self._mem_threshold = resources_per_trial.get(
'mem') if resources_per_trial else None
self._config_constraints = config_constraints
self._metric_constraints = metric_constraints
self._init_search()
def set_search_properties(self,
@ -171,9 +180,8 @@ class BlendSearch(Searcher):
self._points_to_evaluate = state._points_to_evaluate
self._gs = state._gs
self._ls = state._ls
self._resources_per_trial = state._resources_per_trial
self._mem_size = state._mem_size
self._mem_threshold = state._mem_threshold
self._config_constraints = state._config_constraints
self._metric_constraints = state._metric_constraints
def restore_from_dir(self, checkpoint_dir: str):
super.restore_from_dir(checkpoint_dir)
@ -182,6 +190,20 @@ class BlendSearch(Searcher):
error: bool = False):
''' search thread updater and cleaner
'''
if result and not error and self._metric_constraints:
# accout for metric constraints if any
objective = result[self._metric]
for constraint in self._metric_constraints:
metric_constraint, sign, threshold = constraint
value = result.get(metric_constraint)
if value:
# sign is <= or >=
sign_op = 1 if sign == '<=' else -1
violation = (value - threshold) * sign_op
if violation > 0:
# add penalty term to the metric
objective += 1e+10 * violation * self._ls.metric_op
result[self._metric] = objective
thread_id = self._trial_proposed_by.get(trial_id)
if thread_id in self._search_thread_pool:
self._search_thread_pool[thread_id].on_trial_complete(
@ -197,14 +219,15 @@ class BlendSearch(Searcher):
else: # add to result cache
self._result[self._ls.config_signature(config)] = result
# update target metric if improved
if (result[self._metric] - self._metric_target) * self._ls.metric_op < 0:
self._metric_target = result[self._metric]
objective = result[self._metric]
if (objective - self._metric_target) * self._ls.metric_op < 0:
self._metric_target = objective
if not thread_id and self._create_condition(result):
# thread creator
self._search_thread_pool[self._thread_count] = SearchThread(
self._ls.mode,
self._ls.create(config, result[self._metric], cost=result[
self.cost_attr])
self._ls.create(
config, objective, cost=result[self.cost_attr])
)
thread_id = self._thread_count
self._thread_count += 1
@ -362,20 +385,26 @@ class BlendSearch(Searcher):
return config
def _should_skip(self, choice, trial_id, config) -> bool:
''' if config is None or config's result is known or above mem threshold
''' if config is None or config's result is known or constraints are violated
return True; o.w. return False
'''
if config is None:
return True
config_signature = self._ls.config_signature(config)
exists = config_signature in self._result
# check mem constraint
if not exists and self._mem_threshold and self._mem_size(
config) > self._mem_threshold:
# check constraints
if not exists and self._config_constraints:
for constraint in self._config_constraints:
func, sign, threshold = constraint
value = func(config)
if (sign == '<=' and value > threshold
or sign == '>=' and value < threshold):
self._result[config_signature] = {
self._metric: np.inf * self._ls.metric_op, 'time_total_s': 1
self._metric: np.inf * self._ls.metric_op,
'time_total_s': 1,
}
exists = True
break
if exists:
if not self._use_rs:
result = self._result.get(config_signature)

27
flaml/tune/tune.py
View File

@ -3,7 +3,7 @@
* Licensed under the MIT License. See LICENSE file in the
* project root for license information.
'''
from typing import Optional, Union, List, Callable
from typing import Optional, Union, List, Callable, Tuple
import datetime
import time
try:
@ -118,7 +118,10 @@ def run(training_function,
local_dir: Optional[str] = None,
num_samples: Optional[int] = 1,
resources_per_trial: Optional[dict] = None,
mem_size: Callable[[dict], float] = None,
config_constraints: Optional[
List[Tuple[Callable[[dict], float], str, float]]] = None,
metric_constraints: Optional[
List[Tuple[str, str, float]]] = None,
use_ray: Optional[bool] = False):
'''The trigger for HPO.
@ -210,11 +213,19 @@ def run(training_function,
used; or a local dir to save the tuning log.
num_samples: An integer of the number of configs to try. Defaults to 1.
resources_per_trial: A dictionary of the hardware resources to allocate
per trial, e.g., `{'mem': 1024**3}`. When not using ray backend,
only 'mem' is used as approximate resource constraints
(in conjunction with mem_size).
mem_size: A function to estimate the memory size for a given config.
per trial, e.g., `{'cpu': 1}`. Only valid when using ray backend.
config_constraints: A list of config constraints to be satisfied.
e.g.,
.. code-block: python
config_constraints = [(mem_size, '<=', 1024**3)]
mem_size is a function which produces a float number for the bytes
needed for a config.
It is used to skip configs which do not fit in memory.
metric_constraints: A list of metric constraints to be satisfied.
e.g., `['precision', '>=', 0.9]`
use_ray: A boolean of whether to use ray as the backend
'''
global _use_ray
@ -252,8 +263,8 @@ def run(training_function,
prune_attr=prune_attr,
min_resource=min_resource, max_resource=max_resource,
reduction_factor=reduction_factor,
resources_per_trial=resources_per_trial,
mem_size=mem_size)
config_constraints=config_constraints,
metric_constraints=metric_constraints)
if time_budget_s:
search_alg.set_search_properties(metric, mode, config={
'time_budget_s': time_budget_s})

2
flaml/version.py
View File

@ -1 +1 @@
__version__ = "0.3.6"
__version__ = "0.4.0"

4
test/test_automl.py
View File

@ -237,9 +237,11 @@ class TestAutoML(unittest.TestCase):
fake_df = pd.DataFrame({'A': [datetime(1900, 2, 3), datetime(1900, 3, 4)]})
y = np.array([0, 1])
automl_experiment.fit(X_train=fake_df, X_val=fake_df, y_train=y, y_val=y, **automl_settings)
automl_experiment.fit(
X_train=fake_df, X_val=fake_df, y_train=y, y_val=y, **automl_settings)
y_pred = automl_experiment.predict(fake_df)
print(y_pred)
def test_micro_macro_f1(self):
automl_experiment = AutoML()

0
test/tune/__init__.py Normal file
View File

26
test/test_tune.py → test/tune/test_tune.py
View File

@ -1,19 +1,21 @@
'''Require: pip install flaml[test,ray]
'''
import unittest
import time
import os
from sklearn.model_selection import train_test_split
import sklearn.metrics
import sklearn.datasets
try:
from ray.tune.integration.xgboost import TuneReportCheckpointCallback
except ImportError:
print("skip test_tune because ray tune cannot be imported.")
print("skip test_xgboost because ray tune cannot be imported.")
import xgboost as xgb
import logging
logger = logging.getLogger(__name__)
logger.addHandler(logging.FileHandler('test/tune_xgboost.log'))
os.makedirs('logs', exist_ok=True)
logger.addHandler(logging.FileHandler('logs/tune_xgboost.log'))
logger.setLevel(logging.INFO)
def train_breast_cancer(config: dict):
@ -61,6 +63,7 @@ def _test_xgboost(method='BlendSearch'):
for n_cpu in [8]:
start_time = time.time()
ray.init(num_cpus=n_cpu, num_gpus=0)
# ray.init(address='auto')
if method == 'BlendSearch':
analysis = tune.run(
train_breast_cancer,
@ -163,21 +166,28 @@ def test_nested():
}
def simple_func(config):
tune.report(metric=(config["cost_related"]["a"] - 4)**2
* (config["b"] - 0.7)**2)
obj = (config["cost_related"]["a"] - 4)**2 \
+ (config["b"] - config["cost_related"]["a"])**2
tune.report(obj=obj)
tune.report(obj=obj, ab=config["cost_related"]["a"] * config["b"])
tune.run(
analysis = tune.run(
simple_func,
config=search_space,
low_cost_partial_config={
"cost_related": {"a": 1}
},
metric="metric",
metric="obj",
mode="min",
metric_constraints=[("ab", "<=", 4)],
local_dir='logs/',
num_samples=-1,
time_budget_s=1)
best_trial = analysis.get_best_trial()
logger.info(f"Best config: {best_trial.config}")
logger.info(f"Best result: {best_trial.last_result}")
def test_xgboost_bs():
_test_xgboost()
@ -224,4 +234,4 @@ def _test_xgboost_bohb():
if __name__ == "__main__":
unittest.main()
test_xgboost_bs()