Merge pull request #198 from cxhernandez/seed-estimator

pass random_seed to estimator

docs/changelog.rst

@@ -16,10 +16,10 @@ New Features
 + Added random seed via CLI (`#196 <https://github.com/msmbuilder/osprey/pull/196>`_)
 + Added ``DSVDatasetLoader`` (`#175 <https://github.com/msmbuilder/osprey/pull/175>`_)
 + Added ``random_seed`` as a configurable parameter (`#164 <https://github.com/msmbuilder/osprey/pull/164>`_)
-+
 
 Bug Fixes
 ~~~~~~~~~~~~
++ Fixed issue where ``random_seed`` was not passed to estimator (`#198 <https://github.com/msmbuilder/osprey/pull/198>`_)
 + Fixed ``bokeh.io.vplot`` deprecation warning (`#192 <https://github.com/msmbuilder/osprey/pull/192>`_)
 + Fixed ungraceful failures when using GP with a single choice in
   search space (`#191 <https://github.com/msmbuilder/osprey/pull/191>`_)

osprey/execute_worker.py

@@ -30,12 +30,16 @@ def execute(args, parser):
     print_header()
 
     config = Config(args.config)
+    random_seed = args.seed if args.seed is not None else config.random_seed()
     estimator = config.estimator()
+    if 'random_state' in estimator.get_params().keys():
+        estimator.set_params(random_state=random_seed)
+    np.random.seed(random_seed)
     searchspace = config.search_space()
     strategy = config.strategy()
     config_sha1 = config.sha1()
     scoring = config.scoring()
-    random_seed = args.seed if args.seed is not None else config.random_seed()
+
     project_name = config.project_name()
 
     if is_msmbuilder_estimator(estimator):
@@ -74,12 +78,11 @@ def signal_hander(signum, frame):
 
         trial_id, params = initialize_trial(
             strategy, searchspace, estimator, config_sha1=config_sha1,
-            project_name=project_name,
-            sessionbuilder=config.trialscontext)
+            project_name=project_name, sessionbuilder=config.trialscontext)
 
         s = run_single_trial(
             estimator=estimator, params=params, trial_id=trial_id,
-            scoring=scoring, random_seed=random_seed, X=X, y=y, cv=cv,
+            scoring=scoring, X=X, y=y, cv=cv,
             sessionbuilder=config.trialscontext)
 
         statuses[i] = s
@@ -123,15 +126,14 @@ def initialize_trial(strategy, searchspace, estimator, config_sha1,
     return trial_id, params
 
 
-def run_single_trial(estimator, params, trial_id, scoring, random_seed,
-                     X, y, cv, sessionbuilder):
+def run_single_trial(estimator, params, trial_id, scoring, X, y, cv,
+                     sessionbuilder):
 
     status = None
 
     try:
         score = fit_and_score_estimator(
-            estimator, params, cv=cv, scoring=scoring, random_seed=random_seed,
-            X=X, y=y, verbose=1)
+            estimator, params, cv=cv, scoring=scoring, X=X, y=y, verbose=1)
         with sessionbuilder() as session:
             trial = session.query(Trial).get(trial_id)
             trial.mean_test_score = score['mean_test_score']

osprey/fit_estimator.py

@@ -19,8 +19,8 @@
 
 
 def fit_and_score_estimator(estimator, parameters, cv, X, y=None, scoring=None,
-                            random_seed=None, iid=True, n_jobs=1,
-                            verbose=1, pre_dispatch='2*n_jobs'):
+                            iid=True, n_jobs=1, verbose=1,
+                            pre_dispatch='2*n_jobs'):
     """Fit and score an estimator with cross-validation
 
     This function is basically a copy of sklearn's
@@ -40,8 +40,6 @@ def fit_and_score_estimator(estimator, parameters, cv, X, y=None, scoring=None,
         score.
     """
 
-    np.random.seed(random_seed)
-
     scorer = check_scoring(estimator, scoring=scoring)
     n_samples = num_samples(X)
     X, y = check_arrays(X, y, allow_lists=True, sparse_format='csr',

osprey/strategies.py

@@ -177,7 +177,8 @@ def _hyperopt_fmin_random_kwarg(random):
 class GP(BaseStrategy):
     short_name = 'gp'
 
-    def __init__(self, seeds=1, max_feval=5E4, max_iter=1E5):
+    def __init__(self, seed=None, seeds=1, max_feval=5E4, max_iter=1E5):
+        self.seed = seed
         self.seeds = seeds
         self.max_feval = max_feval
         self.max_iter = max_iter
@@ -304,4 +305,3 @@ def suggest(self, history, searchspace):
         # so user should pick correctly number of evaluations
         self.current += 1
         return self.param_grid[self.current % len(self.param_grid)]
-