ENAS enable to add None values in algorithm settings (#1153)

* ENAS enable to add None values as algorithm settings

* Delete TODO

pkg/suggestion/v1alpha3/nas/enas/AlgorithmSettings.py

@@ -11,9 +11,10 @@
     "controller_train_steps":       [int, [1, 'inf']],
     "controller_log_every_steps":   [int, [1, 'inf']],
 }
+enableNoneSettingsList = [
+    "controller_temperature", "controller_tanh_const", "controller_entropy_weight", "controller_skip_weight"]
 
 
-# TODO: Enable to add None values, e.g in controller_temperature parameter
 def parseAlgorithmSettings(settings_raw):
 
     algorithm_settings_default = {
@@ -32,7 +33,10 @@ def parseAlgorithmSettings(settings_raw):
     for setting in settings_raw:
         s_name = setting.name
         s_value = setting.value
-        s_type = algorithmSettingsValidator[s_name][0]
-        algorithm_settings_default[s_name] = s_type(s_value)
+        if s_value == "None":
+            algorithm_settings_default[s_name] = None
+        else:
+            s_type = algorithmSettingsValidator[s_name][0]
+            algorithm_settings_default[s_name] = s_type(s_value)
 
     return algorithm_settings_default

pkg/suggestion/v1alpha3/nas/enas/Controller.py

@@ -190,7 +190,7 @@ def build_trainer(self):
         normalize = tf.dtypes.cast((self.num_layers * (self.num_layers - 1) / 2), tf.float32)
         self.skip_rate = tf.dtypes.cast((self.skip_count / normalize), tf.float32)
 
-        if self.controller_entropy_weight:
+        if self.controller_entropy_weight is not None:
             self.reward += self.controller_entropy_weight * self.sample_entropy
 
         self.sample_log_probs = tf.reduce_sum(self.sample_log_probs)

pkg/suggestion/v1alpha3/nas/enas_service.py

@@ -9,7 +9,8 @@
 from pkg.apis.manager.v1alpha3.python import api_pb2_grpc
 from pkg.suggestion.v1alpha3.nas.enas.Controller import Controller
 from pkg.suggestion.v1alpha3.nas.enas.Operation import SearchSpace
-from pkg.suggestion.v1alpha3.nas.enas.AlgorithmSettings import parseAlgorithmSettings, algorithmSettingsValidator
+from pkg.suggestion.v1alpha3.nas.enas.AlgorithmSettings import (
+    parseAlgorithmSettings, algorithmSettingsValidator, enableNoneSettingsList)
 from pkg.suggestion.v1alpha3.base_health_service import HealthServicer
 
 
@@ -122,8 +123,6 @@ def print_algorithm_settings(self):
                     spec, self.algorithm_settings[spec]))
 
         self.logger.info("")
-        self.logger.info("RequestNumber:\t\t\t{}".format(self.num_trials))
-        self.logger.info("")
 
 
 class EnasService(api_pb2_grpc.SuggestionServicer, HealthServicer):
@@ -205,6 +204,8 @@ def ValidateAlgorithmSettings(self, request, context):
         settings_raw = request.experiment.spec.algorithm.algorithm_setting
         for setting in settings_raw:
             if setting.name in algorithmSettingsValidator.keys():
+                if setting.name in enableNoneSettingsList and setting.value == "None":
+                    continue
                 setting_type = algorithmSettingsValidator[setting.name][0]
                 setting_range = algorithmSettingsValidator[setting.name][1]
                 try:
@@ -243,6 +244,10 @@ def GetSuggestions(self, request, context):
         self.logger.info("-" * 100 + "\nSuggestion Step {} for Experiment {}\n".format(
             experiment.suggestion_step, experiment.experiment_name) + "-" * 100)
 
+        self.logger.info("")
+        self.logger.info(">>> RequestNumber:\t\t{}".format(experiment.num_trials))
+        self.logger.info("")
+
         with experiment.tf_graph.as_default():
             saver = tf.compat.v1.train.Saver()
             ctrl = experiment.controller