diff options
Diffstat (limited to 'lib')
| -rw-r--r-- | lib/loader.py | 2 | ||||
| -rw-r--r-- | lib/model.py | 41 |
2 files changed, 40 insertions, 3 deletions
diff --git a/lib/loader.py b/lib/loader.py index 14b7853..5e9b20a 100644 --- a/lib/loader.py +++ b/lib/loader.py @@ -1974,8 +1974,10 @@ class KConfigAttributes: self.choice[choice.name] = choice self.data = list() + self.configs = list() for config_path, attr_path in experiments: + self.configs.append(config_path) kconf.load_config(config_path) with open(attr_path, "r") as f: attr = json.load(f) diff --git a/lib/model.py b/lib/model.py index f330327..f422204 100644 --- a/lib/model.py +++ b/lib/model.py @@ -1158,7 +1158,11 @@ class PTAModel: class KConfigModel: - """Decision-Tree Model for a specific system attribute such as ROM or RAM usage""" + """ + Decision-Tree Model for a specific system attribute such as ROM or RAM usage. + + See Guo et al: "Data-efficient performance learning for configurable systems", 2017 + """ class Node: pass @@ -1280,6 +1284,8 @@ class KConfigModel: kconf_choice = next( filter(lambda choice: choice.name == self.symbol, kconf.choices) ) + if kconf_choice.selection is None: + return None selection = kconf_choice.selection.name if selection in self.choice: return self.choice[selection].model(kconf) @@ -1299,9 +1305,14 @@ class KConfigModel: return ret @classmethod - def from_benchmark(cls, kconfig_benchmark, attribute): + def from_benchmark(cls, kconfig_benchmark, attribute, indices=None): self = cls() - self.data = kconfig_benchmark.data + if indices is None: + self.data = kconfig_benchmark.data + else: + self.data = list() + for i in indices: + self.data.append(kconfig_benchmark.data[i]) self.symbols = kconfig_benchmark.symbol_names self.choices = kconfig_benchmark.choice_names self.symbol = kconfig_benchmark.symbol @@ -1459,3 +1470,27 @@ class KConfigModel: ) return node + + def assess_benchmark(self, kconfig_benchmark, indices=None): + if indices is None: + indices = range(len(kconfig_benchmark.data)) + + kconf = kconfig_benchmark.kconf + + predictions = list() + values = list() + unpredictable_count = 0 + + for i in indices: + kconf.load_config(kconfig_benchmark.configs[i]) + prediction = self.model.model(kconf) + + if prediction is None: + unpredictable_count += 1 + else: + predictions.append(self.model.model(kconf)) + values.append(self.attr_function(kconfig_benchmark.data[i])) + + measures = regression_measures(np.array(predictions), np.array(values)) + measures["unpredictable_count"] = unpredictable_count + return measures |