feat: added sequential

Author: Darinochka

autointent/generation/utterances/evolution/cli.py

@@ -45,6 +45,14 @@ def _parse_args() -> Namespace:
     parser.add_argument("--seed", type=int, default=0)
     parser.add_argument("--batch-size", type=int, default=4)
     parser.add_argument("--search-space", type=str, default=None)
+    parser.add_argument(
+        "--sequential",
+        action="store_true",
+        help=(
+            "Use sequential evolution. When this option is enabled, solutions "
+            "will evolve one after another, instead of using a parallel approach."
+        ),
+    )
 
     return parser.parse_args()
 
@@ -64,7 +72,11 @@ def main() -> None:
     n_before = len(dataset[args.split])
 
     new_samples = utterance_evolver.augment(
-        dataset, split_name=args.split, n_evolutions=args.n_evolutions, batch_size=args.batch_size
+        dataset,
+        split_name=args.split,
+        n_evolutions=args.n_evolutions,
+        batch_size=args.batch_size,
+        sequential=args.sequential,
     )
     n_after = len(dataset[args.split])
 

autointent/generation/utterances/evolution/evolver.py

@@ -51,9 +51,21 @@ async def _evolve_async(self, utterance: str, intent_data: Intent) -> str:
         chat = maker(utterance, intent_data)
         return await self.generator.get_chat_completion_async(chat)
 
-    def __call__(self, utterance: str, intent_data: Intent, n_evolutions: int = 1) -> list[str]:
+    def __call__(
+        self, utterance: str, intent_data: Intent, n_evolutions: int = 1, sequential: bool = False
+    ) -> list[str]:
         """Apply evolutions multiple times (synchronously)."""
-        return [self._evolve(utterance, intent_data) for _ in range(n_evolutions)]
+        current_utterance = utterance
+        generated_utterances = []
+
+        for _ in range(n_evolutions):
+            gen_utt = self._evolve(current_utterance, intent_data)
+            generated_utterances.append(gen_utt)
+
+            if sequential:
+                current_utterance = gen_utt
+
+        return generated_utterances
 
     def augment(
         self,
@@ -62,13 +74,18 @@ def augment(
         n_evolutions: int = 1,
         update_split: bool = True,
         batch_size: int = 4,
+        sequential: bool = False,
     ) -> HFDataset:
         """
         Augment some split of dataset.
 
         Note that for now it supports only single-label datasets.
         """
         if self.async_mode:
+            if sequential:
+                error = "Sequential and async modes are not compatible"
+                raise ValueError(error)
+
             return asyncio.run(
                 self._augment_async(
                     dataset=dataset,
@@ -85,7 +102,9 @@ def augment(
             utterance = sample[Dataset.utterance_feature]
             label = sample[Dataset.label_feature]
             intent_data = next(intent for intent in dataset.intents if intent.id == label)
-            generated_utterances = self(utterance=utterance, intent_data=intent_data, n_evolutions=n_evolutions)
+            generated_utterances = self(
+                utterance=utterance, intent_data=intent_data, n_evolutions=n_evolutions, sequential=sequential
+            )
             new_samples.extend(
                 [{Dataset.label_feature: intent_data.id, Dataset.utterance_feature: ut} for ut in generated_utterances]
             )

autointent/generation/utterances/evolution/incremental_evolver.py

@@ -67,6 +67,7 @@ def augment(
         n_evolutions: int = 1,
         update_split: bool = True,
         batch_size: int = 4,
+        sequential: bool = False,
     ) -> HFDataset:
         """
         Augment some split of dataset.
@@ -79,7 +80,12 @@ def augment(
 
         for _ in range(n_evolutions):
             new_samples_dataset = super().augment(
-                dataset, split_name=split_name, n_evolutions=1, update_split=False, batch_size=batch_size
+                dataset,
+                split_name=split_name,
+                n_evolutions=1,
+                update_split=False,
+                batch_size=batch_size,
+                sequential=sequential,
             )
             merge_dataset[split_name] = concatenate_datasets([merge_dataset[split_name], new_samples_dataset])
             generated_samples.append(new_samples_dataset)