057: Code Refactoring - Siamese Architectures

2 files changed, 37 insertions, 21 deletions

diff --git a/examples/sentence_level/wmt_2020/ro_en/siamesetransquest.py b/examples/sentence_level/wmt_2020/ro_en/siamesetransquest.py
index e5bdf04..760570f 100755
--- a/examples/sentence_level/wmt_2020/ro_en/siamesetransquest.py
+++ b/examples/sentence_level/wmt_2020/ro_en/siamesetransquest.py
@@ -55,6 +55,9 @@ train = train.rename(columns={'original': 'text_a', 'translation': 'text_b', 'z_
 dev = dev.rename(columns={'original': 'text_a', 'translation': 'text_b', 'z_mean': 'labels'}).dropna()
 test = test.rename(columns={'original': 'text_a', 'translation': 'text_b'}).dropna()
 
+dev_sentence_pairs = list(map(list, zip(dev['text_a'].to_list(), dev['text_b'].to_list())))
+test_sentence_pairs = list(map(list, zip(test['text_a'].to_list(), test['text_b'].to_list())))
+
 train = fit(train, 'labels')
 dev = fit(dev, 'labels')
 
@@ -125,29 +128,31 @@ if siamesetransquest_config["evaluate_during_training"]:
                 inp_example = InputExample(texts=[row['text_a'], row['text_b']], label=score)
                 dev_samples.append(inp_example)
 
-            evaluator = EmbeddingSimilarityEvaluator.from_input_examples(dev_samples, name='dev')
+            evaluator = EmbeddingSimilarityEvaluator.from_input_examples(dev_samples)
             model.evaluate(evaluator,
-                           output_path=os.path.join(siamesetransquest_config['cache_dir'], "dev_result.txt"))
+                           output_path=siamesetransquest_config['cache_dir'])
+            dev_preds[:, i] = model.predict(dev_sentence_pairs)
+            test_preds[:, i] = model.predict(test_sentence_pairs)
 
             # test_data = SentencesDataset(examples=sts_reader.get_examples("test.tsv", test_file=True), model=model)
             # test_dataloader = DataLoader(test_data, shuffle=False, batch_size=8)
 
-            for index, row in test.iterrows():
-                score = random.uniform(0, 1)
-                inp_example = InputExample(texts=[row['text_a'], row['text_b']], label=score)
-                test_samples.append(inp_example)
-
-            evaluator = EmbeddingSimilarityEvaluator.from_input_examples(test_samples, name='dev')
-
-            model.evaluate(evaluator,
-                           output_path=os.path.join(siamesetransquest_config['cache_dir'], "test_result.txt"),
-                           verbose=False)
-
-            with open(os.path.join(siamesetransquest_config['cache_dir'], "dev_result.txt")) as f:
-                dev_preds[:, i] = list(map(float, f.read().splitlines()))
-
-            with open(os.path.join(siamesetransquest_config['cache_dir'], "test_result.txt")) as f:
-                test_preds[:, i] = list(map(float, f.read().splitlines()))
+            # for index, row in test.iterrows():
+            #     score = random.uniform(0, 1)
+            #     inp_example = InputExample(texts=[row['text_a'], row['text_b']], label=score)
+            #     test_samples.append(inp_example)
+            #
+            # evaluator = EmbeddingSimilarityEvaluator.from_input_examples(test_samples)
+            #
+            # model.evaluate(evaluator,
+            #                output_path=siamesetransquest_config['cache_dir'],
+            #                verbose=False)
+
+            # with open(os.path.join(siamesetransquest_config['cache_dir'], "dev_result.txt")) as f:
+            #     dev_preds[:, i] = list(map(float, f.read().splitlines()))
+            #
+            # with open(os.path.join(siamesetransquest_config['cache_dir'], "test_result.txt")) as f:
+            #     test_preds[:, i] = list(map(float, f.read().splitlines()))
 
         dev['predictions'] = dev_preds.mean(axis=1)
         test['predictions'] = test_preds.mean(axis=1)
diff --git a/transquest/algo/sentence_level/siamesetransquest/run_model.py b/transquest/algo/sentence_level/siamesetransquest/run_model.py
index 33fdb91..26ab4b6 100644
--- a/transquest/algo/sentence_level/siamesetransquest/run_model.py
+++ b/transquest/algo/sentence_level/siamesetransquest/run_model.py
@@ -10,6 +10,7 @@ import numpy as np
 from numpy import ndarray
 import transformers
 import torch
+from sklearn.metrics.pairwise import paired_cosine_distances
 from torch import nn, Tensor, device
 from torch.optim.optimizer import Optimizer
 
@@ -227,7 +228,20 @@ class SiameseTransQuestModel(nn.Sequential):
 
         return all_embeddings
 
+    def predict(self, to_predict):
+        sentences1 = []
+        sentences2 = []
 
+        for text_1, text_2 in to_predict:
+            sentences1.append(text_1)
+            sentences2.append(text_2)
+
+        embeddings1 = self.encode(sentences1, show_progress_bar=self.show_progress_bar, convert_to_numpy=True)
+        embeddings2 = self.encode(sentences2, show_progress_bar=self.show_progress_bar, convert_to_numpy=True)
+
+        cosine_scores = 1 - (paired_cosine_distances(embeddings1, embeddings2))
+
+        return cosine_scores
 
     def start_multi_process_pool(self, target_devices: List[str] = None):
         """
@@ -259,7 +273,6 @@ class SiameseTransQuestModel(nn.Sequential):
 
         return {'input': input_queue, 'output': output_queue, 'processes': processes}
 
-
     @staticmethod
     def stop_multi_process_pool(pool):
         """
@@ -275,7 +288,6 @@ class SiameseTransQuestModel(nn.Sequential):
         pool['input'].close()
         pool['output'].close()
 
-
     def encode_multi_process(self, sentences: List[str], pool: Dict[str, object], batch_size: int = 32, chunk_size: int = None):
         """
         This method allows to run encode() on multiple GPUs. The sentences are chunked into smaller packages
@@ -616,7 +628,6 @@ class SiameseTransQuestModel(nn.Sequential):
                 if save_best_model:
                     self.save(output_path)
 
-
     @staticmethod
     def _get_scheduler(optimizer, scheduler: str, warmup_steps: int, t_total: int):
         """