אימון מחודש, 64 + 0.3, לוג משופר

.github/workflows/update_Training.yml

@@ -6,6 +6,7 @@ on:
       - machine-learn
     paths:
       - 'machine-learn/scrape_data/cleaned_new-data.json'
+      - 'machine-learn/creating_model_git.py'
 
 jobs:
   train-ner-model:
@@ -18,11 +19,11 @@ jobs:
       - name: Set up Python environment
         uses: actions/setup-python@v5
         with:
-          python-version: '3.10'
+          python-version: '3.11'
 
       - name: Install dependencies
         run: |
-          pip install spacy==3.7.5 pandas
+          pip install spacy==3.7.5 pandas tqdm
 
       - name: Set working directory
         run: cd machine-learn/

machine-learn/creating_model_git.py

@@ -6,6 +6,18 @@
 from spacy.util import minibatch, compounding
 import json
 import random
+import logging
+from tqdm import tqdm
+
+# הגדרת הלוגר
+logging.basicConfig(
+    level=logging.INFO,
+    format='%(asctime)s - %(levelname)s - %(message)s',
+    handlers=[
+        logging.StreamHandler()
+    ]
+)
+logger = logging.getLogger(__name__)
 
 def custom_tokenizer(nlp):
     default_tokenizer = Tokenizer(nlp.vocab)
@@ -40,87 +52,95 @@ def custom_tokenizer(nlp):
 
     return nlp2.tokenizer
 
-nlp = spacy.blank("he")
-nlp.tokenizer = custom_tokenizer(nlp)
-
-test_text = "תומר כהן- הישראלי הבכיר בלינקדין"
-doc = nlp(test_text)
-print([token.text for token in doc])
-
-ner = nlp.add_pipe("ner")
-ner.add_label("SINGER")
-
-json_files = [
-    '/home/runner/work/Singles-Sorter/Singles-Sorter/machine-learn/scrape_data/cleaned_new-data.json'
-]
-
-training_data = []
-for json_file in json_files:
-    with open(json_file, 'r', encoding='utf-8') as f:
-        data = json.load(f)
-        for example_text, example_entities in data:
-            entities = example_entities.get('entities', [])
-            example = Example.from_dict(nlp.make_doc(example_text), {'entities': entities})
-            training_data.append(example)
-
-
-nlp.begin_training()
-
-patience = 5
-min_delta = 0.01
-best_loss = float('inf')
-patience_counter = 0
-best_model_path = "/home/runner/work/Singles-Sorter/Singles-Sorter/machine-learn/best_model"
-n_iter = 100
-# batch_sizes = compounding(16.0, 64.0, 1.001)
-batch_size = 32
-drop_size = 0.5
-iteration_data = {}
-#initial_lr = 0.001  # שיעור למידה התחלתי
-#lr_decay = 0.95  # קצב דעיכת שיעור הלמידה
-# optimizer.learn_rate = initial_lr
-
-for itn in range(n_iter):
-    random.shuffle(training_data)
-    losses = {}
-    for i in range(0, len(training_data), batch_size):
-        batch = training_data[i:i + batch_size]
-        nlp.update(batch, drop=drop_size, losses=losses)
-    print(f"Iteration {itn}: {losses}")
-    iteration_data[itn] = losses.copy()
-
-    current_loss = losses.get('ner', float('inf'))
-    if current_loss < best_loss - min_delta:
-        best_loss = current_loss
+def main():
+    try:
+        nlp = spacy.blank("he")
+        nlp.tokenizer = custom_tokenizer(nlp)
+
+        test_text = "תומר כהן- הישראלי הבכיר בלינקדין"
+        doc = nlp(test_text)
+        logger.info(f"Tokenized Text: {[token.text for token in doc]}")
+
+        ner = nlp.add_pipe("ner")
+        ner.add_label("SINGER")
+
+        json_files = [
+            '/home/runner/work/Singles-Sorter/Singles-Sorter/machine-learn/scrape_data/cleaned_new-data.json'
+        ]
+
+        training_data = []
+        for json_file in json_files:
+            with open(json_file, 'r', encoding='utf-8') as f:
+                data = json.load(f)
+                for example_text, example_entities in data:
+                    entities = example_entities.get('entities', [])
+                    example = Example.from_dict(nlp.make_doc(example_text), {'entities': entities})
+                    training_data.append(example)
+
+        logger.info(f"Loaded {len(training_data)} training examples.")
+
+        nlp.begin_training()
+
+        patience = 5
+        min_delta = 0.1
+        best_loss = float('inf')
         patience_counter = 0
-        # Save the best model
-        nlp.to_disk(best_model_path)
-    else:
-        patience_counter += 1
+        best_model_path = "/home/runner/work/Singles-Sorter/Singles-Sorter/machine-learn/best_model"
+        n_iter = 60
+        batch_size = 64
+        drop_size = 0.3
+        iteration_data = {}
+
+        logger.info("Starting training...")
+
+        for itn in tqdm(range(n_iter), desc="Training Iterations"):
+            random.shuffle(training_data)
+            losses = {}
+            for i in range(0, len(training_data), batch_size):
+                batch = training_data[i:i + batch_size]
+                nlp.update(batch, drop=drop_size, losses=losses)
+            logger.info(f"Iteration {itn}: {losses}")
+            iteration_data[itn] = losses.copy()
+
+            current_loss = losses.get('ner', float('inf'))
+            if current_loss < best_loss - min_delta:
+                best_loss = current_loss
+                patience_counter = 0
+                # שמירת המודל הטוב ביותר
+                nlp.to_disk(best_model_path)
+                logger.info(f"New best model found at iteration {itn} with loss {current_loss}. Saved to {best_model_path}.")
+            else:
+                patience_counter += 1
+                logger.info(f"No improvement in iteration {itn}. Patience counter: {patience_counter}/{patience}")
+
+            if patience_counter >= patience:
+                logger.info(f"Early stopping at iteration {itn} due to no improvement.")
+                break
+
+        with open("/home/runner/work/Singles-Sorter/Singles-Sorter/machine-learn/model_name.txt", 'r', encoding='utf-8') as f:
+            model_name = f.read().strip()
+            logger.info(f'Final Model Name: {model_name}')
+
+        try:
+            with open('/home/runner/work/Singles-Sorter/Singles-Sorter/machine-learn/iteration_data.json', 'w', encoding='utf-8') as f:
+                json.dump(iteration_data, f, ensure_ascii=False, indent=2)
+            logger.info("Saved iteration data to JSON file.")
+        except Exception as e:
+            logger.error(f'Error saving iteration data to JSON file: {e}')
+
+        # טעינת המודל הטוב ביותר לפני השמירה עם השם הסופי
+        nlp = spacy.load(best_model_path)
+        nlp.meta['name'] = 'singer_ner_he'
+        nlp.meta['description'] = 'Model for recognizing singer names in Hebrew song titles'
+        nlp.meta['author'] = 'nhlocal'
+        nlp.meta['email'] = '[email protected]'
+        nlp.meta['license'] = 'MIT'
+        nlp.meta['tags'] = ['NER', 'Hebrew', 'Singer', 'Named Entity Recognition', 'Text Classification']
+        nlp.to_disk(model_name)
+        logger.info(f"Saved final model to {model_name}.")
 
-    if patience_counter >= patience:
-        print(f"Early stopping at iteration {itn}")
-        break
-
-    # עדכון שיעור הלמידה
-    #optimizer.learn_rate *= lr_decay
-
-with open("/home/runner/work/Singles-Sorter/Singles-Sorter/machine-learn/model_name.txt", 'r', encoding='utf-8') as f:
-    model_name = f.read()
-    print(f'# {model_name}')
-
-try:
-    with open(f'/home/runner/work/Singles-Sorter/Singles-Sorter/machine-learn/iteration_data.json', 'w', encoding='utf-8') as f:
-        json.dump(iteration_data, f, ensure_ascii=False, indent=2)
-except Exception as e:
-    print(f'was error in Save iteration data to a JSON file: {e}')
+    except Exception as e:
+        logger.error(f"An error occurred during training: {e}")
 
-# Load the best model before saving with the final name
-nlp = spacy.load(best_model_path)
-nlp.meta['name'] = 'singer_ner_he'
-nlp.meta['description'] = 'Model for recognizing singer names in Hebrew song titles'
-nlp.meta['author'] = 'nhlocal'
-nlp.meta['email'] = '[email protected]'
-nlp.meta['license'] = 'MIT'
-nlp.meta['tags'] = ['NER', 'Hebrew', 'Singer', 'Named Entity Recognition', 'Text Classification']
-nlp.to_disk(model_name)
+if __name__ == "__main__":
+    main()

machine-learn/model_name.txt

@@ -1 +1 @@
-custom_ner_model23git
+custom_ner_model27-1git