quick-sentiments 0.4.4__tar.gz → 0.4.6__tar.gz

{quick_sentiments-0.4.4/quick_sentiments.egg-info → quick_sentiments-0.4.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: quick-sentiments
-Version: 0.4.4
+Version: 0.4.6
 Summary: Sentiment Analysis pipeline
 Author: Alabhya Dahal
 Author-email: Alabhya Dahal <alabhya.dahal@gmail.com>
@@ -92,8 +92,9 @@ This Python package is designed to streamline natural language processing (NLP)
   - Naive Bayesian
   - Neural Network
   - Naive Bayes
+  - Tensorflow Models
 - **Hyperparameter Tuning Support**:
-  - All models are compatible with GridSearchCV.
+  - All models are compatible with GridSearchCV. Tensorflow has it's own tuning
   - By default, models run with standard parameters for quick testing.
   - Grid search options are built-in and ready to use if needed.
 - **Modular Design**: Each component is cleanly separated into its own module.

{quick_sentiments-0.4.4 → quick_sentiments-0.4.6}/README.md

@@ -46,8 +46,9 @@ This Python package is designed to streamline natural language processing (NLP)
   - Naive Bayesian
   - Neural Network
   - Naive Bayes
+  - Tensorflow Models
 - **Hyperparameter Tuning Support**:
-  - All models are compatible with GridSearchCV.
+  - All models are compatible with GridSearchCV. Tensorflow has it's own tuning
   - By default, models run with standard parameters for quick testing.
   - Grid search options are built-in and ready to use if needed.
 - **Modular Design**: Each component is cleanly separated into its own module.

{quick_sentiments-0.4.4 → quick_sentiments-0.4.6}/pyproject.toml

@@ -5,7 +5,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "quick-sentiments"
-version = "0.4.4" # tensorflow enters !
+version = "0.4.6" # probablities return for evaluate
 authors = [
   { name="Alabhya Dahal", email="alabhya.dahal@gmail.com" },
 ]

{quick_sentiments-0.4.4 → quick_sentiments-0.4.6}/quick_sentiments/__init__.py

@@ -3,6 +3,12 @@ from .pipeline import run_pipeline       # Expose pipeline function
 from .predict import make_predictions # Expose prediction function
 from .preprocess import pre_process_nltk  # Expose NLTK preprocessing function
 from .spacy_preprocess import pre_process_spacy # Expose spaCy preprocessing function
-
-__all__ = ['run_pipeline', 'make_predictions']  # Controls what's available in *
+from .evaluate_performance import evaluate_performance # Expose evaluation function
+__all__ = [
+    'run_pipeline', 
+    'make_predictions',
+    'pre_process_nltk',
+    'pre_process_spacy',
+    'evaluate_performance'
+]
 

quick_sentiments-0.4.6/quick_sentiments/evalute_performance.py

@@ -0,0 +1,69 @@
+import numpy as np
+import pandas as pd
+from sklearn.metrics import roc_curve, precision_recall_curve, accuracy_score
+
+def evaluate_performance(y_true, y_prob, positive_label=1):
+    """
+    Evaluates model performance across different probability thresholds.
+    
+    Args:
+        y_true (list or np.array): The actual ground truth labels.
+        y_prob (list or np.array): The predicted probabilities for the positive class.
+        positive_label (int/str): The value representing the positive class (default is 1).
+        
+    Returns:
+        dict: Contains optimal thresholds and a pandas DataFrame of the decile summary.
+    """
+    y_true = np.array(y_true)
+    y_prob = np.array(y_prob)
+    
+    # 1. Best ROC Threshold (Youden's J statistic: TPR - FPR)
+    fpr, tpr, roc_thresh = roc_curve(y_true, y_prob, pos_label=positive_label)
+    youden_j = tpr - fpr
+    best_roc_idx = np.argmax(youden_j)
+    best_roc_threshold = roc_thresh[best_roc_idx]
+    
+    # 2. Best PR Threshold (Max F1-Score)
+    precision, recall, pr_thresh = precision_recall_curve(y_true, y_prob, pos_label=positive_label)
+    # Ignore warnings for 0 division
+    with np.errstate(divide='ignore', invalid='ignore'):
+        f1_scores = 2 * (precision * recall) / (precision + recall)
+    f1_scores = np.nan_to_num(f1_scores) # Convert NaNs to 0
+    best_pr_idx = np.argmax(f1_scores)
+    # precision_recall_curve returns thresholds len - 1, so we handle the index safely
+    best_pr_threshold = pr_thresh[best_pr_idx] if best_pr_idx < len(pr_thresh) else 1.0
+
+    # 3. Decile Summary Table (0.0 to 1.0)
+    summary_data = []
+    thresholds = np.arange(0.0, 1.1, 0.1)
+    
+    for th in thresholds:
+        # Convert probabilities to hard predictions based on current threshold
+        y_pred = (y_prob >= th).astype(int)
+        is_pos = (y_true == positive_label).astype(int)
+        
+        tp = np.sum((y_pred == 1) & (is_pos == 1))
+        fp = np.sum((y_pred == 1) & (is_pos == 0))
+        tn = np.sum((y_pred == 0) & (is_pos == 0))
+        fn = np.sum((y_pred == 0) & (is_pos == 1))
+        
+        acc = (tp + tn) / len(y_true)
+        prec = tp / (tp + fp) if (tp + fp) > 0 else 0.0
+        rec = tp / (tp + fn) if (tp + fn) > 0 else 0.0
+        f1 = 2 * (prec * rec) / (prec + rec) if (prec + rec) > 0 else 0.0
+        
+        summary_data.append({
+            "Threshold": round(th, 1),
+            "Accuracy": round(acc, 3),
+            "Precision": round(prec, 3),
+            "Recall": round(rec, 3),
+            "F1": round(f1, 3)
+        })
+        
+    df_summary = pd.DataFrame(summary_data)
+    
+    return {
+        "best_roc_threshold": round(best_roc_threshold, 4),
+        "best_pr_threshold": round(best_pr_threshold, 4),
+        "decile_table": df_summary
+    }

{quick_sentiments-0.4.4 → quick_sentiments-0.4.6}/quick_sentiments/pipeline.py

@@ -205,6 +205,7 @@ def run_pipeline(
                                                               X_test_vectorized, 
                                                               perform_tuning=perform_tuning,
                                                               random_state=random_state)
+    y_prob = trained_model_object.predict_proba(X_test_vectorized)[:, 1]
 
     # Evaluate
     print("4. Evaluating model...")
@@ -221,6 +222,7 @@ def run_pipeline(
         "label_encoder": label_encoder,
         "y_test": y_test,
         "y_pred": y_pred,
+        "y_prob": y_prob,
         "accuracy": accuracy_score(y_test, y_pred),
         "report": classification_report(y_test, y_pred, output_dict=True, target_names=label_encoder.classes_)
     }

{quick_sentiments-0.4.4 → quick_sentiments-0.4.6/quick_sentiments.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: quick-sentiments
-Version: 0.4.4
+Version: 0.4.6
 Summary: Sentiment Analysis pipeline
 Author: Alabhya Dahal
 Author-email: Alabhya Dahal <alabhya.dahal@gmail.com>
@@ -92,8 +92,9 @@ This Python package is designed to streamline natural language processing (NLP)
   - Naive Bayesian
   - Neural Network
   - Naive Bayes
+  - Tensorflow Models
 - **Hyperparameter Tuning Support**:
-  - All models are compatible with GridSearchCV.
+  - All models are compatible with GridSearchCV. Tensorflow has it's own tuning
   - By default, models run with standard parameters for quick testing.
   - Grid search options are built-in and ready to use if needed.
 - **Modular Design**: Each component is cleanly separated into its own module.

{quick_sentiments-0.4.4 → quick_sentiments-0.4.6}/quick_sentiments.egg-info/SOURCES.txt

@@ -3,6 +3,7 @@ README.md
 pyproject.toml
 setup.py
 quick_sentiments/__init__.py
+quick_sentiments/evalute_performance.py
 quick_sentiments/pipeline.py
 quick_sentiments/predict.py
 quick_sentiments/preprocess.py
@@ -19,7 +20,7 @@ quick_sentiments/ml_algo/logit.py
 quick_sentiments/ml_algo/nb.py
 quick_sentiments/ml_algo/nn.py
 quick_sentiments/ml_algo/rf.py
-quick_sentiments/ml_algo/ts_model.py
+quick_sentiments/ml_algo/tf_model.py
 quick_sentiments/vect/BOW.py
 quick_sentiments/vect/__init__.py
 quick_sentiments/vect/glove_100.py