pyerualjetwork 2.2.7__py3-none-any.whl → 2.2.9__py3-none-any.whl

plan_bi/__init__.py

@@ -2,4 +2,4 @@
 
 # Bu dosya, plan modülünün ana giriş noktasıdır.
 
-from .plan_bi import auto_balancer, normalization, Softmax, Sigmoid, Relu, weight_identification, fex, fit, evaluate, save_model, load_model, predict_model_ssd, predict_model_ram, get_weights, get_df, get_preds, get_acc, get_pot, synthetic_augmentation, standard_scaler, multiple_evaluate, encode_one_hot, split, metrics, decode_one_hot, roc_curve, confusion_matrix, plot_evaluate
+from .plan_bi import auto_balancer, normalization, Softmax, Sigmoid, Relu, weight_identification, fex, fit, evaluate, save_model, load_model, predict_model_ssd, predict_model_ram, get_weights, get_df, get_preds, get_acc, get_pot, synthetic_augmentation, standard_scaler, multiple_evaluate, encode_one_hot, split, metrics, decode_one_hot, roc_curve, confusion_matrix, plot_evaluate, manuel_balancer

plan_bi/plan_bi.py

@@ -52,7 +52,7 @@ def fit(
     y_train = [tuple(sublist) for sublist in y_train]
     
     neurons = [len(class_count),len(class_count)]
-    layers = ['fex','cat']
+    layers = ['fex']
     
     x_train[0] = np.array(x_train[0])
     x_train[0] = x_train[0].ravel()
@@ -123,8 +123,17 @@ def fit(
         print('Total training time(h): ',calculating_est/3600)
     
     
+    layers.append('cat')
+    trained_W.append(np.eye(len(class_count)))
     
+    for i in range(len(class_count)):
 
+        for j in range(len(layers)):
+    
+            if layers[j] == 'fex':
+        
+                trained_W[j][i,:] = normalization(trained_W[j][i,:])
+    
     return trained_W
         
 # FUNCTIONS -----
@@ -155,8 +164,6 @@ def weight_identification(
     ws = layer_count - 1
     for w in range(ws):
         W[w + 1] = np.ones((neurons[w + 1],neurons[w]))
-    W[layer_count] = np.ones((class_count,neurons[layer_count - 1]))
-    W[len(W) - 1] = np.eye(class_count)
 
     return W
         
@@ -529,7 +536,7 @@ def save_model(model_name,
              weights_type,
              weights_format,
              model_path,
-             scaler,
+             scaler_params,
              W
  ):
     
@@ -545,7 +552,7 @@ def save_model(model_name,
     weights_type (str): Type of weights to save (options: 'txt', 'npy', 'mat').
     WeightFormat (str): Format of the weights (options: 'd', 'f', 'raw').
     model_path (str): Path where the model will be saved. For example: C:/Users/beydili/Desktop/denemePLAN/
-    scaler (bool): trained data it used standard_scaler ? (True or False)
+    scaler_params (int, float): standard scaler params list: mean,std. If not used standard scaler then be: None.
     W: Weights list of the model.
     
     Returns:
@@ -593,7 +600,7 @@ def save_model(model_name,
             'WEIGHTS TYPE': weights_type,
             'WEIGHTS FORMAT': weights_format,
             'MODEL PATH': model_path,
-            'STANDARD SCALER': scaler
+            'STANDARD SCALER': scaler_params
             }
     try:
         
@@ -743,11 +750,11 @@ def predict_model_ssd(Input, model_name, model_path):
     """
     W, activation_potential, df = load_model(model_name,model_path)
     
-    scaler = str(df['STANDARD SCALER'].iloc[0])
+    scaler_params = str(df['STANDARD SCALER'].iloc[0])
 
-    if scaler == 'True':
+    if scaler_params != None:
 
-        Input = standard_scaler(Input, None)
+        Input = standard_scaler(scaler_params, Input, None)
     
     layers = ['fex','cat']
     
@@ -774,7 +781,7 @@ def predict_model_ssd(Input, model_name, model_path):
     return neural_layer
 
 
-def predict_model_ram(Input, activation_potential, scaler, W):
+def predict_model_ram(Input, activation_potential, scaler_params, W):
     
     infopredict_model_ram = """
     Function to make a prediction using a divided pruning learning artificial neural network (PLAN).
@@ -783,15 +790,15 @@ def predict_model_ram(Input, activation_potential, scaler, W):
     Arguments:
     Input (list or ndarray): Input data for the model (single vector or single matrix).
     activation_potential (float): Activation potential.
-    scaler (bool): trained data it used standard_scaler ? (True or False)
+    scaler_params (int, float): standard scaler params list: mean,std. If not used standard scaler then be: None.
     W (list of ndarrays): Weights of the model.
 
     Returns:
     ndarray: Output from the model.
     """
-    if scaler == True:
+    if scaler_params != None:
 
-       Input = standard_scaler(Input, None)
+       Input = standard_scaler(scaler_params, Input, None)
     
     layers = ['fex','cat']
     
@@ -918,7 +925,7 @@ def synthetic_augmentation(x_train, y_train):
     return np.array(x_balanced), np.array(y_balanced)
 
 
-def standard_scaler(x_train, x_test):
+defdef standard_scaler(scaler_params=None, x_train, x_test):
     info_standard_scaler = """
   Standardizes training and test datasets. x_test may be None.
 
@@ -929,23 +936,31 @@ def standard_scaler(x_train, x_test):
       Test data
 
   Returns:
+    list:
+    Scaler parameters: mean and std
     tuple
       Standardized training and test datasets
   """
-    try:
-
-        mean = np.mean(x_train, axis=0)
-        std = np.std(x_train, axis=0)
+        try:
+            
+            if scaler_params == None:
+                mean = np.mean(x_train, axis=0)
+                std = np.std(x_train, axis=0)
+                scaler_params = [mean, std]
 
-        if x_test == None:
+            if x_test == None:
 
-            train_data_scaled = (x_train - mean) / std
-            return train_data_scaled
+                train_data_scaled = (x_train - mean) / std
+                return scaler_params, train_data_scaled
 
-        else:
-            train_data_scaled = (x_train - mean) / std
-            test_data_scaled = (x_test - mean) / std
-            return train_data_scaled, test_data_scaled
+            elif scaler_params == None:
+                train_data_scaled = (x_train - mean) / std
+                test_data_scaled = (x_test - mean) / std
+                return scaler_params, train_data_scaled, test_data_scaled
+                
+            elif scaler_params != None:
+                test_data_scaled = (x_test - scaler_params[0]) / scaler_params[1]
+                return test_data_scaled
 
     except:
         print(
@@ -1325,6 +1340,103 @@ def plot_evaluate(y_test, y_preds, acc_list):
     plt.tight_layout()
     plt.show()
    
+   
+def manuel_balancer(x_train, y_train, target_samples_per_class):
+    """
+    Generates synthetic examples to balance classes to the specified number of examples per class.
+
+    Arguments:
+    x_train -- Input dataset (examples) - NumPy array format
+    y_train -- Class labels (one-hot encoded) - NumPy array format
+    target_samples_per_class -- Desired number of samples per class
+
+    Returns:
+    x_balanced -- Balanced input dataset (NumPy array format)
+    y_balanced -- Balanced class labels (one-hot encoded, NumPy array format)
+    """
+    start_time = time.time()
+    x_train = np.array(x_train)
+    y_train = np.array(y_train)
+    classes = np.arange(y_train.shape[1])
+    class_count = len(classes)
+    
+    x_balanced = []
+    y_balanced = []
+
+    for class_label in range(class_count):
+        class_indices = np.where(np.argmax(y_train, axis=1) == class_label)[0]
+        num_samples = len(class_indices)
+        
+        if num_samples > target_samples_per_class:
+      
+            selected_indices = np.random.choice(class_indices, target_samples_per_class, replace=False)
+            x_balanced.append(x_train[selected_indices])
+            y_balanced.append(y_train[selected_indices])
+            
+        else:
+            
+            x_balanced.append(x_train[class_indices])
+            y_balanced.append(y_train[class_indices])
+
+            if num_samples < target_samples_per_class:
+                
+                samples_to_add = target_samples_per_class - num_samples
+                additional_samples = np.zeros((samples_to_add, x_train.shape[1]))
+                additional_labels = np.zeros((samples_to_add, y_train.shape[1]))
+                
+                for i in range(samples_to_add):
+                    uni_start_time = time.time()
+                    
+                    random_indices = np.random.choice(class_indices, 2, replace=False)
+                    sample1 = x_train[random_indices[0]]
+                    sample2 = x_train[random_indices[1]]
+
+                    
+                    synthetic_sample = sample1 + (sample2 - sample1) * np.random.rand()
+
+                    additional_samples[i] = synthetic_sample
+                    additional_labels[i] = y_train[class_indices[0]]
+                    
+                    uni_end_time = time.time()
+                    calculating_est = round(
+               (uni_end_time - uni_start_time) * (samples_to_add - i), 3)
+    
+                    if calculating_est < 60:
+                        print('\rest......(sec):', calculating_est, '\n', end="")
+             
+                    elif calculating_est > 60 and calculating_est < 3600:
+                        print('\rest......(min):', calculating_est/60, '\n', end="")
+             
+                    elif calculating_est > 3600:
+                        print('\rest......(h):', calculating_est/3600, '\n', end="")
+             
+                    print('Augmenting: ', class_label, '/', class_count, '...', i, "/", samples_to_add, "\n", end="")
+
+                x_balanced.append(additional_samples)
+                y_balanced.append(additional_labels)
+                   
+
+    EndTime = time.time()
+
+    calculating_est = round(EndTime - start_time, 2)
+ 
+    print(Fore.GREEN + " \nBalancing Finished with 0 ERROR\n" + Style.RESET_ALL)
+ 
+    if calculating_est < 60:
+        print('Total balancing time(sec): ', calculating_est)
+ 
+    elif calculating_est > 60 and calculating_est < 3600:
+        print('Total balancing time(min): ', calculating_est/60)
+ 
+    elif calculating_est > 3600:
+        print('Total balancing time(h): ', calculating_est/3600)
+    
+    # Stack the balanced arrays
+    x_balanced = np.vstack(x_balanced)
+    y_balanced = np.vstack(y_balanced)
+
+    return x_balanced, y_balanced  
+
 def get_weights():
         
     return 0

plan_di/__init__.py

@@ -2,4 +2,4 @@
 
 # Bu dosya, plan modülünün ana giriş noktasıdır.
 
-from .plan_di import auto_balancer, normalization, Softmax, Sigmoid, Relu, weight_identification, fex, fit, evaluate, save_model, load_model, predict_model_ssd, predict_model_ram, get_weights, get_df, get_preds, get_acc, synthetic_augmentation, standard_scaler, multiple_evaluate, encode_one_hot, split, metrics, decode_one_hot, roc_curve, confusion_matrix, plot_evaluate
+from .plan_di import auto_balancer, normalization, Softmax, Sigmoid, Relu, weight_identification, fex, fit, evaluate, save_model, load_model, predict_model_ssd, predict_model_ram, get_weights, get_df, get_preds, get_acc, synthetic_augmentation, standard_scaler, multiple_evaluate, encode_one_hot, split, metrics, decode_one_hot, roc_curve, confusion_matrix, plot_evaluate, manuel_balancer

plan_di/plan_di.py

@@ -51,7 +51,7 @@ def fit(
     y_train = [tuple(sublist) for sublist in y_train]
 
     neurons = [len(class_count), len(class_count)]
-    layers = ['fex', 'cat']
+    layers = ['fex']
 
     x_train[0] = np.array(x_train[0])
     x_train[0] = x_train[0].ravel()
@@ -120,6 +120,18 @@ def fit(
     elif calculating_est > 3600:
         print('Total training time(h): ', calculating_est/3600)
     
+    layers.append('cat')
+    trained_W.append(np.eye(len(class_count)))
+    
+    for i in range(len(class_count)):
+    
+        for j in range(len(layers)):
+        
+            if layers[j] == 'fex':
+            
+                trained_W[j][i,:] = normalization(trained_W[j][i,:])
+                
+
     return trained_W
 
 # FUNCTIONS -----
@@ -150,9 +162,7 @@ def weight_identification(
     ws = layer_count - 1
     for w in range(ws):
         W[w + 1] = np.ones((neurons[w + 1], neurons[w]))
-    W[layer_count] = np.ones((class_count, neurons[layer_count - 1]))
-    W[len(W) - 1] = np.eye(class_count)
-    
+   
     return W
 
 
@@ -507,7 +517,7 @@ def save_model(model_name,
                weights_type,
                weights_format,
                model_path,
-               scaler,
+               scaler_params,
                W
                ):
 
@@ -522,7 +532,7 @@ def save_model(model_name,
     weights_type (str): Type of weights to save (options: 'txt', 'npy', 'mat').
     WeightFormat (str): Format of the weights (options: 'd', 'f', 'raw').
     model_path (str): Path where the model will be saved. For example: C:/Users/beydili/Desktop/denemePLAN/
-    scaler (bool): trained data it used standard_scaler ? (True or False)
+    scaler_params (int, float): standard scaler params list: mean,std. If not used standard scaler then be: None.
     W: Weights of the model.
     
     Returns:
@@ -572,7 +582,7 @@ def save_model(model_name,
             'WEIGHTS TYPE': weights_type,
             'WEIGHTS FORMAT': weights_format,
             'MODEL PATH': model_path,
-            'STANDARD SCALER': scaler
+            'STANDARD SCALER': scaler_params
             }
     try:
 
@@ -719,11 +729,11 @@ def predict_model_ssd(Input, model_name, model_path):
     """
     W, df = load_model(model_name, model_path)
 
-    scaler = str(df['STANDARD SCALER'].iloc[0])
+    scaler_params = str(df['STANDARD SCALER'].iloc[0])
 
-    if scaler == 'True':
+    if scaler_params != None:
 
-        Input = standard_scaler(Input, None)
+        Input = standard_scaler(scaler_params, Input, None)
 
     layers = ['fex', 'cat']
 
@@ -751,7 +761,7 @@ def predict_model_ssd(Input, model_name, model_path):
     return neural_layer
 
 
-def predict_model_ram(Input, scaler, W):
+def predict_model_ram(Input, scaler_params, W):
 
     infopredict_model_ram = """
     Function to make a prediction using a divided pruning learning artificial neural network (PLAN).
@@ -759,16 +769,16 @@ def predict_model_ram(Input, scaler, W):
 
     Arguments:
     Input (list or ndarray): Input data for the model (single vector or single matrix).
-    scaler (bool): trained data it used standard_scaler ? (True or False)
+    scaler_params (int, float): standard scaler params list: mean,std. If not used standard scaler then be: None.
     W (list of ndarrays): Weights of the model.
 
     Returns:
     ndarray: Output from the model.
     """
 
-    if scaler == True:
+    if scaler_params != None:
 
-        Input = standard_scaler(Input, None)
+        Input = standard_scaler(scaler_params, Input, None)
 
     layers = ['fex', 'cat']
 
@@ -895,7 +905,7 @@ def synthetic_augmentation(x_train, y_train):
     return np.array(x_balanced), np.array(y_balanced)
 
 
-def standard_scaler(x_train, x_test):
+def standard_scaler(scaler_params=None, x_train, x_test):
     info_standard_scaler = """
   Standardizes training and test datasets. x_test may be None.
 
@@ -906,23 +916,31 @@ def standard_scaler(x_train, x_test):
       Test data
 
   Returns:
+    list:
+    Scaler parameters: mean and std
     tuple
       Standardized training and test datasets
   """
-    try:
-
-        mean = np.mean(x_train, axis=0)
-        std = np.std(x_train, axis=0)
+        try:
+            
+            if scaler_params == None:
+                mean = np.mean(x_train, axis=0)
+                std = np.std(x_train, axis=0)
+                scaler_params = [mean, std]
 
-        if x_test == None:
+            if x_test == None:
 
-            train_data_scaled = (x_train - mean) / std
-            return train_data_scaled
+                train_data_scaled = (x_train - mean) / std
+                return scaler_params, train_data_scaled
 
-        else:
-            train_data_scaled = (x_train - mean) / std
-            test_data_scaled = (x_test - mean) / std
-            return train_data_scaled, test_data_scaled
+            elif scaler_params == None:
+                train_data_scaled = (x_train - mean) / std
+                test_data_scaled = (x_test - mean) / std
+                return scaler_params, train_data_scaled, test_data_scaled
+                
+            elif scaler_params != None:
+                test_data_scaled = (x_test - scaler_params[0]) / scaler_params[1]
+                return test_data_scaled
 
     except:
         print(
@@ -1295,6 +1313,102 @@ def plot_evaluate(y_test, y_preds, acc_list):
     plt.tight_layout()
     plt.show()
 
+def manuel_balancer(x_train, y_train, target_samples_per_class):
+    """
+    Generates synthetic examples to balance classes to the specified number of examples per class.
+
+    Arguments:
+    x_train -- Input dataset (examples) - NumPy array format
+    y_train -- Class labels (one-hot encoded) - NumPy array format
+    target_samples_per_class -- Desired number of samples per class
+
+    Returns:
+    x_balanced -- Balanced input dataset (NumPy array format)
+    y_balanced -- Balanced class labels (one-hot encoded, NumPy array format)
+    """
+    start_time = time.time()
+    x_train = np.array(x_train)
+    y_train = np.array(y_train)
+    classes = np.arange(y_train.shape[1])
+    class_count = len(classes)
+    
+    x_balanced = []
+    y_balanced = []
+
+    for class_label in range(class_count):
+        class_indices = np.where(np.argmax(y_train, axis=1) == class_label)[0]
+        num_samples = len(class_indices)
+        
+        if num_samples > target_samples_per_class:
+      
+            selected_indices = np.random.choice(class_indices, target_samples_per_class, replace=False)
+            x_balanced.append(x_train[selected_indices])
+            y_balanced.append(y_train[selected_indices])
+            
+        else:
+            
+            x_balanced.append(x_train[class_indices])
+            y_balanced.append(y_train[class_indices])
+
+            if num_samples < target_samples_per_class:
+                
+                samples_to_add = target_samples_per_class - num_samples
+                additional_samples = np.zeros((samples_to_add, x_train.shape[1]))
+                additional_labels = np.zeros((samples_to_add, y_train.shape[1]))
+                
+                for i in range(samples_to_add):
+                    uni_start_time = time.time()
+                    
+                    random_indices = np.random.choice(class_indices, 2, replace=False)
+                    sample1 = x_train[random_indices[0]]
+                    sample2 = x_train[random_indices[1]]
+
+                    
+                    synthetic_sample = sample1 + (sample2 - sample1) * np.random.rand()
+
+                    additional_samples[i] = synthetic_sample
+                    additional_labels[i] = y_train[class_indices[0]]
+                    
+                    uni_end_time = time.time()
+                    calculating_est = round(
+               (uni_end_time - uni_start_time) * (samples_to_add - i), 3)
+    
+                    if calculating_est < 60:
+                        print('\rest......(sec):', calculating_est, '\n', end="")
+             
+                    elif calculating_est > 60 and calculating_est < 3600:
+                        print('\rest......(min):', calculating_est/60, '\n', end="")
+             
+                    elif calculating_est > 3600:
+                        print('\rest......(h):', calculating_est/3600, '\n', end="")
+             
+                    print('Augmenting: ', class_label, '/', class_count, '...', i, "/", samples_to_add, "\n", end="")
+
+                x_balanced.append(additional_samples)
+                y_balanced.append(additional_labels)
+                   
+
+    EndTime = time.time()
+
+    calculating_est = round(EndTime - start_time, 2)
+ 
+    print(Fore.GREEN + " \nBalancing Finished with 0 ERROR\n" + Style.RESET_ALL)
+ 
+    if calculating_est < 60:
+        print('Total balancing time(sec): ', calculating_est)
+ 
+    elif calculating_est > 60 and calculating_est < 3600:
+        print('Total balancing time(min): ', calculating_est/60)
+ 
+    elif calculating_est > 3600:
+        print('Total balancing time(h): ', calculating_est/3600)
+    
+    # Stack the balanced arrays
+    x_balanced = np.vstack(x_balanced)
+    y_balanced = np.vstack(y_balanced)
+
+    return x_balanced, y_balanced
+
 def get_weights():
 
     return 0

{pyerualjetwork-2.2.7.dist-info → pyerualjetwork-2.2.9.dist-info}/METADATA

@@ -1,7 +1,7 @@
 Metadata-Version: 2.1
 Name: pyerualjetwork
-Version: 2.2.7
-Summary: Code improvements (Documentation in desc. Examples in GİTHUB: https://github.com/HCB06/PyerualJetwork)
+Version: 2.2.9
+Summary: Weights now noramlized, new function: manuel_balancer. And scaler_params added for scaled models.
 Author: Hasan Can Beydili
 Author-email: tchasancan@gmail.com
 Keywords: model evaluation,classifcation,pruning learning artficial neural networks

pyerualjetwork-2.2.9.dist-info/RECORD

@@ -0,0 +1,8 @@
+plan_bi/__init__.py,sha256=m84PX8yNU6pVlQNyWOqamkQdx_-_aPFsfvtwEnC8U_w,492
+plan_bi/plan_bi.py,sha256=ehSRrQ7_5YRKEVrHNazP8CPZSlJWVxB-lDZUTPO_Q7g,50053
+plan_di/__init__.py,sha256=cp43HHfPlIYVGFULVoiEwjFE8paMHCAgYLSic355gqs,483
+plan_di/plan_di.py,sha256=5IBktVVqKENjDm2gi8ykNOaWfxspoJuG_36FrB0m07k,47444
+pyerualjetwork-2.2.9.dist-info/METADATA,sha256=w1H4NKan5q2vMU7TjOyDVGnMqWHKf3kDFYpQiPMZPf8,319
+pyerualjetwork-2.2.9.dist-info/WHEEL,sha256=2wepM1nk4DS4eFpYrW1TTqPcoGNfHhhO_i5m4cOimbo,92
+pyerualjetwork-2.2.9.dist-info/top_level.txt,sha256=aaXSOcnD62fbXG1x7tw4nV50Qxx9g9zDNLK7OD4BdPE,16
+pyerualjetwork-2.2.9.dist-info/RECORD,,

pyerualjetwork-2.2.7.dist-info/RECORD

@@ -1,8 +0,0 @@
-plan_bi/__init__.py,sha256=rzDe7yWvNlwDVE6xSw8Qk51itcxT6EDBj7iiOeycxMw,475
-plan_bi/plan_bi.py,sha256=ZK2rwYVgYxHygwmgUvb8G1cQ8AHPRm6e2yVHgb9X5Eo,45416
-plan_di/__init__.py,sha256=Omxc07PXPQZOrXBD3PJQT6sPdni6NMykyiQgKVL_IZ0,466
-plan_di/plan_di.py,sha256=nWUc3D9yTDQ6V1TQAGH4Q9F4yKJIndKYGGQPqtysNOU,42786
-pyerualjetwork-2.2.7.dist-info/METADATA,sha256=_XpAwKRGokLPyY5Z5wW3pXLUF9nKFgWDaw510HVKa6E,325
-pyerualjetwork-2.2.7.dist-info/WHEEL,sha256=2wepM1nk4DS4eFpYrW1TTqPcoGNfHhhO_i5m4cOimbo,92
-pyerualjetwork-2.2.7.dist-info/top_level.txt,sha256=aaXSOcnD62fbXG1x7tw4nV50Qxx9g9zDNLK7OD4BdPE,16
-pyerualjetwork-2.2.7.dist-info/RECORD,,