pyerualjetwork 4.3.2.2b0__tar.gz → 4.3.4__tar.gz

{pyerualjetwork-4.3.2.2b0 → pyerualjetwork-4.3.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: pyerualjetwork
-Version: 4.3.2.2b0
+Version: 4.3.4
 Summary: PyerualJetwork is a machine learning library supported with GPU(CUDA) acceleration written in Python for professionals and researchers including with PLAN algorithm, PLANEAT algorithm (genetic optimization). Also includes data pre-process and memory manegament
 Author: Hasan Can Beydili
 Author-email: tchasancan@gmail.com
@@ -23,18 +23,33 @@ PyPi Page: https://pypi.org/project/pyerualjetwork/
 GitHub Page: https://github.com/HCB06/PyerualJetwork
 
 
-      pip install pyerualjetwork==x.x.x (means it is a normal package)
-      pip install pyerualjetwork==x.x.x.x (last x means it is a afterburner package)
-      
-      from pyerualjetwork import plan
-      from pyerualjetwork import planeat
-      from pyerualjetwork import data_operations
-      from pyerualjetwork import model_operations
-
-      from pyerualjetwork import plan_cuda
-      from pyerualjetwork import planeat_cuda
-      from pyerualjetwork import data_operations_cuda
-      from pyerualjetwork import model_operations_cuda
+pip install pyerualjetwork
+
+'use this if your data small or memory management is a problem :'
+
+from pyerualjetwork import plan
+from pyerualjetwork import planeat
+from pyerualjetwork import data_operations
+from pyerualjetwork import model_operations
+
+from pyerualjetwork import plan_cuda
+from pyerualjetwork import planeat_cuda
+from pyerualjetwork import data_operations_cuda
+from pyerualjetwork import model_operations_cuda
+
+'use this if your data large or memory management is not a problem : _afterburner package (afterburner package comes with powerful paralellism,
+afterburner with cuda modules offers super-fast training but some memory managemant features and visualization features discarded.
+Specially designed for LLM training and other massive model training)'
+
+from pyerualjetwork_afterburner import plan
+from pyerualjetwork_afterburner import planeat
+from pyerualjetwork_afterburner import data_operations
+from pyerualjetwork_afterburner import model_operations
+
+from pyerualjetwork_afterburner import plan_cuda
+from pyerualjetwork_afterburner import planeat_cuda
+from pyerualjetwork_afterburner import data_operations_cuda
+from pyerualjetwork_afterburner import model_operations_cuda
 
       Optimized for Visual Studio Code
       

{pyerualjetwork-4.3.2.2b0 → pyerualjetwork-4.3.4}/README.md

@@ -14,18 +14,33 @@ PyPi Page: https://pypi.org/project/pyerualjetwork/
 GitHub Page: https://github.com/HCB06/PyerualJetwork
 
 
-      pip install pyerualjetwork==x.x.x (means it is a normal package)
-      pip install pyerualjetwork==x.x.x.x (last x means it is a afterburner package)
-      
-      from pyerualjetwork import plan
-      from pyerualjetwork import planeat
-      from pyerualjetwork import data_operations
-      from pyerualjetwork import model_operations
-
-      from pyerualjetwork import plan_cuda
-      from pyerualjetwork import planeat_cuda
-      from pyerualjetwork import data_operations_cuda
-      from pyerualjetwork import model_operations_cuda
+pip install pyerualjetwork
+
+'use this if your data small or memory management is a problem :'
+
+from pyerualjetwork import plan
+from pyerualjetwork import planeat
+from pyerualjetwork import data_operations
+from pyerualjetwork import model_operations
+
+from pyerualjetwork import plan_cuda
+from pyerualjetwork import planeat_cuda
+from pyerualjetwork import data_operations_cuda
+from pyerualjetwork import model_operations_cuda
+
+'use this if your data large or memory management is not a problem : _afterburner package (afterburner package comes with powerful paralellism,
+afterburner with cuda modules offers super-fast training but some memory managemant features and visualization features discarded.
+Specially designed for LLM training and other massive model training)'
+
+from pyerualjetwork_afterburner import plan
+from pyerualjetwork_afterburner import planeat
+from pyerualjetwork_afterburner import data_operations
+from pyerualjetwork_afterburner import model_operations
+
+from pyerualjetwork_afterburner import plan_cuda
+from pyerualjetwork_afterburner import planeat_cuda
+from pyerualjetwork_afterburner import data_operations_cuda
+from pyerualjetwork_afterburner import model_operations_cuda
 
       Optimized for Visual Studio Code
       

{pyerualjetwork-4.3.2.2b0/pyerualjetwork_afterburner → pyerualjetwork-4.3.4/pyerualjetwork}/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "4.3.2.2b0-afterburner"
+__version__ = "4.3.4"
 __update__ = "* Changes: https://github.com/HCB06/PyerualJetwork/blob/main/CHANGES\n* PyerualJetwork Homepage: https://github.com/HCB06/PyerualJetwork/tree/main\n* PyerualJetwork document: https://github.com/HCB06/PyerualJetwork/blob/main/Welcome_to_PyerualJetwork/PYERUALJETWORK_USER_MANUEL_AND_LEGAL_INFORMATION(EN).pdf\n* YouTube tutorials: https://www.youtube.com/@HasanCanBeydili"
 
 def print_version(__version__):

{pyerualjetwork-4.3.2.2b0/pyerualjetwork_afterburner → pyerualjetwork-4.3.4/pyerualjetwork}/activation_functions.py

@@ -217,8 +217,8 @@ def scaled_cubic(x, alpha=1.0):
 
 def sine_offset(x, beta=0.0):
     return np.sin(x + beta)
-
-
+    
+    
 def apply_activation(Input, activation_list):
     """
     Applies activation functions for inputs
@@ -277,15 +277,14 @@ def apply_activation(Input, activation_list):
     }
     
     try:
-        valid_activations = [act for act in activation_list if act in activation_functions]
-        
-        activation_outputs = np.stack([activation_functions[act](origin_input) 
-                                     for act in valid_activations])
 
-        result = Input + np.sum(activation_outputs, axis=0)
-        
-        return result
+        valid_mask = np.array([act in activation_functions for act in activation_list])
+        valid_activations = np.array(activation_list)[valid_mask]
+
+        activation_outputs = np.array([activation_functions[act](origin_input) for act in valid_activations])
+
+        return Input + np.sum(activation_outputs, axis=0)
         
     except Exception as e:
         warnings.warn(f"Error in activation processing: {str(e)}", RuntimeWarning)
-        return Input
+        return Input

pyerualjetwork-4.3.4/pyerualjetwork/activation_functions_cuda.py

@@ -0,0 +1,340 @@
+import cupy as cp
+from scipy.special import expit, softmax
+import warnings
+
+# ACTIVATION FUNCTIONS ----
+
+def all_activations():
+    
+    activations_list = ['linear', 'sigmoid', 'relu', 'tanh', 'circular', 'spiral', 'swish', 'sin_plus', 'mod_circular', 'tanh_circular', 'leaky_relu', 'softplus', 'elu', 'gelu', 'selu', 'sinakt', 'p_squared', 'sglu', 'dlrelu', 'exsig', 'acos', 'gla', 'srelu', 'qelu', 'isra', 'waveakt', 'arctan', 'bent_identity', 'sech', 'softsign', 'pwl', 'cubic', 'gaussian', 'sine', 'tanh_square', 'mod_sigmoid', 'quartic', 'square_quartic', 'cubic_quadratic', 'exp_cubic', 'sine_square', 'logarithmic', 'scaled_cubic', 'sine_offset']
+    
+    return activations_list
+
+def spiral_activation(x):
+
+    r = cp.sqrt(cp.sum(x**2))
+    
+    theta = cp.arctan2(x[1:], x[:-1])
+
+    spiral_x = r * cp.cos(theta + r)
+    spiral_y = r * cp.sin(theta + r)
+
+
+    spiral_output = cp.concatenate([cp.array([spiral_x[0]]), spiral_y])
+    
+    return spiral_output
+
+
+def Softmax(
+    x  # num: Input data to be transformed using softmax function.
+):
+    """
+    Applies the softmax function to the input data.
+
+    Args:
+        (num): Input data to be transformed using softmax function.
+
+    Returns:
+       (num): Transformed data after applying softmax function.
+    """
+    
+    return cp.array(softmax(x.get()))
+
+
+def Sigmoid(
+    x  # num: Input data to be transformed using sigmoid function.
+):
+    """
+    Applies the sigmoid function to the input data.
+
+    Args:
+        (num): Input data to be transformed using sigmoid function.
+
+    Returns:
+        (num): Transformed data after applying sigmoid function.
+    """
+    return expit(x)
+
+
+def Relu(
+    x  # num: Input data to be transformed using ReLU function.
+):
+    """
+    Applies the Rectified Linear Unit (ReLU) function to the input data.
+
+    Args:
+        (num): Input data to be transformed using ReLU function.
+
+    Returns:
+        (num): Transformed data after applying ReLU function.
+    """
+
+    return cp.maximum(0, x)
+
+
+def tanh(x):
+    return cp.tanh(x)
+
+def swish(x):
+    return x * (1 / (1 + cp.exp(-x)))
+
+def sin_plus(x):
+    return (cp.sin(x) + 1) / 2
+
+def modular_circular_activation(x, period=2*cp.pi):
+    return cp.mod(x, period) / period
+
+def tanh_circular_activation(x):
+    return (cp.tanh(x) + 1) / 2
+
+def leaky_relu(x, alpha=0.01):
+    return cp.where(x > 0, x, alpha * x)
+
+def softplus(x):
+    return cp.log(1 + cp.exp(x))
+
+def elu(x, alpha=1.0):
+    return cp.where(x > 0, x, alpha * (cp.exp(x) - 1))
+
+def gelu(x):
+    return 0.5 * x * (1 + cp.tanh(cp.sqrt(2 / cp.pi) * (x + 0.044715 * cp.power(x, 3))))
+
+def selu(x, lambda_=1.0507, alpha=1.6733):
+    return lambda_ * cp.where(x > 0, x, alpha * (cp.exp(x) - 1))
+
+def sinakt(x):
+    return cp.sin(x) + cp.cos(x)
+
+def p_squared(x, alpha=1.0, beta=0.0):
+    return alpha * x**2 + beta * x
+
+def sglu(x, alpha=1.0):
+    return cp.array(softmax(alpha * x.get())) * x
+
+# 4. Double Leaky ReLU (DLReLU)
+def dlrelu(x):
+    return cp.maximum(0.01 * x, x) + cp.minimum(0.01 * x, 0.1 * x)
+
+# 5. Exponential Sigmoid (ExSig)
+def exsig(x):
+    return 1 / (1 + cp.exp(-x**2))
+
+# 6. Adaptive Cosine Activation (ACos)
+def acos(x, alpha=1.0, beta=0.0):
+    return cp.cos(alpha * x + beta)
+
+# 7. Gaussian-like Activation (GLA)
+def gla(x, alpha=1.0, mu=0.0):
+    return cp.exp(-alpha * (x - mu)**2)
+
+# 8. Swish ReLU (SReLU)
+def srelu(x):
+    return x * (1 / (1 + cp.exp(-x))) + cp.maximum(0, x)
+
+# 9. Quadratic Exponential Linear Unit (QELU)
+def qelu(x):
+    return x**2 * cp.exp(x) - 1
+
+# 10. Inverse Square Root Activation (ISRA)
+def isra(x):
+    return x / cp.sqrt(cp.abs(x) + 1)
+
+def waveakt(x, alpha=1.0, beta=2.0, gamma=3.0):
+    return cp.sin(alpha * x) * cp.cos(beta * x) * cp.sin(gamma * x)
+
+def arctan(x):
+    return cp.arctan(x)
+
+def bent_identity(x):
+    return (cp.sqrt(x**2 + 1) - 1) / 2 + x
+
+def circular_activation(x, scale=2.0, frequency=1.0, shift=0.0):    
+    
+    n_features = x.shape[0]
+    
+    circular_output = cp.zeros_like(x)
+    
+    for i in range(n_features):
+        
+        r = cp.sqrt(cp.sum(x**2))
+        theta = 2 * cp.pi * (i / n_features) + shift
+        
+        circular_x = r * cp.cos(theta + frequency * r) * scale
+        circular_y = r * cp.sin(theta + frequency * r) * scale
+        
+        if i % 2 == 0:
+            circular_output[i] = circular_x
+        else:
+            circular_output[i] = circular_y
+    
+    return circular_output
+
+def sech(x):
+    return 2 / (cp.exp(x) + cp.exp(-x))
+
+def softsign(x):
+    return x / (1 + cp.abs(x))
+
+def pwl(x, alpha=0.5, beta=1.5):
+    return cp.where(x <= 0, alpha * x, beta * x)
+
+def cubic(x):
+    return x**3
+
+def gaussian(x, alpha=1.0, mu=0.0):
+    return cp.exp(-alpha * (x - mu)**2)
+                
+def sine(x, alpha=1.0):
+    return cp.sin(alpha * x)
+
+def tanh_square(x):
+    return cp.tanh(x)**2
+
+def mod_sigmoid(x, alpha=1.0, beta=0.0):
+    return 1 / (1 + cp.exp(-alpha * x + beta))
+
+def quartic(x):
+    return x**4
+
+def square_quartic(x):
+    return (x**2)**2
+
+def cubic_quadratic(x):
+    return x**3 * (x**2)
+
+def exp_cubic(x):
+    return cp.exp(x**3)
+
+def sine_square(x):
+    return cp.sin(x)**2
+
+def logarithmic(x):
+    return cp.log(x**2 + 1)
+
+def scaled_cubic(x, alpha=1.0):
+    return alpha * x**3
+
+def sine_offset(x, beta=0.0):
+    return cp.sin(x + beta)
+
+
+def safe_add(current_sum, new_value):
+        try:
+            return current_sum + new_value
+        except OverflowError:
+            return cp.array(current_sum) + cp.array(new_value)
+    
+def apply_activation(Input, activation_list):
+    """
+        Applies a sequence of activation functions to the input.
+        
+        Args:
+            Input (cupy.ndarray): The input to apply activations to.
+            activation_list (list): A list of activation function names to apply.
+        
+        Returns:
+            cupy.ndarray: The input after all activations have been applied.
+    """
+    
+    origin_input = cp.copy(Input)
+
+    for i in range(len(activation_list)):
+        try:
+            if activation_list[i] == 'sigmoid':
+                Input = safe_add(Input, Sigmoid(origin_input))
+            elif activation_list[i] == 'swish':
+                Input = safe_add(Input, swish(origin_input))
+            elif activation_list[i] == 'mod_circular':
+                Input = safe_add(Input, modular_circular_activation(origin_input))
+            elif activation_list[i] == 'tanh_circular':
+                Input = safe_add(Input, tanh_circular_activation(origin_input))
+            elif activation_list[i] == 'leaky_relu':
+                Input = safe_add(Input, leaky_relu(origin_input))
+            elif activation_list[i] == 'relu':
+                Input = safe_add(Input, Relu(origin_input))
+            elif activation_list[i] == 'softplus':
+                Input = safe_add(Input, softplus(origin_input))
+            elif activation_list[i] == 'elu':
+                Input = safe_add(Input, elu(origin_input))
+            elif activation_list[i] == 'gelu':
+                Input = safe_add(Input, gelu(origin_input))
+            elif activation_list[i] == 'selu':
+                Input = safe_add(Input, selu(origin_input))
+            elif activation_list[i] == 'tanh':
+                Input = safe_add(Input, tanh(origin_input))
+            elif activation_list[i] == 'sinakt':
+                Input = safe_add(Input, sinakt(origin_input))
+            elif activation_list[i] == 'p_squared':
+                Input = safe_add(Input, p_squared(origin_input))
+            elif activation_list[i] == 'sglu':
+                Input = safe_add(Input, sglu(origin_input, alpha=1.0))
+            elif activation_list[i] == 'dlrelu':
+                Input = safe_add(Input, dlrelu(origin_input))
+            elif activation_list[i] == 'exsig':
+                Input = safe_add(Input, exsig(origin_input))
+            elif activation_list[i] == 'sin_plus':
+                Input = safe_add(Input, sin_plus(origin_input))
+            elif activation_list[i] == 'acos':
+                Input = safe_add(Input, acos(origin_input, alpha=1.0, beta=0.0))
+            elif activation_list[i] == 'gla':
+                Input = safe_add(Input, gla(origin_input, alpha=1.0, mu=0.0))
+            elif activation_list[i] == 'srelu':
+                Input = safe_add(Input, srelu(origin_input))
+            elif activation_list[i] == 'qelu':
+                Input = safe_add(Input, qelu(origin_input))
+            elif activation_list[i] == 'isra':
+                Input = safe_add(Input, isra(origin_input))
+            elif activation_list[i] == 'waveakt':
+                Input = safe_add(Input, waveakt(origin_input))
+            elif activation_list[i] == 'arctan':
+                Input = safe_add(Input, arctan(origin_input))
+            elif activation_list[i] == 'bent_identity':
+                Input = safe_add(Input, bent_identity(origin_input))
+            elif activation_list[i] == 'sech':
+                Input = safe_add(Input, sech(origin_input))
+            elif activation_list[i] == 'softsign':
+                Input = safe_add(Input, softsign(origin_input))
+            elif activation_list[i] == 'pwl':
+                Input = safe_add(Input, pwl(origin_input))
+            elif activation_list[i] == 'cubic':
+                Input = safe_add(Input, cubic(origin_input))
+            elif activation_list[i] == 'gaussian':
+                Input = safe_add(Input, gaussian(origin_input))
+            elif activation_list[i] == 'sine':
+                Input = safe_add(Input, sine(origin_input))
+            elif activation_list[i] == 'tanh_square':
+                Input = safe_add(Input, tanh_square(origin_input))
+            elif activation_list[i] == 'mod_sigmoid':
+                Input = safe_add(Input, mod_sigmoid(origin_input))
+            elif activation_list[i] == 'linear':
+                Input = safe_add(Input, origin_input)
+            elif activation_list[i] == 'quartic':
+                Input = safe_add(Input, quartic(origin_input))
+            elif activation_list[i] == 'square_quartic':
+                Input = safe_add(Input, square_quartic(origin_input))
+            elif activation_list[i] == 'cubic_quadratic':
+                Input = safe_add(Input, cubic_quadratic(origin_input))
+            elif activation_list[i] == 'exp_cubic':
+                Input = safe_add(Input, exp_cubic(origin_input))
+            elif activation_list[i] == 'sine_square':
+                Input = safe_add(Input, sine_square(origin_input))
+            elif activation_list[i] == 'logarithmic':
+                Input = safe_add(Input, logarithmic(origin_input))
+            elif activation_list[i] == 'scaled_cubic':
+                Input = safe_add(Input, scaled_cubic(origin_input, 1.0))
+            elif activation_list[i] == 'sine_offset':
+                Input = safe_add(Input, sine_offset(origin_input, 1.0))
+            elif activation_list[i] == 'spiral':
+                Input = safe_add(Input, spiral_activation(origin_input))
+            elif activation_list[i] == 'circular':
+                Input = safe_add(Input, circular_activation(origin_input))
+        
+        except Exception as e:
+            warnings.warn(f"Error in activation {activation_list[i]}: {str(e)}", RuntimeWarning)
+            if not isinstance(Input, cp.ndarray):
+                Input = cp.array(Input)
+            if not isinstance(origin_input, cp.ndarray):
+                origin_input = cp.array(origin_input)
+            continue
+ 
+    return Input