pyerualjetwork 4.7.2__py3-none-any.whl → 4.8__py3-none-any.whl

pyerualjetwork/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "4.7.2"
+__version__ = "4.8"
 __update__ = """* Changes: https://github.com/HCB06/PyerualJetwork/blob/main/CHANGES
 * PyerualJetwork Homepage: https://github.com/HCB06/PyerualJetwork/tree/main
 * PyerualJetwork document: https://github.com/HCB06/PyerualJetwork/blob/main/Welcome_to_PyerualJetwork/PYERUALJETWORK_USER_MANUEL_AND_LEGAL_INFORMATION(EN).pdf

pyerualjetwork/plan.py

@@ -16,6 +16,7 @@ PyerualJetwork document: https://github.com/HCB06/PyerualJetwork/blob/main/Welco
 """
 
 import numpy as np
+import copy
 
 ### LIBRARY IMPORTS ###
 from .ui import loading_bars, initialize_loading_bar
@@ -68,7 +69,7 @@ def fit(
     """
 
     # Pre-check
-    
+
     if len(x_train) != len(y_train): raise ValueError("x_train and y_train must have the same length.")
 
     weight = np.zeros((len(y_train[0]), len(x_train[0].ravel()))).astype(dtype, copy=False) if W is None else W
@@ -86,12 +87,12 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
            weight_evolve=True, neural_web_history=False, show_current_activations=False, auto_normalization=False,
            neurons_history=False, early_stop=False, show_history=False, target_loss=None,
            interval=33.33, target_acc=None, loss='categorical_crossentropy', acc_impact=0.9, loss_impact=0.1,
-           start_this_act=None, start_this_W=None, dtype=np.float32):
+           start_this_act=None, start_this_W=None, neurons=None, hidden=0, activation_functions=None, dtype=np.float32):
     """
     Optimizes the activation functions for a neural network by leveraging train data to find 
-    the most accurate combination of activation potentiation for the given dataset using genetic algorithm NEAT (Neuroevolution of Augmenting Topologies). But modifided for PLAN version. Created by me: PLANEAT. 
+    the most accurate combination of activation potentiation(or activation function) & weight values for the given dataset.
 
-    Why genetic optimization and not backpropagation?
+    Why genetic optimization ENE(Eugenic NeuroEvolution) and not backpropagation?
     Because PLAN is different from other neural network architectures. In PLAN, the learnable parameters are not the weights; instead, the learnable parameters are the activation functions.
     Since activation functions are not differentiable, we cannot use gradient descent or backpropagation. However, I developed a more powerful genetic optimization algorithm: PLANEAT.
 
@@ -120,10 +121,10 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
     :param batch_size: (float, optional): Batch size is used in the prediction process to receive train feedback by dividing the train data into chunks and selecting activations based on randomly chosen partitions. This process reduces computational cost and time while still covering the entire train set due to random selection, so it doesn't significantly impact accuracy. For example, a batch size of 0.08 means each train batch represents %8 of the train set. Default is 1. (%100 of train)
     :param pop_size: (int, optional): Population size of each generation. Default: count of activation functions
     :param weight_evolve: (bool, optional): Activation combinations already optimizes by PLANEAT genetic search algorithm. Should the weight parameters also evolve or should the weights be determined according to the aggregating learning principle of the PLAN algorithm? Default: True (Evolves Weights)
-    :param neural_web_history: (bool, optional): Draws history of neural web. Default is False.
+    :param neural_web_history: (bool, optional): Draws history of neural web. Default is False. [ONLY FOR PLAN MODELS]
     :param show_current_activations: (bool, optional): Should it display the activations selected according to the current strategies during learning, or not? (True or False) This can be very useful if you want to cancel the learning process and resume from where you left off later. After canceling, you will need to view the live training activations in order to choose the activations to be given to the 'start_this' parameter. Default is False
     :param auto_normalization: (bool, optional): Normalization may solves overflow problem. Default: False
-    :param neurons_history: (bool, optional): Shows the history of changes that neurons undergo during the TFL (Test or Train Feedback Learning) stages. True or False. Default is False.
+    :param neurons_history: (bool, optional): Shows the history of changes that neurons undergo during the ENE optimization process. True or False. Default is False. [ONLY FOR PLAN MODELS]
     :param early_stop: (bool, optional): If True, implements early stopping during training.(If accuracy not improves in two gen stops learning.) Default is False.
     :param show_history: (bool, optional): If True, displays the training history after optimization. Default is False.
     :param target_loss: (float, optional): The target loss to stop training early when achieved. Default is None.
@@ -134,7 +135,10 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
     :param loss_impact: (float, optional): Impact of loss for optimization [0-1]. Default: 0.1
     :param start_this_act: (list, optional): To resume a previously canceled or interrupted training from where it left off, or to continue from that point with a different strategy, provide the list of activation functions selected up to the learned portion to this parameter. Default is None
     :param start_this_W: (numpy.array, optional): To resume a previously canceled or interrupted training from where it left off, or to continue from that point with a different strategy, provide the weight matrix of this genome. Default is None
-    :param dtype: (numpy.dtype): Data type for the arrays. np.float32 by default. Example: np.float64 or np.float16.
+    :param hidden: (int, optional): If you dont want train PLAN model this parameter represents a hidden layer count for MLP model. Default: 0 (PLAN)
+    :param neurons: (list[int], optional): If you dont want train PLAN model this parameter represents neuron count of each hidden layer for MLP. Default: None (PLAN)
+    :param activation_functions: (list[str], optional): If you dont want train PLAN model this parameter represents activation function of each hidden layer for MLP. Default: None (PLAN) NOTE: THIS EFFECTS HIDDEN LAYERS OUTPUT. NOT OUTPUT LAYER!
+    :param dtype: (numpy.dtype): Data type for the Weight matrices. np.float32 by default. Example: np.float64 or np.float16.
 
     Returns:
         tuple: A list for model parameters: [Weight matrix, Test loss, Test Accuracy, [Activations functions]].
@@ -151,8 +155,6 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
     # Pre-checks
 
     if pop_size is None: pop_size = activation_potentiation_len
-
-    x_train = x_train.astype(dtype, copy=False)
     y_train = optimize_labels(y_train, cuda=False)
 
     if pop_size < activation_potentiation_len: raise ValueError(f"pop_size must be higher or equal to {activation_potentiation_len}")
@@ -162,6 +164,14 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
     if target_acc is not None and (target_acc < 0 or target_acc > 1): raise ValueError('target_acc must be in range 0 and 1')
     if fit_start is not True and fit_start is not False: raise ValueError('fit_start parameter only be True or False. Please read doc-string')
 
+    if hidden is not 0 and neurons is not None and activation_functions is not None:
+        fit_start = False
+        weight_evolve = True
+        is_mlp = True
+        activation_potentiation = activation_functions
+    else:
+        is_mlp = False
+
     # Initialize visualization components
     viz_objects = initialize_visualization_for_learner(show_history, neurons_history, neural_web_history, x_train, y_train)
 
@@ -177,7 +187,7 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
     progress = initialize_loading_bar(total=activation_potentiation_len, desc="", ncols=77, bar_format=bar_format_learner)
 
     if fit_start is False or pop_size > activation_potentiation_len:
-        weight_pop, act_pop = define_genomes(input_shape=len(x_train[0]), output_shape=len(y_train[0]), population_size=pop_size, dtype=dtype)
+        weight_pop, act_pop = define_genomes(input_shape=len(x_train[0]), output_shape=len(y_train[0]), neurons=neurons, hidden=hidden, activation_functions=activation_functions, population_size=pop_size, dtype=dtype)
 
     else:
         weight_pop = [0] * pop_size
@@ -210,8 +220,8 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
             
             if weight_evolve is False:
                 weight_pop[j] = fit(x_train_batch, y_train_batch, activation_potentiation=act_pop[j], auto_normalization=auto_normalization, dtype=dtype)
-                
-            model = evaluate(x_train_batch, y_train_batch, W=weight_pop[j], activation_potentiation=act_pop[j], auto_normalization=auto_normalization)
+
+            model = evaluate(x_train_batch, y_train_batch, W=weight_pop[j], activation_potentiation=act_pop[j], auto_normalization=auto_normalization, is_mlp=is_mlp)
             acc = model[get_acc()]
             
             if loss == 'categorical_crossentropy':
@@ -229,11 +239,11 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
                 best_fitness = fitness
                 best_acc = acc
                 best_loss = train_loss
-                best_weight = np.copy(weight_pop[j])
+                best_weight = np.copy(weight_pop[j]) if is_mlp is False else copy.deepcopy(weight_pop[j])
                 best_model = model
 
                 final_activations = act_pop[j].copy() if isinstance(act_pop[j], list) else act_pop[j]
-                final_activations = [final_activations[0]] if len(set(final_activations)) == 1 else final_activations # removing if all same
+                if is_mlp is False: final_activations = [final_activations[0]] if len(set(final_activations)) == 1 else final_activations # removing if all same
 
                 if batch_size == 1:
                     postfix_dict[f"{data} Accuracy"] = np.round(best_acc, 4)
@@ -268,7 +278,7 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
                 if target_acc is not None and best_acc >= target_acc:
                     progress.close()
                     train_model = evaluate(x_train, y_train, W=best_weight, 
-                                        activation_potentiation=final_activations, auto_normalization=auto_normalization)
+                                        activation_potentiation=final_activations, auto_normalization=auto_normalization, is_mlp=is_mlp)
                     if loss == 'categorical_crossentropy':
                         train_loss = categorical_crossentropy(y_true_batch=y_train, 
                                                            y_pred_batch=train_model[get_preds_softmax()])
@@ -288,7 +298,7 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
                 if target_loss is not None and best_loss <= target_loss:
                     progress.close()
                     train_model = evaluate(x_train, y_train, W=best_weight,
-                                        activation_potentiation=final_activations, auto_normalization=auto_normalization)
+                                        activation_potentiation=final_activations, auto_normalization=auto_normalization, is_mlp=is_mlp)
 
                     if loss == 'categorical_crossentropy':
                         train_loss = categorical_crossentropy(y_true_batch=y_train, 
@@ -310,7 +320,7 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
             progress.update(1)
 
         if batch_size != 1:
-            train_model = evaluate(x_train, y_train, best_weight, final_activations)
+            train_model = evaluate(x_train, y_train, best_weight, final_activations, auto_normalization=auto_normalization, is_mlp=is_mlp)
     
             if loss == 'categorical_crossentropy':
                 train_loss = categorical_crossentropy(y_true_batch=y_train, 
@@ -330,7 +340,10 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
             best_acc_per_gen_list.append(best_acc)
             loss_list.append(best_loss)
 
-        weight_pop, act_pop = optimizer(np.array(weight_pop, copy=False, dtype=dtype), act_pop, i, np.array(target_pop, dtype=dtype, copy=False), weight_evolve=weight_evolve, bar_status=False)
+        if is_mlp is False: weight_pop = np.array(weight_pop, copy=False, dtype=dtype)
+        else: weight_pop = np.array(weight_pop, copy=False, dtype=object)
+
+        weight_pop, act_pop = optimizer(weight_pop, act_pop, i, np.array(target_pop, dtype=dtype, copy=False), weight_evolve=weight_evolve, is_mlp=is_mlp, bar_status=False)
         target_pop = []
 
         # Early stopping check
@@ -338,7 +351,7 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
             if best_acc_per_gen_list[i] == best_acc_per_gen_list[i-1]:
                 progress.close()
                 train_model = evaluate(x_train, y_train, W=best_weight, 
-                                    activation_potentiation=final_activations, auto_normalization=auto_normalization)
+                                    activation_potentiation=final_activations, auto_normalization=auto_normalization, is_mlp=is_mlp)
                 
             if loss == 'categorical_crossentropy':
                 train_loss = categorical_crossentropy(y_true_batch=y_train, 
@@ -358,8 +371,9 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
 
     # Final evaluation
     progress.close()
+
     train_model = evaluate(x_train, y_train, W=best_weight,
-                        activation_potentiation=final_activations, auto_normalization=auto_normalization)
+                        activation_potentiation=final_activations, auto_normalization=auto_normalization, is_mlp=is_mlp)
 
     if loss == 'categorical_crossentropy':
         train_loss = categorical_crossentropy(y_true_batch=y_train, 
@@ -382,7 +396,8 @@ def evaluate(
     y_test,
     W,
     activation_potentiation=['linear'],
-    auto_normalization=False
+    auto_normalization=False,
+    is_mlp=False
 ) -> tuple:
     """
     Evaluates the neural network model using the given test data.
@@ -398,6 +413,8 @@ def evaluate(
 
         auto_normalization (bool, optional): Normalization for x_test ? Default = False.
         
+        is_mlp (bool, optional): Evaluate PLAN model or MLP model ? Default: False (PLAN)
+
     Returns:
         tuple: Model (list).
     """
@@ -406,12 +423,22 @@ def evaluate(
 
     if isinstance(activation_potentiation, str):
         activation_potentiation = [activation_potentiation]
+    elif isinstance(activation_potentiation, list):
+        activation_potentiation = [item if isinstance(item, list) or isinstance(item, str) else [item] for item in activation_potentiation]
+
+    if is_mlp:
+        layer = x_test
+        for i in range(len(W)):
+            if i != len(W) - 1 and i != 0: layer = apply_activation(layer, activation_potentiation[i])
+
+            layer = layer @ W[i].T
+
+        result = layer
+
     else:
-        activation_potentiation = [item if isinstance(item, list) else [item] for item in activation_potentiation]
 
-    x_test = apply_activation(x_test, activation_potentiation)
-    
-    result = x_test @ W.T
+        x_test = apply_activation(x_test, activation_potentiation)
+        result = x_test @ W.T
     
     max_vals = np.max(result, axis=1, keepdims=True)
     

pyerualjetwork/plan_cuda.py

@@ -16,6 +16,8 @@ PyerualJetwork document: https://github.com/HCB06/PyerualJetwork/blob/main/Welco
 """
 
 import cupy as cp
+import numpy as np
+import copy
 
 ### LIBRARY IMPORTS ###
 from .ui import loading_bars, initialize_loading_bar
@@ -80,12 +82,12 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
            weight_evolve=True, neural_web_history=False, show_current_activations=False, auto_normalization=False, target_acc=None,
            neurons_history=False, early_stop=False, show_history=False, loss='categorical_crossentropy',
            interval=33.33, target_loss=None, loss_impact=0.1, acc_impact=0.9,
-           start_this_act=None, start_this_W=None, dtype=cp.float32, memory='gpu'):
+           start_this_act=None, start_this_W=None, neurons=None, hidden=0, activation_functions=None, dtype=cp.float32, memory='gpu'):
     """
     Optimizes the activation functions for a neural network by leveraging train data to find 
-    the most accurate combination of activation potentiation for the given dataset.
+    the most accurate combination of activation potentiation(or activation function) & weight values for the given dataset.
 
-    Why genetic optimization and not backpropagation?
+    Why genetic optimization ENE(Eugenic NeuroEvolution) and not backpropagation?
     Because PLAN is different from other neural network architectures. In PLAN, the learnable parameters are not the weights; instead, the learnable parameters are the activation functions.
     Since activation functions are not differentiable, we cannot use gradient descent or backpropagation. However, I developed a more powerful genetic optimization algorithm: PLANEAT.
 
@@ -114,11 +116,11 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
     :param batch_size: (float, optional): Batch size is used in the prediction process to receive train feedback by dividing the train data into chunks and selecting activations based on randomly chosen partitions. This process reduces computational cost and time while still covering the entire train set due to random selection, so it doesn't significantly impact accuracy. For example, a batch size of 0.08 means each train batch represents %8 of the train set. Default is 1. (%100 of train)
     :param pop_size: (int, optional): Population size of each generation. Default: count of activation functions
     :param weight_evolve: (bool, optional): Activation combinations already optimizes by PLANEAT genetic search algorithm. Should the weight parameters also evolve or should the weights be determined according to the aggregating learning principle of the PLAN algorithm? Default: True (Evolves Weights)
-    :param neural_web_history: (bool, optional): Draws history of neural web. Default is False.
+    :param neural_web_history: (bool, optional): Draws history of neural web. Default is False. [ONLY FOR PLAN MODELS]
     :param show_current_activations: (bool, optional): Should it display the activations selected according to the current strategies during learning, or not? (True or False) This can be very useful if you want to cancel the learning process and resume from where you left off later. After canceling, you will need to view the live training activations in order to choose the activations to be given to the 'start_this' parameter. Default is False
     :param auto_normalization: (bool, optional): Normalization may solves overflow problem. Default: False
     :param target_acc: (float, optional): The target accuracy to stop training early when achieved. Default is None.
-    :param neurons_history: (bool, optional): Shows the history of changes that neurons undergo during the TFL (Test or Train Feedback Learning) stages. True or False. Default is False.
+    :param neurons_history: (bool, optional): Shows the history of changes that neurons undergo during the ENE process. True or False. Default is False. [ONLY FOR PLAN MODELS]
     :param early_stop: (bool, optional): If True, implements early stopping during training.(If train accuracy not improves in two gen stops learning.) Default is False.
     :param show_history: (bool, optional): If True, displays the training history after optimization. Default is False.
     :param loss: (str, optional): For visualizing and monitoring. PLAN neural networks doesn't need any loss function in training. options: ('categorical_crossentropy' or 'binary_crossentropy') Default is 'categorical_crossentropy'.
@@ -128,7 +130,10 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
     :param acc_impact: (float, optional): Impact of accuracy for optimization [0-1]. Default: 0.9
     :param start_this_act: (list, optional): To resume a previously canceled or interrupted training from where it left off, or to continue from that point with a different strategy, provide the list of activation functions selected up to the learned portion to this parameter. Default is None
     :param start_this_W: (cupy.array, optional): To resume a previously canceled or interrupted training from where it left off, or to continue from that point with a different strategy, provide the weight matrix of this genome. Default is None
-    :param dtype: (cupy.dtype): Data type for the arrays. np.float32 by default. Example: cp.float64 or cp.float16.
+    :param hidden: (int, optional): If you dont want train PLAN model this parameter represents a hidden layer count for MLP model. Default: 0 (PLAN)
+    :param neurons: (list[int], optional): If you dont want train PLAN model this parameter represents neuron count of each hidden layer for MLP. Default: None (PLAN)
+    :param activation_functions: (list[str], optional): If you dont want train PLAN model this parameter represents activation function of each hidden layer for MLP. Default: None (PLAN) NOTE: THIS EFFECTS HIDDEN LAYERS OUTPUT. NOT OUTPUT LAYER!
+    :param dtype: (cupy.dtype): Data type for the Weight matrices. np.float32 by default. Example: cp.float64 or cp.float16.
     :param memory: (str): The memory parameter determines whether the dataset to be processed on the GPU will be stored in the CPU's RAM or the GPU's RAM. Options:  'gpu', 'cpu'. Default: 'gpu'.
 
     Returns:
@@ -151,13 +156,13 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
     if gen is None: gen = activation_potentiation_len
 
     if memory == 'gpu':
-        x_train = transfer_to_gpu(x_train, dtype=dtype)
+        x_train = transfer_to_gpu(x_train, dtype=x_train.dtype)
         y_train = transfer_to_gpu(y_train, dtype=y_train.dtype)
 
         from .data_operations_cuda import batcher
 
     elif memory == 'cpu':
-        x_train = transfer_to_cpu(x_train, dtype=dtype)
+        x_train = transfer_to_cpu(x_train, dtype=x_train.dtype)
         y_train = transfer_to_cpu(y_train, dtype=y_train.dtype)
 
         from .data_operations import batcher
@@ -168,6 +173,14 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
     if target_acc is not None and (target_acc < 0 or target_acc > 1): raise ValueError('target_acc must be in range 0 and 1')
     if fit_start is not True and fit_start is not False: raise ValueError('fit_start parameter only be True or False. Please read doc-string')
 
+    if hidden is not 0 and neurons is not None and activation_functions is not None:
+        fit_start = False
+        weight_evolve = True
+        is_mlp = True
+        activation_potentiation = activation_functions
+    else:
+        is_mlp = False
+
     # Initialize visualization components
     viz_objects = initialize_visualization_for_learner(show_history, neurons_history, neural_web_history, x_train, y_train)
 
@@ -183,7 +196,7 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
     progress = initialize_loading_bar(total=activation_potentiation_len, desc="", ncols=79, bar_format=bar_format_learner)
 
     if fit_start is False or pop_size > activation_potentiation_len:
-        weight_pop, act_pop = define_genomes(input_shape=len(x_train[0]), output_shape=len(y_train[0]), population_size=pop_size, dtype=dtype)
+        weight_pop, act_pop = define_genomes(input_shape=len(x_train[0]), output_shape=len(y_train[0]), neurons=neurons, hidden=hidden, activation_functions=activation_functions, population_size=pop_size, dtype=dtype)
         
     else:
         weight_pop = [0] * pop_size
@@ -206,7 +219,7 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
 
             x_train_batch, y_train_batch = batcher(x_train, y_train, batch_size=batch_size)
 
-            x_train_batch = cp.array(x_train_batch, dtype=dtype, copy=False)
+            x_train_batch = cp.array(x_train_batch, dtype=x_train_batch.dtype, copy=False)
             y_train_batch = cp.array(y_train_batch, dtype=y_train.dtype)
 
             if fit_start is True and i == 0 and j < activation_potentiation_len:
@@ -220,7 +233,7 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
             if weight_evolve is False:
                 weight_pop[j] = fit(x_train_batch, y_train_batch, activation_potentiation=act_pop[j], auto_normalization=auto_normalization, dtype=dtype) 
             
-            model = evaluate(x_train_batch, y_train_batch, W=weight_pop[j], activation_potentiation=act_pop[j], auto_normalization=auto_normalization)
+            model = evaluate(x_train_batch, y_train_batch, W=weight_pop[j], activation_potentiation=act_pop[j], auto_normalization=auto_normalization, is_mlp=is_mlp)
             acc = model[get_acc()]
  
             if loss == 'categorical_crossentropy':
@@ -232,18 +245,18 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
 
 
             fitness  = wals(acc, train_loss, acc_impact, loss_impact)
-            target_pop.append(fitness)
+            target_pop.append(fitness.get())
 
             if fitness >= best_fitness:
 
                 best_fitness = fitness
                 best_acc = acc
                 best_loss = train_loss
-                best_weight = cp.copy(weight_pop[j])
+                best_weight = cp.copy(weight_pop[j]) if is_mlp is False else copy.deepcopy(weight_pop[j])
                 best_model = model
 
                 final_activations = act_pop[j].copy() if isinstance(act_pop[j], list) else act_pop[j]
-                final_activations = [final_activations[0]] if len(set(final_activations)) == 1 else final_activations # removing if all same
+                if is_mlp is False: final_activations = [final_activations[0]] if len(set(final_activations)) == 1 else final_activations # removing if all same
 
                 if batch_size == 1:
                     postfix_dict[f"{data} Accuracy"] = cp.round(best_acc, 4)
@@ -278,7 +291,7 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
                 if target_acc is not None and best_acc >= target_acc:
                     progress.close()
                     train_model = evaluate(x_train, y_train, W=best_weight, 
-                                        activation_potentiation=final_activations, auto_normalization=auto_normalization)
+                                        activation_potentiation=final_activations, auto_normalization=auto_normalization, is_mlp=is_mlp)
                     if loss == 'categorical_crossentropy':
                         train_loss = categorical_crossentropy(y_true_batch=y_train, 
                                                            y_pred_batch=train_model[get_preds_softmax()])
@@ -298,7 +311,7 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
                 if target_loss is not None and best_loss <= target_loss:
                     progress.close()
                     train_model = evaluate(x_train, y_train, W=best_weight,
-                                        activation_potentiation=final_activations, auto_normalization=auto_normalization)
+                                        activation_potentiation=final_activations, auto_normalization=auto_normalization, is_mlp=is_mlp)
 
                     if loss == 'categorical_crossentropy':
                         train_loss = categorical_crossentropy(y_true_batch=y_train, 
@@ -320,7 +333,7 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
             progress.update(1)
         
         if batch_size != 1:
-            train_model = evaluate(x_train, y_train, best_weight, final_activations)
+            train_model = evaluate(x_train, y_train, best_weight, final_activations, auto_normalization=auto_normalization, is_mlp=is_mlp)
     
             if loss == 'categorical_crossentropy':
                 train_loss = categorical_crossentropy(y_true_batch=y_train, 
@@ -340,7 +353,9 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
             best_acc_per_gen_list.append(best_acc)
             loss_list.append(best_loss)
 
-        weight_pop, act_pop = optimizer(cp.array(weight_pop, copy=False, dtype=dtype), act_pop, i, cp.array(target_pop, dtype=dtype, copy=False), weight_evolve=weight_evolve, bar_status=False)
+        if is_mlp is False: weight_pop = cp.array(weight_pop, copy=False, dtype=dtype)
+    
+        weight_pop, act_pop = optimizer(weight_pop, act_pop, i, np.array(target_pop), weight_evolve=weight_evolve, is_mlp=is_mlp, bar_status=False)
         target_pop = []
 
         # Early stopping check
@@ -348,7 +363,7 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
             if best_acc_per_gen_list[i] == best_acc_per_gen_list[i-1]:
                 progress.close()
                 train_model = evaluate(x_train, y_train, W=best_weight, 
-                                    activation_potentiation=final_activations, auto_normalization=auto_normalization)
+                                    activation_potentiation=final_activations, auto_normalization=auto_normalization, is_mlp=is_mlp)
                 
             if loss == 'categorical_crossentropy':
                 train_loss = categorical_crossentropy(y_true_batch=y_train, 
@@ -369,7 +384,7 @@ def learner(x_train, y_train, optimizer, fit_start=True, gen=None, batch_size=1,
     # Final evaluation
     progress.close()
     train_model = evaluate(x_train, y_train, W=best_weight,
-                        activation_potentiation=final_activations, auto_normalization=auto_normalization)
+                        activation_potentiation=final_activations, auto_normalization=auto_normalization, is_mlp=is_mlp)
 
     if loss == 'categorical_crossentropy':
         train_loss = categorical_crossentropy(y_true_batch=y_train, 
@@ -391,7 +406,8 @@ def evaluate(
     y_test,
     W,
     activation_potentiation=['linear'],
-    auto_normalization=False
+    auto_normalization=False,
+    is_mlp=False
 ) -> tuple:
     """
     Evaluates the neural network model using the given test data.
@@ -407,6 +423,8 @@ def evaluate(
 
         auto_normalization (bool, optional): Normalization for x_test ? Default = False.
         
+        is_mlp (bool, optional): Evaluate PLAN model or MLP model ? Default: False (PLAN)
+
     Returns:
         tuple: Model (list).
     """
@@ -415,12 +433,22 @@ def evaluate(
 
     if isinstance(activation_potentiation, str):
         activation_potentiation = [activation_potentiation]
+    elif isinstance(activation_potentiation, list):
+        activation_potentiation = [item if isinstance(item, list) or isinstance(item, str) else [item] for item in activation_potentiation]
+
+    if is_mlp:
+        layer = x_test
+        for i in range(len(W)):
+            if i != len(W) - 1 and i != 0: layer = apply_activation(layer, activation_potentiation[i])
+
+            layer = layer @ W[i].T
+
+        result = layer
+
     else:
-        activation_potentiation = [item if isinstance(item, list) else [item] for item in activation_potentiation]
 
-    x_test = apply_activation(x_test, activation_potentiation)
-    
-    result = x_test @ W.T
+        x_test = apply_activation(x_test, activation_potentiation)
+        result = x_test @ W.T
 
     max_vals = cp.max(result, axis=1, keepdims=True)
 

pyerualjetwork/planeat.py

@@ -15,6 +15,7 @@ PyerualJetwork document: https://github.com/HCB06/Anaplan/blob/main/Welcome_to_A
 import numpy as np
 import random
 import math
+import copy
 
 ### LIBRARY IMPORTS ###
 from .data_operations import normalization, non_neg_normalization
@@ -134,6 +135,7 @@ def evolver(weights,
             weight_mutate_threshold=16,
             weight_mutate_prob=1,
             is_mlp=False,
+            save_best_genome=False,
             dtype=np.float32):
    """
    Applies the evolving process of a population of genomes using selection, crossover, mutation, and activation function potentiation.
@@ -222,6 +224,8 @@ def evolver(weights,
 
       is_mlp (bool, optional): Evolve PLAN model or MLP model ? Default: False (PLAN)
       
+      save_best_genome (bool, optional): Save the best genome of the previous generation to the next generation. Default: False
+
       dtype (numpy.dtype, optional): Data type for the arrays. Default: np.float32. 
          Example: np.float64 or np.float16 [fp32 for balanced devices, fp64 for strong devices, fp16 for weak devices: not recommended!].
 
@@ -316,6 +320,8 @@ def evolver(weights,
 
 ### FITNESS IS SORTED IN ASCENDING ORDER, AND THE WEIGHT AND ACTIVATIONS OF EACH GENOME ARE SORTED ACCORDING TO THIS ORDER:
 
+   previous_best_genome_index = np.argmax(fitness)
+
    sort_indices = np.argsort(fitness)
 
    fitness = fitness[sort_indices]
@@ -327,7 +333,7 @@ def evolver(weights,
 
    good_weights = weights[slice_center:]
    bad_weights = weights[:slice_center]
-   best_weight = np.copy(good_weights[-1])
+   best_weight = np.copy(good_weights[-1]) if is_mlp is False else copy.deepcopy(good_weights[-1])
 
    good_activations = list(activation_potentiations[slice_center:])
    bad_activations = list(activation_potentiations[:slice_center])
@@ -354,7 +360,7 @@ def evolver(weights,
    for i in range(len(bad_weights)):
       
       if policy == 'aggressive':
-         first_parent_W = best_weight
+         first_parent_W = np.copy(best_weight)
          first_parent_act = best_activations
          first_parent_fitness = best_fitness
 
@@ -457,9 +463,6 @@ def evolver(weights,
                                                 )
 
       if bar_status: progress.update(1)
-   
-   child_W[0] = best_weight
-   child_act[0] = best_activations
 
    if is_mlp:
       weights = np.vstack((child_W, mutated_W), dtype=object)
@@ -468,7 +471,10 @@ def evolver(weights,
    
    activation_potentiations = child_act + mutated_act
 
-
+   if save_best_genome:
+      weights[0] = best_weight
+      activation_potentiations[0] = best_activations
+   
    ### INFO PRINTING CONSOLE
    
    if show_info == True:
@@ -491,7 +497,8 @@ def evolver(weights,
       print("  ACTIVATION SELECTION THRESHOLD:", str(activation_selection_threshold))
       print("  ACTIVATION SELECTION ADD PROB: ", str(activation_selection_add_prob))
       print("  ACTIVATION SELECTION CHANGE PROB: ", str(activation_selection_change_prob))
-      print("  FITNESS BIAS: ", str(fitness_bias) + '\n')
+      print("  FITNESS BIAS: ", str(fitness_bias))
+      print("  SAVE BEST GENOME: ", str(save_best_genome) + '\n')
 
 
       print("*** Performance ***")
@@ -500,8 +507,8 @@ def evolver(weights,
       print("  MIN FITNESS: ", str(round(min(fitness), 2)) + '\n')
 
       print("  BEST GENOME ACTIVATION LENGTH: ", str(len(best_activations)))
-      print("  BEST GENOME INDEX: ", str(0))
-      print("  NOTE: The returned genome at the first index is the best of the previous generation." + '\n')
+      print("  PREVIOUS BEST GENOME ACTIVATION LENGTH: ", str(len(best_activations)))
+      print("  PREVIOUS BEST GENOME INDEX: ", str(previous_best_genome_index) + '\n')
       
    if weight_evolve is False: weights = origin_weights
 
@@ -536,17 +543,18 @@ def evaluate(Input, weights, activation_potentiations, is_mlp=False):
       the output for each genome in population.
    """
    ### THE OUTPUTS ARE RETURNED, WHERE EACH GENOME'S OUTPUT MATCHES ITS INDEX:
-
-
+   
    if isinstance(activation_potentiations, str):
       activation_potentiations = [activation_potentiations]
    elif isinstance(activation_potentiations, list):
       activation_potentiations = [item if isinstance(item, list) or isinstance(item, str) else [item] for item in activation_potentiations]
 
+
    if is_mlp:
       layer = Input
       for i in range(len(weights)):
          if i != len(weights) - 1 and i != 0: layer = apply_activation(layer, activation_potentiations[i])
+
          layer = layer @ weights[i].T
       
       return layer
@@ -733,7 +741,7 @@ def cross_over(first_parent_W,
       
       threshold = abs(activation_selection_threshold / succes)
       new_threshold = threshold
-      
+
       while True:
          
          random_index_undominant = int(random.uniform(0, len(undominant_parent_act)-1))

pyerualjetwork/planeat_cuda.py

@@ -16,7 +16,7 @@ import cupy as cp
 import numpy as np
 import random
 import math
-
+import copy
 
 ### LIBRARY IMPORTS ###
 from .data_operations_cuda import normalization, non_neg_normalization
@@ -135,6 +135,7 @@ def evolver(weights,
             weight_mutate_threshold=16,
             weight_mutate_prob=1,
             is_mlp=False,
+            save_best_genome=False,
             dtype=cp.float32):
    """
    Applies the evolving process of a population of genomes using selection, crossover, mutation, and activation function potentiation.
@@ -223,6 +224,8 @@ def evolver(weights,
 
       is_mlp (bool, optional): Evolve PLAN model or MLP model ? Default: False (PLAN)
 
+      save_best_genome (bool, optional): Save the best genome of the previous generation to the next generation. Default: False
+
       dtype (cupy.dtype): Data type for the arrays. Default: cp.float32. 
          Example: cp.float64 or cp.float16 [fp32 for balanced devices, fp64 for strong devices, fp16 for weak devices: not recommended!].
 
@@ -307,7 +310,7 @@ def evolver(weights,
    else:
       raise ValueError("genome population size must be even number. for example: not 99, make 100 or 98.")
 
-   if weight_evolve is False: origin_weights = cp.copy(weights)
+   if weight_evolve is False: origin_weights = cp.copy(weights) if is_mlp else copy.deepcopy(weights)
 
    if is_mlp:
       activation_mutate_add_prob = 0
@@ -326,6 +329,8 @@ def evolver(weights,
 
 ### FITNESS LIST IS SORTED IN ASCENDING ORDER, AND THE WEIGHT AND ACTIVATIONS OF EACH GENOME ARE SORTED ACCORDING TO THIS ORDER:
 
+   previous_best_genome_index = np.argmax(fitness)
+
    sort_indices = cp.argsort(fitness)
 
    fitness = fitness[sort_indices]
@@ -471,9 +476,6 @@ def evolver(weights,
             
       if bar_status: progress.update(1)
 
-   child_W[0] = best_weight
-   child_act[0] = best_activations
-
    if is_mlp:
       for i in range(len(child_W)):
          for j in range(len(child_W[i])):
@@ -487,6 +489,10 @@ def evolver(weights,
 
    activation_potentiations = child_act + mutated_act
 
+   if save_best_genome:
+      weights[0] = best_weight
+      activation_potentiations[0] = best_activations
+
    ### INFO PRINTING CONSOLE
    
    if show_info == True:
@@ -509,7 +515,8 @@ def evolver(weights,
       print("  ACTIVATION SELECTION THRESHOLD:", str(activation_selection_threshold))
       print("  ACTIVATION SELECTION ADD PROB: ", str(activation_selection_add_prob))
       print("  ACTIVATION SELECTION CHANGE PROB: ", str(activation_selection_change_prob))
-      print("  FITNESS BIAS: ", str(fitness_bias) + '\n')
+      print("  FITNESS BIAS: ", str(fitness_bias))
+      print("  SAVE BEST GENOME: ", str(save_best_genome) + '\n')
 
       print("*** Performance ***")
       print("  MAX FITNESS: ", str(cp.round(max(fitness), 2)))
@@ -517,8 +524,8 @@ def evolver(weights,
       print("  MIN FITNESS: ", str(cp.round(min(fitness), 2)) + '\n')
 
       print("  BEST GENOME ACTIVATION LENGTH: ", str(len(best_activations)))
-      print("  BEST GENOME INDEX: ", str(0))
-      print("  NOTE: The returned genome at the first index is the best of the previous generation." + '\n')
+      print("  PREVIOUS BEST GENOME ACTIVATION LENGTH: ", str(len(best_activations)))
+      print("  PREVIOUS BEST GENOME INDEX: ", str(previous_best_genome_index) + '\n')
       
    if weight_evolve is False: weights = origin_weights
 
@@ -557,9 +564,9 @@ def evaluate(Input, weights, activation_potentiations, is_mlp=False):
    ### THE OUTPUTS ARE RETURNED WHERE EACH GENOME'S OUTPUT MATCHES ITS INDEX:
    
    if isinstance(activation_potentiations, str):
-      activation_potentiations = [activation_potentiations]
-   elif isinstance(activation_potentiations, list):
-      activation_potentiations = [item if isinstance(item, list) or isinstance(item, str) else [item] for item in activation_potentiations]
+    activation_potentiations = [activation_potentiations]
+   else:
+    activation_potentiations = [item if isinstance(item, list) else [item] for item in activation_potentiations]
 
    if is_mlp:
       

{pyerualjetwork-4.7.2.dist-info → pyerualjetwork-4.8.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: pyerualjetwork
-Version: 4.7.2
+Version: 4.8
 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

{pyerualjetwork-4.7.2.dist-info → pyerualjetwork-4.8.dist-info}/RECORD

@@ -1,4 +1,4 @@
-pyerualjetwork/__init__.py,sha256=hgsCywyfA2zoxYuwVY5frkHQxzZiurKJH7nWkbg4_pY,1279
+pyerualjetwork/__init__.py,sha256=S8sXWrlXoix9qh1HnSSCbj-VRK0hY3_yCQb8IN9Ms4c,1277
 pyerualjetwork/activation_functions.py,sha256=X7Kv8qv8oZq8hvTdUiV-GkFjKHRlKIQypRPXh6gdkm4,7614
 pyerualjetwork/activation_functions_cuda.py,sha256=pefklsl9QuSVbKwiUUHeF_ExN0bICH7QIF1MfoMU40Q,7665
 pyerualjetwork/data_operations.py,sha256=XKYG9-mLa3qKAXUjejuD7V8aJKjpl5PdQwKzPFjpKgs,15437
@@ -12,14 +12,14 @@ pyerualjetwork/metrics.py,sha256=q7MkhnZDRbCjFBDDfUgrl8lBYnUT_1ro1LxeBq105pI,607
 pyerualjetwork/metrics_cuda.py,sha256=73h9GC7XwmnFCVzFEEiPQfF8CwHIz2wsCbxpZrJtYgw,5061
 pyerualjetwork/model_operations.py,sha256=kTLhHlu8NSFBTTWDVyMdN7PsDk2emFU9_6JgH6_Mh8k,15268
 pyerualjetwork/model_operations_cuda.py,sha256=bK1SKi_ar7UVL7jwDftSyg4g5wMLsYT1z_ZPqFAsrKo,16329
-pyerualjetwork/plan.py,sha256=cjVblo8TxTHX-GZPvgQvJZ34nOmzxSvtCrQi9K-Mhog,23268
-pyerualjetwork/plan_cuda.py,sha256=ZEU_b_EoA-zPk7Gn94L_XBZz1v4mn8DOUsjTNV6fp8Q,24230
-pyerualjetwork/planeat.py,sha256=_AkG-EJj8M7M1fCoXLQpttXud34X6bbaFkaD117UmAc,44717
-pyerualjetwork/planeat_cuda.py,sha256=YEpqxwhFYxHBxZNbDFXlI4d-r2S46OalLmKhk86pFBI,45311
+pyerualjetwork/plan.py,sha256=h7_00bqAP0VOWpXUumAtw71BEhtYZQPVc_iTzVoJ5fo,24922
+pyerualjetwork/plan_cuda.py,sha256=rffz03OmN3wdB9pdxSFML62okD8WuPilLxGCy-nOR5A,26028
+pyerualjetwork/planeat.py,sha256=J7tgHoM3P2tU43RJNgwDNi8ocsE-Cq5tIXvDCYXtMwY,45138
+pyerualjetwork/planeat_cuda.py,sha256=Uczjey02GknfNAUObJVkZwQTro2L_z46ypcEo34Y0aw,45634
 pyerualjetwork/ui.py,sha256=JBTFYz5R24XwNKhA3GSW-oYAoiIBxAE3kFGXkvm5gqw,656
 pyerualjetwork/visualizations.py,sha256=utnX9zQhzmtvBJLOLNGm2jecVVk4zHXABQdjb0XzJac,28352
 pyerualjetwork/visualizations_cuda.py,sha256=gnoaaazZ-nc9E1ImqXrZBRgQ4Rnpi2qh2yGJ2eLKMlE,28807
-pyerualjetwork-4.7.2.dist-info/METADATA,sha256=qNbqS7ETCQ3sKRg5of3a_EkvW-rth4zDvaaCaMjKmwo,7505
-pyerualjetwork-4.7.2.dist-info/WHEEL,sha256=2wepM1nk4DS4eFpYrW1TTqPcoGNfHhhO_i5m4cOimbo,92
-pyerualjetwork-4.7.2.dist-info/top_level.txt,sha256=BRyt62U_r3ZmJpj-wXNOoA345Bzamrj6RbaWsyW4tRg,15
-pyerualjetwork-4.7.2.dist-info/RECORD,,
+pyerualjetwork-4.8.dist-info/METADATA,sha256=KTYIGCZhUJIWnAIMFtHGUcLuWtd6DYBw-656ynLEfTI,7503
+pyerualjetwork-4.8.dist-info/WHEEL,sha256=2wepM1nk4DS4eFpYrW1TTqPcoGNfHhhO_i5m4cOimbo,92
+pyerualjetwork-4.8.dist-info/top_level.txt,sha256=BRyt62U_r3ZmJpj-wXNOoA345Bzamrj6RbaWsyW4tRg,15
+pyerualjetwork-4.8.dist-info/RECORD,,