pyerualjetwork 3.3.4__py3-none-any.whl → 4.1.9__py3-none-any.whl

pyerualjetwork/visualizations.py

@@ -68,131 +68,117 @@ def draw_neural_web(W, ax, G, return_objs=False):
         return art1, art2, art3
 
 
-def draw_model_architecture(model_name, model_path='', style='basic'):
+def draw_model_architecture(model_name, model_path=''):
     """
-    Visualizes the architecture of a neural network model.
-
-    Parameters
-    ----------
-    model_name : str
-        The name of the model to be visualized, which will be displayed in the title or label.
-    
-    model_path : str
-        The file path to the model, from which the architecture is loaded. Default is ''
-    
-    style : str, optional
-        The style of the visualization. 
-        Options:
-            - 'basic': Displays a simplified view of the model architecture.
-            - 'detailed': Shows a more comprehensive view, including layer details and parameters.
-        Default is 'basic'.
-    
-    Returns
-    -------
-    None
-        Draws and displays the architecture of the specified model.
-
-        
-    Examples
-    --------
-    >>> draw_model_architecture("MyModel", "path/to/model", style='detailed')
+    The `draw_model_architecture` function visualizes the architecture of a neural network model with
+    multiple inputs based on activation functions.
+    
+    :param model_name: The `model_name` parameter in the `draw_model_architecture` function is used to
+    specify the name of the neural network model whose architecture you want to visualize. This function
+    visualizes the architecture of a neural network model with multiple inputs based on activation
+    functions
+    :param model_path: The `model_path` parameter in the `draw_model_architecture` function is used to
+    specify the path where the neural network model is saved. If the model is saved in a specific
+    directory or file location, you can provide that path as a string when calling the function. If the
+    model is saved
     """
-    from .plan import get_scaler, get_act_pot, get_weights
-    from .model_operations import load_model
+    """
+    Visualizes the architecture of a neural network model with multiple inputs based on activation functions.
+    """
+    
+    from .model_operations import load_model, get_scaler, get_act_pot, get_weights
     
     model = load_model(model_name=model_name, model_path=model_path)
     
     W = model[get_weights()]
     activation_potentiation = model[get_act_pot()]
     scaler_params = model[get_scaler()]
-
-    text_1 = f"Input Shape:\n{W.shape[1]}"
-    text_2 = f"Output Shape:\n{W.shape[0]}"
-
-    if scaler_params is None:
-        bottom_left_text = 'Standard Scaler=No'
-    else:
-        bottom_left_text = 'Standard Scaler=Yes'
-
-    if len(activation_potentiation) != 1 or (len(activation_potentiation) == 1 and activation_potentiation[0] != 'linear'):
-
-        bottom_left_text_1 = f'Aggregation Layers(Aggregates All Conversions)={len(activation_potentiation)}'
-
-    else:
-
-        bottom_left_text_1 = 'Aggregation Layers(Aggregates All Conversions)=0'
-
-    bottom_left_text_2 = 'Potentiation Layer(Fully Connected)=1'
-
-    if scaler_params is None:
-        bottom_left_text = 'Standard Scaler=No'
-    else:
-        bottom_left_text = 'Standard Scaler=Yes'
-
-    num_middle_axes = len(activation_potentiation)
-
-    if style == 'detailed':
-
-        col = 1
-
-    elif style == 'basic':
     
-        col = 2
+    # Calculate dimensions based on number of activation functions
+    num_activations = len(activation_potentiation)
+    input_groups = num_activations  # Number of input groups equals number of activations
+    num_inputs = W.shape[1]
 
-    fig, axes = plt.subplots(1, num_middle_axes + col, figsize=(5 * (num_middle_axes + 2), 5))
-
-    fig.suptitle("Model Architecture", fontsize=16, fontweight='bold')
-
-    for i, activation in enumerate(activation_potentiation):
-        x = np.linspace(-100, 100, 100)
-        translated_x_train = draw_activations(x, activation)
-        y = translated_x_train
-
-        axes[i].plot(x, y, color='b', markersize=6, linewidth=2, label='Activations Over Depth')
-        axes[i].set_title(activation_potentiation[i])
-
-        axes[i].spines['top'].set_visible(False)
-        axes[i].spines['right'].set_visible(False)
-        axes[i].spines['left'].set_visible(False)
-        axes[i].spines['bottom'].set_visible(False)
-        axes[i].get_xaxis().set_visible(False)
-        axes[i].get_yaxis().set_visible(False)
+    # Create figure
+    fig = plt.figure(figsize=(15, 10))
+    
+    # Calculate positions for nodes
+    def get_node_positions():
+        positions = {}
         
-
-        if i < num_middle_axes - 1:
-            axes[i].annotate('', xy=(1.05, 0.5), xytext=(0.95, 0.5), 
-                            xycoords='axes fraction', textcoords='axes fraction',
-                            arrowprops=dict(arrowstyle="->", color='black', lw=1.5))
+        # Input layer positions
+        total_height = 0.8  # Maksimum dikey alan
+        group_height = total_height / input_groups  # Her grup için ayrılan dikey alan
+        input_spacing = min(group_height / (num_inputs + 1), 0.1)  # Her girdi arasındaki mesafe
+        
+        for group in range(input_groups):
+            group_start_y = 0.9 - (group * group_height)  # Grubun başlangıç y koordinatı
+            for i in range(num_inputs):
+                y_pos = group_start_y - ((i + 1) * input_spacing)
+                positions[f'input_{group}_{i}'] = (0.2, y_pos)
+        
+        # Aggregation layer positions
+        agg_spacing = total_height / (num_inputs + 1)
+        for i in range(num_inputs):
+            positions[f'summed_{i}'] = (0.5, 0.9 - ((i + 1) * agg_spacing))
+        
+        # Output layer positions
+        output_spacing = total_height / (W.shape[0] + 1)
+        for i in range(W.shape[0]):
+            positions[f'output_{i}'] = (0.8, 0.9 - ((i + 1) * output_spacing))
+        
+        return positions
+    
+    # Draw the network
+    pos = get_node_positions()
+    
+    # Draw nodes
+    for group in range(input_groups):
+        # Draw input nodes
+        for i in range(num_inputs):
+            plt.plot(*pos[f'input_{group}_{i}'], 'o', color='lightgreen', markersize=20)
+            plt.text(pos[f'input_{group}_{i}'][0] - 0.05, pos[f'input_{group}_{i}'][1], 
+                    f'Input #{i+1} ({activation_potentiation[group]})', ha='right', va='center')
+            
+            # Draw connections from input to summed input directly
+            plt.plot([pos[f'input_{group}_{i}'][0], pos[f'summed_{i}'][0]],
+                    [pos[f'input_{group}_{i}'][1], pos[f'summed_{i}'][1]], 'k-')
+            # Draw aggregation nodes
+            if group == 0:
+                plt.plot(*pos[f'summed_{i}'], 'o', color='lightgreen', markersize=20)
+                plt.text(pos[f'summed_{i}'][0], pos[f'summed_{i}'][1] + 0.02,
+                        f'Summed\nInput #{i+1}', ha='center', va='bottom')
+    
+    # Draw output nodes and connections
+    for i in range(W.shape[0]):
+        plt.plot(*pos[f'output_{i}'], 'o', color='gold', markersize=20)
+        plt.text(pos[f'output_{i}'][0] + 0.05, pos[f'output_{i}'][1],
+                f'Output #{i+1}', ha='left', va='center', color='purple')
+        
+        # Connect all aggregation nodes to each output
+        for group in range(num_inputs):
+            plt.plot([pos[f'summed_{group}'][0], pos[f'output_{i}'][0]],
+                    [pos[f'summed_{group}'][1], pos[f'output_{i}'][1]], 'k-')
     
-    if style == 'detailed':
+    # Add labels and annotations
+    plt.text(0.2, 0.95, 'Input Layer', ha='center', va='bottom', fontsize=12)
+    plt.text(0.5, 0.95, 'Aggregation\nLayer', ha='center', va='bottom', fontsize=12)
+    plt.text(0.8, 0.95, 'Output Layer', ha='center', va='bottom', fontsize=12)
     
-        G = nx.Graph()
-        draw_neural_web(W=W, ax=axes[num_middle_axes], G=G)
-
-    elif style == 'basic':
-    
-        circle1 = plt.Circle((0.5, 0.5), 0.4, color='skyblue', ec='black', lw=1.5)
-        axes[num_middle_axes].add_patch(circle1)
-        axes[num_middle_axes].text(0.5, 0.5, text_1, ha='center', va='center', fontsize=12)
-        axes[num_middle_axes].set_xlim(0, 1)
-        axes[num_middle_axes].set_ylim(0, 1)
-        axes[num_middle_axes].axis('off') 
-
-        circle2 = plt.Circle((0.5, 0.5), 0.4, color='lightcoral', ec='black', lw=1.5)
-        axes[-1].add_patch(circle2)
-        axes[-1].text(0.5, 0.5, text_2, ha='center', va='center', fontsize=12)
-        axes[-1].set_xlim(0, 1)
-        axes[-1].set_ylim(0, 1)
-        axes[-1].axis('off') 
-  
-     
-    fig.text(0.01, 0, bottom_left_text, ha='left', va='bottom', fontsize=10)
-    fig.text(0.01, 0.04, bottom_left_text_1, ha='left', va='bottom', fontsize=10)
-    fig.text(0.01, 0.08, bottom_left_text_2, ha='left', va='bottom', fontsize=10)
+    # Remove axes
+    plt.axis('off')
+    
+    # Add model information
+    if scaler_params is None:
+        plt.text(0.95, 0.05, 'Standard Scaler=No', fontsize=10, ha='right', va='bottom')
+    else:
+        plt.text(0.95, 0.05, 'Standard Scaler=Yes', fontsize=10, ha='right', va='bottom')
 
+    # Add model architecture title
+    plt.text(0.95, 0.1, f"PLAN Model Architecture: {model_name}", fontsize=12, ha='right', va='bottom', fontweight='bold')
     plt.tight_layout()
     plt.show()
-    
+
     
 def draw_activations(x_train, activation):
 
@@ -332,10 +318,12 @@ def draw_activations(x_train, activation):
 
     elif activation == 'spiral':
         result = af.spiral_activation(x_train)
-
-    return result
-
-
+    
+    try: return result
+    except:
+        print('WARNING: error in drawing some activation.')
+        return x_train
+ 
 def plot_evaluate(x_test, y_test, y_preds, acc_list, W, activation_potentiation):
     
     from .metrics import metrics, confusion_matrix, roc_curve
@@ -479,7 +467,7 @@ def plot_decision_boundary(x, y, activation_potentiation, W, artist=None, ax=Non
                          np.arange(y_min, y_max, h))
     
     grid = np.c_[xx.ravel(), yy.ravel()]
-    grid_full = np.zeros((grid.shape[0], x.shape[1]))
+    grid_full = np.zeros((grid.shape[0], x.shape[1]), dtype=np.float32)
     grid_full[:, feature_indices] = grid
     
     Z = [None] * len(grid_full)
@@ -487,7 +475,7 @@ def plot_decision_boundary(x, y, activation_potentiation, W, artist=None, ax=Non
     for i in range(len(grid_full)):
         Z[i] = np.argmax(predict_model_ram(grid_full[i], W=W, activation_potentiation=activation_potentiation))
 
-    Z = np.array(Z)
+    Z = np.array(Z, dtype=np.int32)
     Z = Z.reshape(xx.shape)
 
     if ax is None:
@@ -513,11 +501,11 @@ def plot_decision_boundary(x, y, activation_potentiation, W, artist=None, ax=Non
         
         except:
 
-            art1_1 = ax[0].contourf(xx, yy, Z, alpha=0.8)
-            art1_2 = ax[0].scatter(x[:, feature_indices[0]], x[:, feature_indices[1]], c=decode_one_hot(y), edgecolors='k', marker='o', s=20, alpha=0.9)
-            ax[0].set_xlabel(f'Feature {0 + 1}')
-            ax[0].set_ylabel(f'Feature {1 + 1}')
-            ax[0].set_title('Decision Boundary')
+            art1_1 = ax.contourf(xx, yy, Z, alpha=0.8)
+            art1_2 = ax.scatter(x[:, feature_indices[0]], x[:, feature_indices[1]], c=decode_one_hot(y), edgecolors='k', marker='o', s=20, alpha=0.9)
+            ax.set_xlabel(f'Feature {0 + 1}')
+            ax.set_ylabel(f'Feature {1 + 1}')
+            ax.set_title('Decision Boundary')
 
 
             return art1_1, art1_2
@@ -558,7 +546,7 @@ def plot_decision_space(x, y, y_preds=None, s=100, color='tab20'):
             if y_preds[i] == cls:
                 class_points.append(X_pca[i])
         
-        class_points = np.array(class_points)
+        class_points = np.array(class_points, dtype=y.dtype)
         
 
         if len(class_points) > 2:
@@ -573,8 +561,8 @@ def plot_decision_space(x, y, y_preds=None, s=100, color='tab20'):
 
     plt.draw()
     
-        
-def neuron_history(LTPW, ax1, row, col, class_count, artist5, data, fig1, acc=False, loss=False):
+
+def update_neuron_history_for_learner(LTPW, ax1, row, col, class_count, artist5, data, fig1, acc=False, loss=False):
 
     for j in range(len(class_count)):
         
@@ -601,15 +589,40 @@ def neuron_history(LTPW, ax1, row, col, class_count, artist5, data, fig1, acc=Fa
 
     return artist5
 
+def update_neuron_history(LTPW, ax1, row, col, class_count, artist5, fig1, acc=False, loss=False):
+
+    for j in range(class_count):
+        
+            if acc != False and loss != False:
+                suptitle_info = ' Accuracy:' + str(acc) + '\n' + '\nNeurons Memory:'
+            else:
+                suptitle_info = 'Neurons Memory:'
+
+            mat = LTPW[j,:].reshape(row, col)
+
+            title_info = f'{j+1}. Neuron'
+            
+            art5 = ax1[j].imshow(mat, interpolation='sinc', cmap='viridis')
+
+            ax1[j].set_aspect('equal')
+            ax1[j].set_xticks([])
+            ax1[j].set_yticks([])
+            ax1[j].set_title(title_info)
+
+           
+            artist5.append([art5])
+
+    fig1.suptitle(suptitle_info, fontsize=16)
+
 
 def initialize_visualization_for_fit(val, show_training, neurons_history, x_train, y_train):
     """Initializes the visualization setup based on the parameters."""
     from .data_operations import find_closest_factors
     visualization_objects = {}
 
-    if show_training:
+    if show_training or neurons_history:
         if not val:
-            raise ValueError("For showing training, 'val' parameter must be True.")
+            raise ValueError("For showing training or neurons history, 'val' parameter must be True.")
 
         G = nx.Graph()
         fig, ax = plt.subplots(2, 2)
@@ -626,7 +639,7 @@ def initialize_visualization_for_fit(val, show_training, neurons_history, x_trai
 
     if neurons_history:
         row, col = find_closest_factors(len(x_train[0]))
-        fig1, ax1 = plt.subplots(1, len(set(y_train)), figsize=(18, 14))
+        fig1, ax1 = plt.subplots(1, len(y_train[0]), figsize=(18, 14))
         visualization_objects.update({
             'fig1': fig1,
             'ax1': ax1,
@@ -637,6 +650,16 @@ def initialize_visualization_for_fit(val, show_training, neurons_history, x_trai
 
     return visualization_objects
 
+        
+
+def update_neural_web_for_fit(W, ax, G, artist):
+    """
+    The function `update_neural_web_for_fit` updates a neural web visualization for fitting.
+    """
+    art5_1, art5_2, art5_3 = draw_neural_web(W=W, ax=ax, G=G, return_objs=True)
+    art5_list = [art5_1] + [art5_2] + list(art5_3.values())
+    artist.append(art5_list)
+
 
 def update_weight_visualization_for_fit(ax, LTPW, artist2):
     """Updates the weight visualization plot."""
@@ -673,11 +696,12 @@ def update_validation_history_for_fit(ax, val_list, artist3):
 def display_visualization_for_fit(fig, artist_list, interval):
     """Displays the animation for the given artist list."""
     ani = ArtistAnimation(fig, artist_list, interval=interval, blit=True)
+    return ani
+
+def show():
     plt.tight_layout()
     plt.show()
 
-
-
 def initialize_visualization_for_learner(show_history, neurons_history, neural_web_history, x_train, y_train):
     """Initialize all visualization components"""
     from .data_operations import find_closest_factors
@@ -765,7 +789,7 @@ def display_visualizations_for_learner(viz_objects, best_weights, data, best_acc
     if 'neurons' in viz_objects:
         neurons = viz_objects['neurons']
         for _ in range(10):
-            neurons['artists'] = neuron_history(
+            neurons['artists'] = update_neuron_history_for_learner(
                 np.copy(best_weights), 
                 neurons['ax'],
                 neurons['row'],