visualizations.py

import matplotlib.pyplot as plt
import seaborn as sns
import plotly.express as px
from sklearn.decomposition import PCA

def plot_attention_heatmaps(attention_data, tokens, layer_index):
    """
    Plots all attention heads for a given layer as heatmaps.
    """
    # Select the tensor for the chosen layer.
    attention_for_layer = attention_data[layer_index].squeeze(0)
    num_heads = attention_for_layer.shape[0]

    # Create a 3x4 grid for the 12 attention heads.
    fig, axes = plt.subplots(3, 4, figsize=(15, 12))
    fig.suptitle(f"Dikkat Desenleri - Katman {layer_index+1}", fontsize=16)

    axes = axes.flatten()

    for i in range(num_heads):
        ax = axes[i]
        head_attention = head_attention = attention_for_layer[i].detach().numpy()

        sns.heatmap(head_attention, xticklabels=tokens, yticklabels=tokens,
                    cmap='viridis', ax=ax, cbar=False)

        ax.set_title(f"Dikkat Kafası {i+1}")
        ax.tick_params(axis='x', rotation=90) # Improve label readability.
        ax.tick_params(axis='y', rotation=0)

    for i in range(num_heads, len(axes)):
        axes[i].axis('off')

    plt.tight_layout(rect=[0, 0, 1, 0.96])
    return fig

def plot_embedding_space(hidden_states, tokens):
    """
    Visualizes word embeddings in 2D space using PCA.
    """
    # Get last layer's embeddings and convert to numpy.
    last_layer_embeddings = hidden_states[-1].squeeze(0).detach().numpy()

    # Reduce 768 dimensions to 2 for plotting.
    pca = PCA(n_components=2)
    embeddings_2d = pca.fit_transform(last_layer_embeddings)

    # Create the plot with Plotly for interactivity.
    fig = px.scatter(
        x=embeddings_2d[:, 0],
        y=embeddings_2d[:, 1],
        text=tokens,
        title="Kelime Gömülmeleri Uzayı (PCA ile 2D)"
    )
    
    fig.update_traces(textposition='top center')
    fig.update_layout(
        xaxis_title="Ana Bileşen 1 (Principal Component 1)",
        yaxis_title="Ana Bileşen 2 (Principal Component 2)"
    )

    return fig