| import numpy as np | |
| from matrices import construir_matriz_Q, generar_k_transpuesta_y_v | |
| from attention import encoder_layer, calcular_pesos_proyeccion, scaled_dot_product_attention_2D | |
| def transform_word(word): | |
| significant_numbers = [] | |
| for char in word: | |
| Q = construir_matriz_Q(char) | |
| K_transpose, V = generar_k_transpuesta_y_v(Q, char) | |
| output = scaled_dot_product_attention_2D(Q, K_transpose, V) | |
| Wq, Wk, Wv = calcular_pesos_proyeccion(output) | |
| encoder_output = encoder_layer(Q, K_transpose, V, Wq, Wk, Wv) | |
| significant_number = np.sum(np.max(encoder_output, axis=1)) | |
| significant_numbers.append(significant_number) | |
| return significant_numbers | |