"""Small, dependency-free helpers for the IndoBERT embedding demo.""" from __future__ import annotations from math import sqrt import numpy as np def parse_input_lines(raw_text: str, max_samples: int = 6, max_chars: int = 280) -> list[str]: """Parse newline-separated samples and enforce the demo's input limits.""" texts = [line.strip() for line in raw_text.splitlines() if line.strip()] if not texts: raise ValueError("Masukkan setidaknya satu kalimat Bahasa Indonesia.") if len(texts) > max_samples: raise ValueError(f"Maksimal {max_samples} kalimat dalam satu kali analisis.") if any(len(text) > max_chars for text in texts): raise ValueError(f"Setiap kalimat dibatasi hingga {max_chars} karakter.") return texts def cosine_similarity_matrix(vectors: list[list[float]]) -> list[list[float]]: """Return a symmetric cosine-similarity matrix; zero vectors score 0.""" if not vectors: return [] dimensions = len(vectors[0]) if any(len(vector) != dimensions for vector in vectors): raise ValueError("Semua embedding harus memiliki dimensi yang sama.") norms = [sqrt(sum(value * value for value in vector)) for vector in vectors] matrix: list[list[float]] = [] for i, left in enumerate(vectors): row = [] for j, right in enumerate(vectors): denominator = norms[i] * norms[j] similarity = sum(a * b for a, b in zip(left, right)) / denominator if denominator else 0.0 row.append(round(similarity, 6)) matrix.append(row) return matrix def principal_component_projection(vectors: list[list[float]]) -> list[list[float]]: """Project rows to two centered PCA coordinates for a small-sample chart.""" if not vectors: return [] values = np.asarray(vectors, dtype=np.float64) if values.ndim != 2: raise ValueError("Embedding harus berupa matriks dua dimensi.") centered = values - values.mean(axis=0, keepdims=True) if values.shape[0] == 1: return [[0.0, 0.0]] left, singular_values, _ = np.linalg.svd(centered, full_matrices=False) component_count = min(2, len(singular_values)) coordinates = left[:, :component_count] * singular_values[:component_count] if component_count == 1: coordinates = np.column_stack([coordinates, np.zeros(values.shape[0])]) return coordinates.tolist()