import pandas as pd import numpy as np import matplotlib.pyplot as plt from sklearn.neighbors import LocalOutlierFactor from sklearn.preprocessing import StandardScaler from sklearn.model_selection import train_test_split from sklearn.metrics import r2_score, mean_absolute_error, mean_squared_error from scipy.interpolate import RBFInterpolator import warnings from tqdm import tqdm warnings.filterwarnings('ignore') # ================= Font and Font Size Control Center ================= FONT_NAME = 'Calibri' # Global font AXES_LABEL_SIZE = 16 # X/Y axis label font size TICK_LABEL_SIZE = 14 # Tick number font size LEGEND_FONT_SIZE = 14 # Legend font size # ================= Global Settings for Academic-level Charts (Top Conference PDF) ================= plt.rcParams['font.family'] = 'sans-serif' plt.rcParams['font.sans-serif'] = [FONT_NAME] plt.rcParams['mathtext.fontset'] = 'custom' plt.rcParams['mathtext.rm'] = FONT_NAME plt.rcParams['mathtext.it'] = FONT_NAME plt.rcParams['mathtext.bf'] = FONT_NAME plt.rcParams['axes.unicode_minus'] = False plt.rcParams['xtick.labelsize'] = TICK_LABEL_SIZE plt.rcParams['ytick.labelsize'] = TICK_LABEL_SIZE plt.rcParams['xtick.direction'] = 'in' plt.rcParams['ytick.direction'] = 'in' plt.rcParams['pdf.fonttype'] = 42 plt.rcParams['ps.fonttype'] = 42 def get_blind_test_scores(excel_filename, magnet_label, n_iterations=1000): """Process a single dataset and return accuracy score list and average error metrics""" print(f"\n{'=' * 50}") print(f"Calculating for [{magnet_label}] dataset ({n_iterations} random blind tests)...") try: df = pd.read_excel(excel_filename) df = df.dropna(subset=['X(mm)', 'Y(mm)', 'Z(mm)', 'Fx', 'Fy', 'Fz']) except FileNotFoundError: print(f"Error: File '{excel_filename}' not found.") return None, None, None, None X_raw = df[['X(mm)', 'Y(mm)', 'Z(mm)']].values Y_raw = df[['Fx', 'Fy', 'Fz']].values # Clean outliers data_scaled = StandardScaler().fit_transform(np.column_stack((X_raw, Y_raw))) lof = LocalOutlierFactor(n_neighbors=10, contamination=0.01) good_mask = lof.fit_predict(data_scaled) == 1 X_clean = X_raw[good_mask] Y_clean = Y_raw[good_mask] overall_scores_per_iter = [] mae_per_iter = [] rmse_per_iter = [] # True random fluctuations for _ in tqdm(range(n_iterations), desc=f"[{magnet_label}] Progress"): X_train, X_test, Y_train, Y_test = train_test_split(X_clean, Y_clean, test_size=0.03) # Spatial compression Y_train_transformed = np.arcsinh(Y_train) # TPS physics engine rbf = RBFInterpolator(X_train, Y_train_transformed, kernel='thin_plate_spline', smoothing=0.0005) # Predict and decompress Y_pred_transformed = rbf(X_test) Y_pred = np.sinh(Y_pred_transformed) # ---------------------------------------------------- # [Key Modification]: Physical Defense (Range Clipping and Negative Score Zeroing) # Update clipping boundary from original [-50.0, 50.0] to [-25.0, 25.0] # ---------------------------------------------------- # Physical Defense (Axis-wise range clipping and negative score zeroing) Y_pred[:, 0] = np.clip(Y_pred[:, 0], -10.0, 10.0) # Clip Fx Y_pred[:, 1] = np.clip(Y_pred[:, 1], -10.0, 10.0) # Clip Fy Y_pred[:, 2] = np.clip(Y_pred[:, 2], 0.0, 25.0) # Clip Fz # 1. Calculate fitting score iter_score = r2_score(Y_test, Y_pred, multioutput='raw_values') iter_score = np.maximum(0, iter_score) iter_overall_acc = np.mean(iter_score) * 100 overall_scores_per_iter.append(iter_overall_acc) # 2. Calculate MAE and RMSE for current iteration iter_mae = mean_absolute_error(Y_test, Y_pred) iter_rmse = np.sqrt(mean_squared_error(Y_test, Y_pred)) mae_per_iter.append(iter_mae) rmse_per_iter.append(iter_rmse) final_median = np.median(overall_scores_per_iter) # Calculate average MAE and RMSE over 1000 iterations avg_mae = np.mean(mae_per_iter) avg_rmse = np.mean(rmse_per_iter) return overall_scores_per_iter, final_median, avg_mae, avg_rmse # ================= Start Calculation and Independent Plotting ================= if __name__ == "__main__": datasets = [ ('sensor_data_3mm.xlsx', '3mm'), ('sensor_data_2.5mm.xlsx', '2.5mm') ] print("\nRendering single academic chart (PDF vector version) and calculating physical errors, please wait...") x_axis = np.arange(1, 1001) for file_name, label in datasets: scores, median, avg_mae, avg_rmse = get_blind_test_scores(file_name, label, n_iterations=1000) if scores is not None: # ---> Print core error data to terminal <--- print(f"\n[{label}] Dataset 1000 Blind Test Results Summary:") print(f" => Median Score: {median:.2f}%") print(f" => Average MAE: {avg_mae:.4f} N") print(f" => Average RMSE: {avg_rmse:.4f} N") print("-" * 50) # Create independent canvas for each dataset (6x5 size) fig, ax = plt.subplots(figsize=(6, 5)) # Plot scatter and line ax.plot(x_axis, scores, color='#1f77b4', alpha=0.3, linewidth=1, label='Single Iteration') ax.scatter(x_axis, scores, color='#1f77b4', alpha=0.2, s=5) # Plot 50-iteration moving average line moving_avg = pd.Series(scores).rolling(window=50, min_periods=1).mean() ax.plot(x_axis, moving_avg, color='#ff7f0e', linewidth=2, alpha=0.9, label='Moving Average (w=50)') # Plot red median crosshair line ax.axhline(y=median, color='#d62728', linestyle='--', linewidth=2, label=f'Median Score: {median:.2f}%') # Axis label decoration ax.set_xlabel('Blind Test Iteration', fontsize=AXES_LABEL_SIZE) ax.set_ylabel('Predictive Performance Score (%)', fontsize=AXES_LABEL_SIZE) ax.set_ylim(-5, 105) ax.set_xlim(0, 1010) ax.grid(True, linestyle=':', alpha=0.6) # Borderless legend ax.legend(loc='lower right', fontsize=LEGEND_FONT_SIZE, frameon=False) plt.tight_layout() output_filename = f'score_distribution_{label}_random.pdf' plt.savefig(output_filename, format='pdf', bbox_inches='tight') plt.close() print(f"[{label}] Independent PDF chart generated: {output_filename}")