#Pearson correlation coefficient import argparse import numpy as np import json import pandas as pd import scipy.stats import os import math def compute_correlation(set1, set2): corr_coefficient = np.corrcoef(set1, set2)[0, 1] return corr_coefficient def compute_max_correlation(fix, var): #(350,3) set1 = fix[:,0].flatten() set2 = var[:,0].flatten() max = np.corrcoef(set1,set2)[0, 1] selected_human_score = set2 selected_human_index = np.zeros(num_images) for i in range(num_images): for j in range(3): set2[i] = var[i,j] corr_coefficient = np.corrcoef(set1, set2)[0, 1] if corr_coefficient > max: selected_human_score[i] = var[i,j] selected_human_index[i] = j max = corr_coefficient return max,selected_human_index,selected_human_score # Normalizing function using Min-Max scaling def min_max_normalize(scores): min_score = np.min(scores) max_score = np.max(scores) return (scores - min_score) / (max_score - min_score) def scale_dict_values(input_dict): # 获取字典中所有值 values = list(input_dict.values()) # 找到最小值和最大值 min_value = min(values) max_value = max(values) # 定义缩放函数 def scale_value(value): return 1 + (value - min_value) * (5 - 1) / (max_value - min_value) # 创建一个新的字典，其中所有值都缩放到1到5之间 scaled_dict = {key: scale_value(value) for key, value in input_dict.items()} return scaled_dict def parse_args(): parser = argparse.ArgumentParser(description="Demo") parser.add_argument( "--auto_evaluation", type=str, help="csv", # default="/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_5_2/frame_score.csv" # TODO pkc # default="/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_9_2/frame_score.csv" # TODO # default="/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_2_2/frame_score_mechanism_3_1.csv" # TODO # default="/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_2_2/frame_score_mechanism.csv" # TODO pick # default="/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_6_2_gt/grid_score.csv" # TODO pick # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_6_2_gt/frame_score.csv" # default="/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/pllava/csv_output_6_2/video_score.csv" # # default="/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_5_2/grid_score.csv" # TODO pick # default="/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/clip/csv_output_6_2_gt/clip.csv" # TODO # default ="/group/xihuiliu/sky/T2V-Compbench/metric/models/GroundingDINO/csv_output_9_2_user_study/thresh_04_2_score.csv" # default ="/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/b_clip/csv_output_2_2/b_clip.csv" # default ="/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/clip/csv_output_2_2/clip.csv" # default ="/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/pllava/csv_output_2_2/video_score.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_1_2/grid_score_again.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_1_2/grid_score_6_again.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_1_2/grid_score_4_adjust_again.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_1_2/grid_score_8.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_1_2/frame_score_8.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/BLIP-vqa/csv_output_1_2_blipvqa.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/pllava/csv_output_5_2/video_score.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_2_2/frame_score_mechanism_test.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_2_2/grid_score.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/models/GroundingDINO/csv_output_3_2_user_study_2d/user_study_2d_03_combine_frame.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/models/GroundingDINO/csv_output_3_2_user_study_2d/user_study_2d_03_321split_combine_frame.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/models/GroundingDINO/csv_output_3_2_user_study_2d/user_study_2d_03_321split_algorithm_035_combine_frame.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/models/GroundingDINO/csv_output_3_2_user_study_2d/user_study_2d_03_321split_algorithm_035_2_combine_frame.csv", # default = "/group/xihuiliu/sky/T2V-Compbench/metric/models/Grounded-Segment-Anything/csv_3_depth_user_study/algorithm_035_combine_frame.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/models/Grounded-Segment-Anything/csv_3_depth_user_study/algorithm_035_addscore_combine_frame.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/clip/csv_output_3_2/clip.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/b_clip/csv_output_6_2/b_clip.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/models/dot/csv_4_2_user_study/4_2_user_study_score.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/clip/csv_output_4_2/clip.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/b_clip/csv_output_4_2/b_clip.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_4_2/frame_score.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_4_2/grid_score.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_1_2/frame_score_corrected.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/models/GroundingDINO/csv_output_3_2_user_study_2d/user_study_2d_03_321split_algorithm_035_productscore_combine_frame.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/models/GroundingDINO/csv_output_3_2_user_study_2d/user_study_2d_03_321split_algorithm_035_2_combine_frame.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/models/Grounded-Segment-Anything/csv_3_depth_user_study/algorithm_035_addscore_combine_frame.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/groundingDINO_3/spatial_final_2.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/models/GroundingDINO/csv_output_3_2_user_study_2d/user_study_2d_03_321split_algorithm_035_corrected_combine_frame.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/models/GroundingDINO/csv_output_3_2_user_study_2d/user_study_2d_03_321split_algorithm_035_corrected_combine_frame.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/models/GroundingDINO/csv_output_3_2_user_study_2d/user_study_2d_03_321split_algorithm_035_corrected_2_combine_frame.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/models/Grounded-Segment-Anything/csv_3_depth_user_study/algorithm_035_corrected_combine_frame.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/b_clip/csv_output_3_2/b_clip.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_3_2/frame_score.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/pllava/csv_output_3_2/video_score.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/pllava/csv_output_4_2/video_score.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_1_2/grid_score_8.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_1_2/frame_score_corrected_8_withinitial.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/pllava/csv_output_1_2/video_score.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/BLIP-vqa/csv_output_1_2_blipvqa.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/b_clip/csv_output_1_2/b_clip.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/clip/csv_output_6_2_gt/clip.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/groundingDINO_3/spatial_method1.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/groundingDINO_3/spatial_method2.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_5_2/frame_score.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_5_2/grid_score_old.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/b_clip/csv_output_5_2/b_clip.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/clip/csv_output_5_2/clip.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/pllava/csv_output_5_2/video_score.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava/csv_output_5_2/frame_score_new.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/viclip/consistent_attr_rebuttal.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/viclip/dynamic_attr_rebuttal.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/viclip/spatial_relationship_rebuttal.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/viclip/motion_binding_rebuttal.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/viclip/action_rebuttal.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/viclip/interaction_rebuttal.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/viclip/numeracy_rebuttal.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/rebuttal_models/EvalCrafter/blipbleu_rebuttal/consistent_attr_rebuttal.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/rebuttal_models/EvalCrafter/blipbleu_rebuttal/dynamic_attr_rebuttal.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/rebuttal_models/EvalCrafter/blipbleu_rebuttal/spatial_relationship_rebuttal.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/rebuttal_models/EvalCrafter/blipbleu_rebuttal/motion_binding_rebuttal.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/rebuttal_models/EvalCrafter/blipbleu_rebuttal/action_rebuttal.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/rebuttal_models/EvalCrafter/blipbleu_rebuttal/interaction_rebuttal.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/rebuttal_models/EvalCrafter/blipbleu_rebuttal/numeracy_rebuttal.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/rebuttal_models/VPEval/csv_rebuttal/vpeval_user_study_video.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/models/GroundingDINO/rebuttal_videodirectorgpt_csv/M-GDino_user_study_rebuttal.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/metric/models/LLaVA/cvpr_csv/csv_6/user_study/user-t0_object_interactions_score_0.0.csv", # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava_cvpr/csv_output_1_2/frame_score.csv", # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava_cvpr/csv_output_5_2/frame_score.csv", # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava_cvpr/csv_output_6_2_gt/frame_score.csv", # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava_cvpr/csv_output_1_2/frame_score_1st_last_2.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/auto_eval/llava_cvpr/csv_output_3_2/frame_score.csv", # default = "/group/xihuiliu/sky/T2V-Compbench/metric/models/LLaVA/cvpr_csv/csv_1/user_study/cvpr_conv/user-t005_consistent_attr_score_0.05_4.csv", ) parser.add_argument( "--human_evaluation", type=str, help="csv", # default="/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/action_grid_corrected.csv" # pick!!! # default="/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/dynamic_attr.csv" # default="/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/interaction_llava_grid.csv" # default="/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/interaction_llava_grid_corrected_gt_filter2.csv" #pick!!! # default="/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/dynamic_attr_mechanism_corrected.csv" #pick!!! # default="/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/interaction_gt.csv" # default="/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/consistent spatial_human.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/consistent spatial_human_corrected2.csv" # default="/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/numeracy_human.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/numeracy_human_adjust.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/dynamic_spatial_human_1.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/consistent attribute_human.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/consistent attribute_human_2.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/consistent attribute_human_corrected.csv" default = "/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/consistent attribute_human_corrected_kaiyue.csv", #ok final # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/dynamic_attr_mechanism_corrected_kaiyue.csv" #ok final # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/consistent spatial_human_corrected2_kaiyue.csv" #ok final # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/dynamic_spatial_human_1_kaiyue.csv" #ok final # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/action_adjust_kaiyue_arxiv.csv" #ok final # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/interaction_llava_grid_corrected_gt_filter2_kaiyue.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/numeracy_human_adjust_corrected_kaiyue.csv" # default = "/group/xihuiliu/sky/T2V-Compbench/user_study/human_eval/interaction_adjust_kaiyue.csv", #ok final ) args = parser.parse_args() return args if __name__ == "__main__": args = parse_args() auto_evaluation = args.auto_evaluation method = ["videocrafter2", "latte", "opensora11", "opensoraplan", "show1", "modelscope", "gt"] dic_human = {} # store by model name, for calculating avg score for each model dic_auto = {}# store by model name dic_score_category = {} dic_human_avg = {} # store by video name, for calculating human corr with avg score dic_auto_avg = {} # store by video name # hky # get AMT score # Define the path to your CSV file csv_file = args.human_evaluation # Read the CSV file into a pandas DataFrame data_frame = pd.read_csv(csv_file) try: column_index = data_frame.columns.get_loc('Input.video') except: column_index = data_frame.columns.get_loc('Input.url') column_name = data_frame.iloc[:,column_index] column_name = column_name.to_numpy() try: score_index = data_frame.columns.get_loc('Answer.score') except: score_index = data_frame.columns.get_loc('Answer.correctness_answer') score_AMT = data_frame.iloc[:,score_index] score_AMT = score_AMT.to_numpy() score_dic = {} # to store {method: {id: score,... },... } # read from auto eval csv_file_auto = args.auto_evaluation data_frame_auto = pd.read_csv(csv_file_auto) # read key = id, value = Score from csv for i in range(len(data_frame_auto)): # print(data_frame_auto.iloc[i]) # score_dic[(data_frame_auto.iloc[i]['name'].split(".")[0])] = float(data_frame_auto.iloc[i][-1]) score_dic[(data_frame_auto.iloc[i]['name'].split(".")[0])] = float(data_frame_auto.iloc[i]['Score']) # score_dic[(data_frame_auto[0].split(".")[0])] = float(data_frame_auto.iloc[i]['Score']) # score_dic[(data_frame_auto.iloc[i]['name'])] = float(data_frame_auto.iloc[i]['Score']) # score_dic[(data_frame_auto.iloc[i]['id'])] = float(data_frame_auto.iloc[i]['Score']) # # read from human eval's csv (col. "Input.video"), store {method: [score,...],...} for me in range(len(method)): # initialize dic_human[method[me]] = [] dic_auto[method[me]] = [] dic_score_category[method[me]] = [] print(score_dic) record = np.empty(0) score_human = np.empty(0) # for calculating human corr with all scores score_norm = np.empty(0) score_human_avg = np.empty(0) #f or calculating human corr with avg score score_norm_avg = np.empty(0) for i in range(len(column_name)): # initialize c = column_name[i] c = c.replace(".mp4","") dic_human_avg[c] = [] dic_auto_avg[c] = [] # import pdb; pdb.set_trace() # score_dic = scale_dict_values(score_dic) # TODO rescale to 1~5 for i in range(len(column_name)): c = column_name[i] c = c.replace(".mp4","") s = score_AMT[i] method_name = c.split("_")[-1] # hky index_name = method.index(method_name) # print(method_name, index_num) # score_norm = np.append(score_norm, score_dic[method_name][index_num]) # print(score_dic) # print(c) # print(score_dic[c]) try: score_norm = np.append(score_norm, score_dic[c]) #TODO round or not # import pdb; pdb.set_trace() # auto_score = round(score_dic[c]) # score_norm = np.append(score_norm, auto_score) score_human = np.append(score_human, (s/5)) record = np.append(record, c) dic_human[method_name].append(s/5) # length 150 dic_auto[method_name].append(score_dic[c]) # dic_score_category[method_name].append(score_dic[c]) dic_human_avg[c].append(s/5) dic_auto_avg[c].append(score_dic[c]) #TODO round or not # dic_auto_avg[c].append(auto_score) except: # not for gt continue # import pdb; pdb.set_trace() for k, v in dic_human_avg.items(): score_human_avg = np.append(score_human_avg, np.average(dic_human_avg[k])) score_norm_avg = np.append(score_norm_avg, np.average(dic_auto_avg[k])) Pearson_correlation_coefficient_all = compute_correlation(score_human,score_norm) #TODO change score_human, score_norm print("Pearson correlation coefficient all:",f'{Pearson_correlation_coefficient_all: .4}') score_diff = np.empty(0) # Normalize the scores to the range [0, 1], and pick the larger different score score_norm_norm = min_max_normalize(score_norm) score_human_norm = min_max_normalize(score_human) # Calculate the threshold threshold = np.mean(np.abs(score_human_norm - score_norm_norm)) # Iterate through the scores and apply the condition for i in range(len(score_norm)): score_diff = np.append(score_diff, score_human_norm[i] - score_norm_norm[i]) if abs(score_norm_norm[i] - score_human_norm[i]) > threshold * 2: score_diff = np.append(score_diff, [score_norm[i], score_human[i]]) print("Not aligned:", i, "score ori:", score_norm[i], "score from AMT", score_human[i]) print("##############################") # score_diff = np.reshape(score_diff,(num_images,3)) # best_index = np.argmin(score_diff, axis=1) # #score_diff_best = np.choose(best_index,score_diff.T) # score_norm_reshape = np.reshape(score_norm,(num_images,3)) # score_norm_best = np.choose(best_index,score_norm_reshape.T) # score_human_reshape = np.reshape(score_human,(num_images,3)) # score_human_best = np.choose(best_index,score_human_reshape.T) * 5 #best human score # Pearson_correlation_coefficient_abs_diff = compute_correlation(score_human_best,score_norm_best) # print("Pearson correlation coefficient with min abs diff:",f'{Pearson_correlation_coefficient_abs_diff: .4}') # Pearson_correlation_coefficient_search_max,selected_human_index,selected_human_score = compute_max_correlation(score_norm_reshape, score_human_reshape) #(350,3) # print("Pearson correlation coefficient with search max:",f'{Pearson_correlation_coefficient_search_max: .4}') # score_norm_average = np.average(score_norm_reshape, axis=1) # score_human_average = np.average(score_human_reshape, axis=1) Pearson_correlation_coefficient_all = compute_correlation(score_norm, score_human) print("Pearson correlation coefficient all:",f'{Pearson_correlation_coefficient_all: .4}') Pearson_correlation_coefficient_avg = compute_correlation(score_norm_avg, score_human_avg) print("Pearson correlation coefficient avg:",f'{Pearson_correlation_coefficient_avg: .4}') kendalltau_all,p_value_all = scipy.stats.kendalltau(score_human, score_norm) print("kendall's Tau_all: ", f'{kendalltau_all: .4}', " p_value_all: ", f'{p_value_all: .4}') # kendalltau_abs_diff,p_value_abs_diff = scipy.stats.kendalltau(score_human_best,score_norm_best) # print("kendall's Tau with abs diff: ", f'{kendalltau_abs_diff: .4}', " p_value: ", f'{p_value_abs_diff: .4}') #kendalltau_search_max = scipy.stats.kendalltau(x, y) kendalltau_avg, p_value_avg = scipy.stats.kendalltau(score_human_avg, score_norm_avg) print("kendall's Tau avg: ", f'{kendalltau_avg: .4}', " p_value: ", f'{p_value_avg: .4}') spearmanr_all = scipy.stats.spearmanr(score_human,score_norm) print("spearmanr's rho all", f'{spearmanr_all.statistic: .4}', f'{spearmanr_all.pvalue: .4}') # spearmanr_abs_diff = scipy.stats.spearmanr(score_human_best, score_norm_best) # print("spearmanr's rho with abs diff", f'{spearmanr_abs_diff.statistic: .4}', f'{spearmanr_abs_diff.pvalue: .4}') spearmanr_avg = scipy.stats.spearmanr(score_human_avg, score_norm_avg) print("spearmanr's rho avg", f'{spearmanr_avg.statistic: .4}', f'{spearmanr_avg.pvalue: .4}') #auto avg per model print("##############################") # import pdb; pdb.set_trace() for i in range(len(method)): auto_avg = np.average(dic_auto[method[i]]) human_avg = np.average(dic_human[method[i]]) print(method[i], " auto_avg: ", f'{auto_avg: .4}', " human_avg:", f'{human_avg: .4}'," human eval length / method", len(dic_human[method[i]])) # for i in range (len(method)): # #corr within per model # mPearson_all = compute_correlation(dic_human[method[i]], dic_score_category[method[i]]) # mkendalltau_all, mp_value_all = scipy.stats.kendalltau(dic_human[method[i]], dic_score_category[method[i]]) # mspearmanr_all = scipy.stats.spearmanr(dic_human[method[i]], dic_score_category[method[i]]) # m_human_reshape = np.reshape(dic_human[method[i]], (50, 3)) # m_auto_reshape = np.reshape(dic_score_category[method[i]], (50, 3)) # m_human_average = np.average(m_human_reshape, axis=1) # m_auto_average = np.average(m_auto_reshape, axis=1) # mPearson_average = compute_correlation(m_human_average, m_auto_average) # mkendalltau_avg, mp_value_avg = scipy.stats.kendalltau(m_human_average, m_auto_average) # mspearmanr_avg = scipy.stats.spearmanr(m_human_average, m_auto_average) # print(method[i]) # print(" Pearson all:", f'{mPearson_all: .4}', " Pearson avg", f'{mPearson_average: .4}') # print(" kendall all:", f'{mkendalltau_all: .4}', " p value", f'{mp_value_all: .4}',end=" ") # print(" spearman all:", f'{mspearmanr_all.statistic: .4}', f'{mspearmanr_all.pvalue: .4}') # print(" kendall avg:", f'{mkendalltau_avg: .4}', " p value", f'{mp_value_avg: .4}',end=" ") # print(" spearman avg:", f'{mspearmanr_avg.statistic: .4}',f'{mspearmanr_avg.pvalue: .4}')