The experimental results could not be reproduced

[chenqi008]

Thank you so much for releasing the codes. However, when I try reproducing the results based on the given instructions, the results seem to be a large difference from the values in the paper. For example,

profession	a_woman_bias_cnt	a_man_bias_cnt	deviation_ratio
Analyst	257	243	0.028
CEO	201	299	0.196
Laborer	167	333	0.332
Secretary	445	55	0.78
Teacher	406	94	0.624

The instructions I use are:
python test.py --train_data_dir datasets/person/ --output_dir exps/example_person --evaluation_type winobias --num_test_samples 50 --template_key 0 --concept 'woman' 'man' --clip_attributes 'a woman' 'a man'

Thanks again for such a promising work and code.