Function main

utils/plot-mini.py:195–291

(args)

    return avg / len(data.index)

def main(args):
    os.makedirs(args.out_dir, exist_ok=True)

    # setup plots
    # sns.set_style('white')
    sns.set_style("ticks")
    # sns.set()
    sns.set(font_scale=1.7) 

    #list of all datasets wanted
    datasets = [
                'resisc', 'ucmerced', 'viper', 'bdd', 
                'domain_net_painting', 'domain_net_clipart', 'domain_net_infograph', 'domain_net_sketch', 
                'domain_net_quickdraw', 'domain_net_real', 'flowers', 'chest_xray_kids', 
                'chexpert', 'xview', 'coco_2014', 'pascal',  
                ]

    #

    #get files for all datasets wanted
    result_files = []
    for dataset in datasets:
        result_files.append(os.path.join(args.results_dir, dataset + "_results.json"))

    print(result_files)

    #creates a datasets list to concatenate all pd to form one large pandas df
    datasets_pd = []

    #createsd dictionary for moco transfer result and bn result for each dataset
    moco_transfers = {}
    data_bn_points = {}

    for resfile in result_files:
        with open(resfile, 'r') as infile:
            raw_data = json.load(infile)
        #get data into pandas dataframe
        data = pd.DataFrame(raw_data.values())

        #takes only linear evals
        data = data[data.result_type =='linear-eval']

        #ignores all imagenet basetrain models
        data = data[data.basetrain != "imagenet_r50_supervised"]

        #renames the basetrain
        data.basetrain = data.basetrain.replace("moco_v2_800ep", "HPT")
        data.basetrain = data.basetrain.replace("no", "MoCo Random Init")
        data.basetrain = data.basetrain.replace("none", "MoCo Random Init")


        #ignores all imagenet basetrain models
        # data = data[~((data.pretrain_iters == "5000") & (data.basetrain == "MoCo Random Init"))]

        #gets all bn data 
        data_bn = data[data.pretrain_iters.str.contains("bn")]

        #all non-bn data