plot.py

import os
import argparse
import re

import numpy as np
from matplotlib import pyplot as plt
import seaborn as sns


def load_random(fold):
    rsvs = ['1.0', '0.95', '0.9', '0.85', '0.8', '0.75', '0.7', '0.65', '0.6', '0.55', '0.5', '0.45', '0.4', '0.35', '0.3', '0.25', '0.2']#, '0.15', '0.1', '0.09', '0.08', '0.07', '0.0625', '0.06', '0.0575', '0.055', '0.0525', '0.05', '0.0475', '0.045', '0.0425', '0.04', '0.0375', '0.035', '0.0325', '0.03']
    accs = []
    for para in rsvs:
        fn = os.path.join(fold, para+".out")
        with open(fn, 'r') as ips:
            for line in ips:
                mobj = re.findall("\d+\.\d+", line)
                if len(mobj) == 2:
                    acc = float(mobj[0])
            accs.append(acc)
    return np.asarray([float(elem) for elem in rsvs]), np.asarray(accs)


def load_al(fold):
    rsvs = []
    accs = []
    for fn in os.listdir(fold):
        if fn.endswith(".out"):
            params = re.findall("\d+\.\d+", fn)
            rsv, alpha = float(params[0]), float(params[1])
            if rsv == alpha:
                rsvs.append(rsv)
                with open(os.path.join(fold, fn), 'r') as ips:
                    for line in ips:
                        mobj = re.findall("\d+\.\d+", line)
                        if len(mobj) == 2:
                            acc = float(mobj[0])
                    accs.append(acc)
    results = sorted([(x, y) for x, y in zip(rsvs, accs)], key=lambda tp:tp[0])
    xs = [tp[0] for tp in results]
    ys = [tp[1] for tp in results]
    return np.asarray(xs), np.asarray(ys)


def main():
    parser = argparse.ArgumentParser(description='Plot Results')
    parser.add_argument('--dataset', type=str, default='products')
    parser.add_argument('--folds', type=str, default='logs')
    args = parser.parse_args()
    print(args)

    if args.dataset == 'products':
        V = 235938
    elif args.dataset == 'papers100M':
        V = 1546782
    elif args.dataset == 'mag240m':
        V = 0
    else:
        raise ValueError('{}'.format(args.dataset))

    xs, ys, lgs = [], [], []
    for dirname in args.folds.split(','):
        #for lg in ['Random', 'EL2N', 'Mem', 'Infl-max', 'DDD', 'AGE']:#'Infl-sum-abs', 'DDD', 'AGE']:
        #for lg in ['Random', 'EL2N', 'Mem', 'Infl-max', 'AGE', 'Dist-Greedy']:
        for lg in ['Random', 'Infl-max', 'Ours']:
            fn = lg.lower()
            #if fn == 'random':
            #    tp = load_random(os.path.join(args.fold, fn))
            #else:
            #    tp = load_al(os.path.join(args.fold, fn))
            tp = load_al(os.path.join("on_{}".format(args.dataset), os.path.join(dirname, fn)))
            if lg in lgs:
                idx = lgs.index(lg)
                assert len(ys[idx]) == len(tp[1]), "inconsistent len {}".format(len(tp[1]))
                ys[idx] += tp[1]
            else:
                lgs.append(lg)
                xs.append(tp[0])
                ys.append(tp[1])

    num_folds = len(args.folds.split(','))
    min_error_rate = 100
    method_id = -1
    ratio = -1
    for i in range(len(xs)):
        xs[i] = V * xs[i]
        ys[i] = 100.0 - (ys[i] / num_folds)
        idx = np.argmin(ys[i])
        if ys[i][idx] < min_error_rate and i != 0:
            min_error_rate = ys[i][idx]
            method_id = i
            ratio = xs[i][idx]
    print(lgs[method_id], ratio, min_error_rate)

    # plt.axvline(np.sqrt(N)/2)
    with sns.color_palette('viridis_r', len(xs)):
        plt.figure(figsize=(6, 5))
        for i in range(len(xs)):
            plt.scatter(xs[i], ys[i], label=lgs[i])
            plt.plot(xs[i], ys[i])
    plt.xscale('log')
    plt.yscale('log')
    plt.ylabel('Error')
    #plt.xlabel(r'$\alpha$')
    plt.xlabel('Training examples')
    if args.dataset == 'products':
        plt.ylim(0.0, 23.5)
    plt.legend();
    #plt.xticks([1,2,3],[1,2,3])
    #plt.yticks([2,10,20,50],[2,10,20,50])
    #plt.ylim([2,50])
    plt.grid(True,which='both',alpha=0.2)
    sns.despine()
    plt.savefig("cmp.pdf", transparent='True')


if __name__ == "__main__":
    main()