vissl_plot_losses_log.txt

gistfile1.txt

from classy_vision.generic.util import (
    copy_model_to_gpu,
    load_and_broadcast_checkpoint,
    load_checkpoint,
)
import torch
from fvcore.common.file_io import PathManager

# Train loss
import matplotlib.pyplot as plt
import os
import numpy as np
import json
from plotly import __version__
from plotly.offline import download_plotlyjs, init_notebook_mode, plot, iplot
init_notebook_mode(connected=True)
import plotly.tools as tls

def get_lr_values(log_file):
    with PathManager.open(log_file, 'r') as fopen:
        iters, lr_values = [], []
        for line in fopen:
            try:
                if 'lr:' in line and 'eta:' in line:
                    start = line.find('iter:') + 6
                    end = start + line[start:].find(';')
                    iter_num = int(line[start:end])
                    
                    start = line.find('lr:') + 4
                    end = start + line[start:].find(';')
                    curr_lr = float(line[start:end])

                    iters.append(iter_num)
                    lr_values.append(curr_lr)
            except Exception:
                pass
    iters, lr_values = zip(*sorted(zip(iters, lr_values)))
    return iters, lr_values

def get_loss_iters(log_file, normalize, multiplier=1.0):
    with PathManager.open(log_file, 'r') as fopen:
        loss, iters = [], []
        for line in fopen:
            try:
                if 'eta:' in line:
                    start = line.find('loss:') + 6
                    end = start + line[start:].find(';')
                    loss_amount = float(line[start:end])

                    start = line.find('iter:') + 6
                    end = start + line[start:].find(';')
                    iter_num = int(line[start:end])

                    loss.append(loss_amount * multiplier)
                    iters.append(iter_num)
            except Exception:
                pass

        if normalize:
            N = 17
            rem = int(len(iters) % N)
            if rem == 0:
                rem = int(len(iters))
            loss = list(np.mean(np.array(loss[:-rem]).reshape(-1, N), axis=1))
            iters = list(np.mean(np.array(iters[:-rem]).reshape(-1, N), axis=1))
        iters, loss = zip(*sorted(zip(iters, loss)))
    return loss, iters


def get_matlab_colors():
    matlabColors = [
        tuple([0, 0.4470, 0.7410]),
        tuple([0.8500, 0.3250, 0.0980]),
        tuple([0.9290, 0.6940, 0.1250]),
        tuple([0.4940, 0.1840, 0.5560]),
        tuple([0.4660, 0.6740, 0.1880]),
        tuple([0.3010, 0.7450, 0.9330]),
        tuple([0.6350, 0.0780, 0.1840]),
        tuple([0.300,  0.300, 0.300]),
        tuple([0.600, 0.600, 0.600]),
        tuple([1.000, 0.000, 0.000]),
        tuple([1.000, 0.500, 0.000]),
        tuple([0.749, 0.749, 0.000]),
        tuple([0.000, 1.000, 0.000]),
        tuple([0.000, 0.000, 1.000]),
        tuple([0.667, 0.000, 1.000]),
        tuple([0.333, 0.333, 0.000]),
        tuple([0.333, 0.667, 0.000]),
    ];
    return matlabColors;


def prepare_legend(file_list, legend_list, normalize, multipliers):
    assert len(file_list) == len(legend_list)
    legends = []
    all_colors = get_matlab_colors()
    for idx in range(len(file_list)):
        file_path = os.path.join(file_list[idx])
        print(file_path)

        loss, iters = get_loss_iters(file_path, normalize, multipliers[idx])
        leg_name = legend_list[idx]
        leg, = plt.plot(iters, loss, linewidth=20, label=leg_name, color=all_colors[idx])
        legends.append(leg)
        
    return legends


FILE_DICT = {
    '1': '/absolute/path/to/log.txt',
    '2': '/absolute/path/to/log.txt',
}


legend_dict = {
    '1': 'loss_curve_1',
    '2': 'loss_curve_2',
}