A nested progress with plotting options for fastai
Project description
fastprogress
A fast and simple progress bar for Jupyter Notebook and console. Created by Sylvain Gugger for fast.ai.
Copyright 2017 onwards, fast.ai. Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. A copy of the License is provided in the LICENSE file in this repository.
Install
To install simply use
pip install fastprogress
or:
conda install c fastai fastprogress
Note that this requires python 3.6 or later.
Usage
Example 1
Here is a simple example. Each bar takes an iterator as a main argument, and we can specify the second bar is nested with the first by adding the argument parent=mb
. We can then:
 add a comment in the first bar by changing the value of
mb.main_bar.comment
 add a comment in the first bar by changing the value of
mb.child.comment
 write a line between the two bars with
mb.write('message')
from fastprogress.fastprogress import master_bar, progress_bar
from time import sleep
mb = master_bar(range(10))
for i in mb:
for j in progress_bar(range(100), parent=mb):
sleep(0.01)
mb.child.comment = f'second bar stat'
mb.main_bar.comment = f'first bar stat'
mb.write(f'Finished loop {i}.')
#mb.update_graph(graphs, x_bounds, y_bounds)
Example 2
To add a graph that get plots as the training goes, just use the command mb.update_graphs
. It will create the figure on its first use. Arguments are:
graphs
: a list of graphs to be plotted (each of the form[x,y]
)x_bounds
: the min and max values of the x axis (ifNone
, it will those given by the graphs)y_bounds
: the min and max values of the y axis (ifNone
, it will those given by the graphs)
Note that it's best to specify x_bounds
and y_bounds
, otherwise the box will change as the loop progresses.
Additionally, we can give the label of each graph via the command mb.names
(should have as many elements as the graphs argument).
import numpy as np
mb = master_bar(range(10))
mb.names = ['cos', 'sin']
for i in mb:
for j in progress_bar(range(100), parent=mb):
if j%10 == 0:
k = 100 * i + j
x = np.arange(0, 2*k*np.pi/1000, 0.01)
y1, y2 = np.cos(x), np.sin(x)
graphs = [[x,y1], [x,y2]]
x_bounds = [0, 2*np.pi]
y_bounds = [1,1]
mb.update_graph(graphs, x_bounds, y_bounds)
mb.child.comment = f'second bar stat'
mb.main_bar.comment = f'first bar stat'
mb.write(f'Finished loop {i}.')
Here is the rendering in console:
If the script using this is executed with a redirect to a file, only the results of the .write
method will be printed in that file.
Example 3
Here is an example that a typical machine learning training loop can use. It also demonstrates how to set y_bounds
dynamically.
def plot_loss_update(epoch, epochs, mb, train_loss, valid_loss):
""" dynamically print the loss plot during the training/validation loop.
expects epoch to start from 1.
"""
x = range(1, epoch+1)
y = np.concatenate((train_loss, valid_loss))
graphs = [[x,train_loss], [x,valid_loss]]
x_margin = 0.2
y_margin = 0.05
x_bounds = [1x_margin, epochs+x_margin]
y_bounds = [np.min(y)y_margin, np.max(y)+y_margin]
mb.update_graph(graphs, x_bounds, y_bounds)
And here is an emulation of a training loop that uses this function:
from fastprogress.fastprogress import master_bar, progress_bar
from time import sleep
import numpy as np
import random
epochs = 5
mb = master_bar(range(1, epochs+1))
# optional: graph legend: if not set, the default is 'train'/'valid'
# mb.names = ['first', 'second']
train_loss, valid_loss = [], []
for epoch in mb:
# emulate train subloop
for batch in progress_bar(range(2), parent=mb): sleep(0.2)
train_loss.append(0.5  0.06 * epoch + random.uniform(0, 0.04))
# emulate validation subloop
for batch in progress_bar(range(2), parent=mb): sleep(0.2)
valid_loss.append(0.5  0.03 * epoch + random.uniform(0, 0.04))
plot_loss_update(epoch, epochs, mb, train_loss, valid_loss)
And the output:
Project details
Release history Release notifications  RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for fastprogress0.2.3py3noneany.whl
Algorithm  Hash digest  

SHA256  8b4d7a6af31bafbe1f17a8e5c29befe514a6d84a920c2d215cb5fac016c8e661 

MD5  73de7ec7013d4e1920ba7ec17f4f7c45 

BLAKE2b256  a3daffd8fe0daf7e679804a32a1e8654ac2988e2ef85937fc1d223e98eee736e 