PrettyTimer is a simple yet useful tool for you to collect the running time of the code blocks.
If you want to evaluate the running time of your code blocks, for example, in a machine learning project, the program has several stages: data load, model forward, gradient backward, etc. Even you will statistic running time more than ten code blocks. The general way is to implement it like this:
import time
start_time_load_data = time.time()
time.sleep(3)
end_time_load_data = time.time()
print(end_time_load_data - start_time_load_data)
start_time_forward = time.time()
time.sleep(3)
end_time_forward = time.time()
print(end_time_forward - start_time_forward)
start_time_backward = time.time()
time.sleep(3)
end_time_backward = time.time()
print(end_time_backward - start_time_backward)
start_time_clloct = time.time()
time.sleep(3)
end_time_clloct = time.time()
print(end_time_clloct - start_time_clloct)
PrettyTimer will reproduce it efficiently and elegantly as follows:
import time
from prettytimer import PrettyTimer
timer = PrettyTimer()
timer.start('load_data')
time.sleep(3)
timer.end('load_data')
timer.start('forward')
time.sleep(3)
timer.end('forward')
timer.start('backward')
time.sleep(3)
timer.end('backward')
timer.start('collect')
time.sleep(3)
timer.end('collect')
timer.collect()
The collected information is a Markdown table style which can be copied and pasted to your Markdown document.
Moreover, Prettytimer provides an ETA (Estimated Time of Arrival) method:
timer.eta(NAME, ITER, TOTAL)Where NANE is a unique identifier to a timer, ITER is the current iteration and TOTAL is the total iteration of your training schedule.
That's all! Pretty concise, right?
Maybe there is a more graceful way to implement this, if yes, please tell me.
Install latest development version prettytable via pip:
pip install -U git+https://github.com/jazzband/prettytable
Install via pip:
pip install -U prettytimer
Install latest development version:
pip install -U git+git@github.com:kinredon/prettytimer.git