I installed Python + Chainer on Windows to start deep learning, which is popular now, and when I ran the sample mnist triumphantly, for some reason it ended with an error and camphor for those who tend to lose their temper. It is an agent. Although the detailed cause has not been clarified, mnist can be executed without any error for the time being.
When mninst is executed, an error always occurs after epoch ends (when it reaches 100%). The following error message is thrown.
FileExistsError: [WinError 183]You cannot create a file that already exists.: '<PATH_TO_CHAINER_INSTALL_DIRECTORY>\chainer\examples\mnist\result\logofw0i659' -> 'result\log'
An error occurred because I tried to rename another file to "log" even though a file named "log" existed.
Do not rename above. (Is it originally wanted to overwrite and save?)
See the error message on the line above it.
File "C:\Anaconda3\lib\site-packages\chainer\training\extensions\log_report.py", line 94, in __call__
os.rename(path, os.path.join(trainer.out, log_name))
Comment out the number of lines (line 94) in the file (log_report.py).
log_report.py
#Comment out the following line.
#os.rename(path, os.path.join(trainer.out, log_name))
With the above, the error will not occur. Instead, a large number of log files are generated in the log directory and should be cleaned accordingly. Please be careful.
It is the execution time measurement result with the default parameter without gpu of mnist. As a reference value.
Execution time: less than 20 minutes (18 minutes) (This, especially the execution time is not displayed on the screen ...)
Recommended Posts