Nim-Cifar10
A test place for cifar-10 classification using Arraymancer.
How to use
bash cifar10_downloader.sh # This downloads cifar-10 to cifar10 folder.
nim c -r -d:release demoCifar10.nim # compile and start to train
While training, csv files which include accuracy, val-loss, and train-loss are stored in log folder. For visualization, visualize.py is placed here.
python3 visualize.py ./log/23481250.csv # example
# csv-name = [start-time(hour,minute,secound) + epoch].csv
This fig shows the performance of the current model. The accuracy can be improved of course.
Arraymancer -> v0.4.0
Arraymancer -> v0.5.0. This versrion looks working correclty.
This implementation works on my MBP(2013-Late, HighSierra) and Ubuntu18.
Nim -> v0.19.0
Arraymancer -> 0.5.0 or (0.4.0)
If you are interested in this, please try to get better models.
Note
The performance of Arraymancer v0.4.0 is not good, but the performance of Arraymancer v0.5.0 is better.
You should use newer Arraymancer.