I'm a beginner in Machine Learning.
I've been learning about YOLO and DarkFlow from the following links with Ubuntu 20.04: darkflow and Tiny YOLO.
I successfully executed the code, and got the results like this:
Statistics:
car: 436
person: 73
Dataset size: 2599
Dataset of 2599 instance(s)
Training statistics:
Learning rate : 1e-05
Batch size : 16
Epoch number : 1000
Backup every : 2000
It's not bad, but the training's taking way too much time.
But I wanna know if there is any powerful IDE or other tools that can help me reduce time.
I searched in google and tried to find many ways to improve.
I heard that there are many ways to make training faster,(including Azure AI ML service) but since I'm a beginner, I cannot decide which will be the perfect choice to run YOLO and DarkFlow.
I would appreciate advices about robust development environments, especially the ones that would be suitable in my current specific condition.
Thanks in advance!
++) Since I'm a mere sophomore, the level of the hardware that I can use is very limited. I would also appreciate tools that can help me overcome the limitations of my hardware!
The IDE usually won't decrease computing time, but is rather based around the compiler. Likely, you'll just have to deal with intensive training times if you can't get better hardware. However, you might look into using a gpu to do computations(if you have one) rather than the standard way it runs on the cpu. Here's a link as to how to achieve this in Tensorflow: https://stackoverflow.com/a/51307381/14392018 . The general rule is the more data and the more complex the ML/DL model(i.e. the more layers in a neural network), the longer it takes to compute 1 epoch on the dataset. Also, the more data you're working with, the higher the computational intensity.