[COMMUNITY PROJECT] Machine Learning!

Question

Hey everyone,

I'm in the process of creating an image classifier that attempts to identify a number from a handwritten image. At the moment it only has 4 sets of training data (4x 0 to 9 handwritten images) and I'm looking for more training data.

As it stands, it's not extremely accurate but it works - this is why I need more training data to increase efficiency. Do you think this could work as a community project?

Let's see if we can add more training data to the classifier :]

Not sure how to go about it, maybe you awesome people can help. But best bet would be to send through to me (let me know the best ways) your training data which includes:

- 10 handwritten images (I did mine on a touchscreen tablet)
- the digits 0 to 9, one on each image
- size: 28x28 pixels
- colour: RGB

Here's the code below if you want to know what I'm on about:

https://code.sololearn.com/cR8OUhsdTCpm/#py

Really hope we can find a way to make this work!

Answer 1

you don't have to create them on your own!

here are some useful links:

https://deeplearning4j.org/opendata
http://www.cs.nyu.edu/~roweis/data.html
https://www.google.at/url?sa=t&source=web&rct=j&url=https://stackoverflow.com/questions/963041/data-sets-for-neural-network-training&ved=0ahUKEwj889Gn8qnXAhUJJOwKHWDEBh8QjjgIKjAA&usg=AOvVaw2Hapg1YnOWEdXnkYzL72Le

Answer 2

Awesome, thanks guys!

Answer 3

Thanks James. I'm not in the industry anyway, coding is just a hobby for the time being. I guess I just like challenging myself, not a big fan of using things when I don't understand their inner working

Answer 4

I've tried really hard to not use modules, wanted to see if I could get it working on it's own :) what's the difference between using the numpy array and the flattening I did btw?

Answer 5

Thanks for the advice man. Haven't looked too deeply into it, but do the docs explain what the module functions do (like break it down) or is there anywhere I can go to find out the inner working of the modules? Knowing how it works would make me feel heaps better about relying on it

Answer 6

Awesome reference, thanks James :)