Posted on • Originally published at github.com
4
1
Implementation of a digit recognition using my neural network with MNIST
Live version here
I made this to test how long it would take to train my neural network with MNIST and accurate it would be at the end. Turns out I have a long way to go.
Draw the digit on the white rectangle (try to use to whole area) and press predict for the neural network indentify the digit. Press clear to clear the white drawing area.
Check out the code here
SurveyJS UI libraries let you build a JSON-based form management system that integrates with any backend, giving you full control over your data with no user limits. Includes support for custom question types, skip logic, an integrated CSS editor, PDF export, real-time analytics, and more.
For further actions, you may consider blocking this person and/or reporting abuse
Docusign is now extensible. Overcome challenges with disconnected products and inaccessible data by bringing your solutions into Docusign and publishing to 1.6M customers in the App Center.
Mike Young -
Mike Young -
Mike Young -
Mike Young -
Top comments (9)
Love it!
It was actually working really well for me, until I threw it a curve ball with a seriffed 1.
Someone made an example using Tensor Flow a while back. Theirs are not perfect either. Mighty have something to do with MNIST data set.
gogul09.github.io/software/digit-r...
Yeah, some times it does that. To be honest the error on the training it is still high, but it was taking too long to converge.
The live version doesn't perform well. What features did you use to train this network?
The MNIST data set
Did you use pixels for features?
Yes, indeed that's how you train a neural network with images.
Doesn't feel very resilient, though.
I've updated the code to use the last version of my neural network. It's a little better this time.