Amazing progress has been made in deep learning. I have been Tensorflow for a while now. I started out with tf0.6 then upgraded to tf0.12 then to tf1.0. The latest version is tf1.10 which is supposed to provide a stable API. I have a lot of code which has now become incompatible. The tf0.6’s saver API was totally different. tf.slim was added in tf0.12 (which I loved btw). But as of tf1.10 it is gone. It is recommended to use Keras instead. In this blog post, I am organizing some of my old examples and other useful code and making as much as possible tf1.10 compatible.
A while ago I had written a few articles on neural networks from first principles. All of them can be accessed here.
My tensorflow examples are available. They are compatible with tf1.10.
Here are the links to all my public code related to Neural Nets basics. I no more actively maintain it at this moment.
- ResNet – https://gist.github.com/mpkuse/6f9dcd419effa707422eb2c5097f51b4. Still useful in some of my projects. But I do not recommend using it anymore. Best is to move to Keras to build your networks.
- Toy Neural Net from first principles (in Matlab) to classify XOR: https://github.com/mpkuse/learning_xor. A useful resource to understanding the core.
- 3-Category classification from scratch: https://github.com/mpkuse/nn_classify.
- MNIST with caffe : https://github.com/mpkuse/mnist_in_memory_training. I no
longer use caffe.
- CIFAR with Tensorflow: https://github.com/mpkuse/cifar_tf. This was working with tf0.06. Pretty sure it is incompatible with tf1.10.