This project is based on the mnist dataset - using TensorFlow in a convoluted neural network
The mnist dataset is a collection of handwritten numbers, being loaded into a cNN
For this project I've trained the model with 7058 steps, being able to get a loss = 0.29477254(which is okay, I might say)
The images are loaded into the model 28*28 pixel on colorchannel 1 -> black/white (grayscale)
The model is using the ReLu activation function, to filter out negative values, and normalise the data.
Information about dead neurons due to ReLu being used is not provided.
The datatype is set to an int32. (standard)
The last training provided follow information: INFO:tensorflow:loss = 0.12128864, step = 7701 (18.603 sec) INFO:tensorflow:loss = 0.18814294, step = 8960 (16.587 sec)
During training, it's training on 100 examples on each step, which would translate to -> 896.000 examples that have been trained on. The model is set to train 20.000 times.
On step 22.629 I got the following: INFO:tensorflow:loss = 0.095737554, step = 22629 (19.141 sec)