This experiment takes a very simple and straight forward approach of training a Neural Network to classify the MNIST dataset, consisting of handwritten numbers from 0-9 saved in 28x28 images. Here's what a sample of what the numbers in MNIST look like:
To get started, use the terminal to navigate to ml-in-tf/experiments/nn-mnist/and run python nn-mnist.py.
To see the graph and plots using tensorboard, use the terminal to navigate to ml-in-tf/ and run tensorboard --logdir logs/. Wait for the following message:
Starting TensorBoard on port <port>
And then open up a browser and go to localhost:<port>.
The network in this experiment has one fully connected layer. It takes every single image pixel as input represented as a one-dimensional of size 784 (=28x28) and outputs a softmax layer of size 10, one for each number-class.
| Input | Output |
|---|---|
| 784 | 10 |
The customizable parameters of this experiment - and their default values - are as follows:
minibatches-1000-Number of minibatches to run the training on.minibatch_size-100- Number of samples in each minibatch.status_update-100- How often to print an status update.random_seed-123- Number of minibatches to run the training on.learning_rate-0.05- Learning rate of the optimizer.optimizer-gradient_descent- If another optimizer should be used [adam, rmsprop, gradient_descent]. Defaults togradient_descent.run_test-True- If the final model should be tested.use_gpu-False- If TensorFlow operations should run on GPU rather than CPU.
The implementation is really straight forward and should receive an accuracy of ≈88% on the test set. The plot below shows a standard run will all default parameters.
