#Fashion-MNIST ##Introduction The clue of the task is to implement a model that allows classification of thumbnails of photos representing clothes from Fashion-MNIST. Fashion-MNIST is a dataset of Zalando's article images-consisting of a training set of 60,000 examples and a test set of 10,000 examples. Each example is a 28x28 grayscale image, associated with a label from 10 classes. We'll be covering three classifiers: KNN and MLP.
Here we have an example of how the data looks:
##Methods First classifier we'll be covering is KNN (k-nearest neighbours). The Distance Metric I decided to use is Manhattan Distance. The KNN algorithm assumes that similar things exist in close proximity. K parameter represents number of neighbours we take into consideration. On my computer it takes around 30 minutes to calculate Manhattan distance matrix.
Next, we have MLP classifier (Multi-layer Perceptron), which is one of the simplest Neural Network classifier. Multi-layer Perceptron classifier consists of:
- input layer
- hidden layers
- output layer
Most of the magic happens in hidden layers. In my algorithm i use:
- Activation function I use in hidden layers is the ReLU(Rectified Linear Units) function, because it's cheap to compute as there is no complicated math and hence easier to optimize.
- Dropout, which is a computationally cheap way to regularize our neural network. It's a technique where randomly selected neurons are ignored during training.
- Dense layer, which is fully connected with previous layer.
- Kernel regularizer to apply a penalty on the layer's kernel.
On output layer we use Softmax function because our output is multidimensional.
##Results
| Classifier | my Accuracy | Accuracy on benchmark |
|---|---|---|
| KNN{k = 7} | 86,28% | 86% |
| MLP | 89,5% | 87,7% |
##Usage To run our classifier, we'll need to download Fashion-MNIST data from here and put it in 'fashion' directory in the project. We need to have Tensorflow and Numpy installed. After these steps, we can run our classifiers by running: 'classifierMLP.py', 'classifierKNN.py'. To use best fitted MLP classifier, uncomment 57th line and comment 58th.