The Fashion MNIST it's a dataset developed by Zalando SE and can be found in this link. The dataset has a 700000 images of Clothes and Accessories splited in tem 10 classes:
T-shirt/topTrouserPulloverDressCoatSandalShirtSneakerBagAnkle boot
- Reshape the image vector from 2D images to 3D with this distribuition (inputs size, height, columns, colors). That reshape basicly add one more dimesion corresponding the color, just is a pattern,.
- Rescale the value of each one pixel. THe variace between 0 and 255 probably will incresse too much the value of the weight, making the train more difficult. To solve that problem we rescale into 0 and 1.
- Tranform the result to categorial. The Fashion MNIST split the classes of the clothes using values between 0 and 9, but isn't scalar problem, so it's necessary tranform into a vector with mutiples results. E.g.: Case the value is 4 the result will be [0, 0, 0, 0, 1, 0, 0, 0, 0, 0]
- After train and retrain a lot of diferentes layers that was the sequencial whitch got more accuracy with less epochs and time trainning:
Developed by Anderson Laurentino @ 2019