Gemstones Neural Network - multiclass classification

Build a simple convolutional neural network (CNN) from scratch over the dataset of gemstones images

Can your program understand is it Ruby, Amethyst or Emerald?

Full dataset with gemstones images can be found at MY KAGGLE PAGE: Gemstones Images: it's already divided into train and test data. This dataset contains 3,000+ images of different gemstones.

• I. Create dataset: parse images: from static and dynamic resources, create train and test folders (90%:10%)
• II. Install libraries for building Neural Network (+ GPU)
• III. Prepare to build Convolutional Neural Network: working over train folder (read and convert images and labels into numpy-arrays, split train data into train and validation sets)
• IV. Build a Convolutional Neural Network for multiclass classification: import necessary modules, provide Hyperparameters, provide a model: choose an architecture (layers) and compile it.
• V. Fit the CNN: provide more data - image augmentation, fit the model with train data and wait fit_generator to complete its work.
• VI. Undestand the model: plot accuracy and loss, score model, understand confusion matrix, save weights as .h5 file to reuse it in the future.
• VII Evaluate on testing folder: understand test data, some conclusion.

Feel free to give any suggestions to improve my code.
Upvote this repo or/and my Kaggle solutions if you find them useful!