zhavei/image_clasification — reverse-engineered prompt

Build me a beginner friendly Google Colab notebook for a rock paper scissors image classifier.

I want it to download or use the rockpaperscissors dataset from the Dicoding GitHub link, split it so 40 percent is validation data, and train a model that can tell whether a hand image is rock, paper, or scissors. Please include image augmentation so the model is not too fragile, and keep the training fast enough to finish in under 30 minutes.

Use a simple sequential neural network with more than one hidden layer, Adam optimizer, and categorical crossentropy. The goal is at least 85 percent accuracy. Show the training and validation accuracy and loss clearly, and include a final evaluation so I can see how well it worked.

Please make the notebook easy to run from top to bottom in Colab, with short explanations in plain English for each step. If anything has changed in the current libraries, look up the current docs and adjust it.