A Deep learning Neural Network from the scratch via Numpy/Pandas

The basic complete deep learning neural network architecture includes:

• Forward propagation process with activation function (sigmoid and softmax) in the hidden layer and output laryer.
• Backward propagation process with derivative calculation and cross-entropy of each parameter (weight and bias).
• Update each parameter (weight and bias) simultaneously.

Task

1. Goal: train a multiple-label classification neural network based on collected data.
2. Method: handwritten the whole process.
3. Expected output: the "difference" between the prediction and the reality can gradually decrease.

Dataset

3 samples, 2 features, 3 target labels.

Neural Network architecture