Neural-Net
An implementation of neural-net with just numpy.
Requirements
Use pip install -U numpy to install numpy.
Guide
Activations
Activation functions available are:
- ReLU
- Sigmoid
- Tanh
- LReLU
Losses
Loss functions available are:
- MSE
- BinaryCrossEntropy
- SoftmaxCrossEntropy
- SparseSoftmaxCrossEntropy
Optimizers
Loss functions available are:
- SGD
- SGDMom
- AdaGrad
- RMSProp
- Adam
Others
- You can pass keep_probs (list) which can initialize dropouts in layers.
- For activations pass a list of activation functions as strings (for eg. ['relu', 'lrelu', 'sigmoid', 'tanh', 'linear']) during initializations.
- Either pass a string for optimizer (for eg. 'adam') along with learning_rate or assign your own optimizer from optimizers.py and pass that.
- Similarly for loss function, either pass a string (for eg. 'mean_squared_error') or assign a loss function from losses.py.
- For training use model.fit() and pass the required arguments.
- For predicion use model.predict() and pass the required arguments.