Networks - A Machine/Deep Learning Library
Machine Learning and Deep Learning Models from Scratch.
This Library allows users to create the following models:
- Feed-Forward Neural Networks
- Convolution Neural Networks
- Linear Regression
- Logistic Regression
Without having to write any backpropagation code.
To install the Networks Library
pip install networks
Layers in the Library & their Parameters in Add function
Activation Layers
1. Relu Layer
No Params
2. Sigmoid Layer
No Params
3. Tanh Layer
No Params
4. Leaky Relu Layer
No Params
Normalization Layers
1. Batch Normalization Layer
batch_params={
'mode':'train'/'test',
'momentum':0.9,
'eps':1e-8
}
2. Spatial Batch Normalization Layer
batch_params={
'mode':'train'/'test',
'momentum':0.9,
'eps':1e-8
}
Convolution Layers
1. Max Pooling Layer
pooling_params={
'pooling_height':2,
'pooling_width':2,
'pooling_stride_height':2,
'pooling_stride_width':2
}
2. Convolution Layer
num_kernels=64,
kernel_h=3,
kernel_w=3,
convolution_params={
'stride':1
}
3. Padding Layer
padding_h=2, padding_w=2
Loss Layers
1. Softmax Loss Layer
No params
2. SVM Loss Layer
No params
3. Mean Squared Error Layer
No params
4. Cross Entropy Loss Layer
No params
Fully Connected Layer
1. Affine Layer
affine_out = 64
2. Flatten Layer
No params
Example Usage
from networks.network import network
model = network(input_shape=(64,1,50,50),initialization="xavier2",
update_params={
'alpha':1e-3,
'method':'adam',
'epoch':100,
'reg':0.01,
'reg_type':'L2',
'offset':1e-7
})
To Add Padding Layer
model.add("padding",padding_h=3,padding_w=3)
To Add Convolution Layer
model.add("convolution",num_kernels=64,kernel_h=3,kernel_w=3,
convolution_params:{
'stride':1
})
To Add Relu Layer
model.add("relu")
To Add Pooling Layer
model.add("pooling",pooling_params={
"pooling_height":2,
"pooling_width":2,
"pooling_stride_height":2,
'pooling_stide_width':2
})
To Add Batch Normalization Layer
model.add("batch_normalization",
batch_params={'mode':'train'/'test','momentum':0.9,'eps':1e-8})
To Add Spatial Batch Normalization Layer
model.add("spatial_batch",
batch_params={'mode':'train'/'test','momentum':0.9,'eps':1e-8})
To Add a Flatten Layer
model.add("flatten")
To Add Affine Layer
model.add("affine",affine_out=128)
To Add Softmax Loss Layer
model.add("softmax")
To Add SVM Loss Layer
model.add("svm")
To Add MSE Loss Layer
model.add("mse")
To Add Cross Entropy Loss Layer
model.add("cross_entropy")
To Save Model
model.save("model.json")
To Load Model
model = network.load("model.json")
To Train Model
model.train(X,y)
To Get Accuracy & Loss After Training
accuracy,loss = model.test(validX,validY)
To Predict
predictions = model.predict(X)
