The dataset has been taken from thttps://www.kaggle.com/c/digit-recognizer.
1.Python
2.Angular 8
layers : Conv2D-->MaxPool2D-->Dropout-->Conv2D-->MaxPool2D-->Dropout-->Conv2D--->MaxPool2D---->Flatten--->Dense---->Dropout--->Dense
activation : (relu) (relu) (relu) (relu) (sigmoid)
fliter : 32,(5,5)
pool_size = (2,2)
input_shape = (28,,3)
output_shape = 128
padding="same"
loss='binary_crossentropy'
optimizer='adam'
metrics=['accuracy']
I have used h5 format for saving and later converted into tensorflowjs format inorder for interfernce in javascript.