Aug-20-2021, 07:09 PM
(This post was last modified: Aug-20-2021, 07:10 PM by AlekseyPython.)
Python 3.9.6, Pytorch 1.9.0+cu111
I have several text fields describing various aspects of a patient's treatment (filled in by the doctors). The entire samples is about 20 thousand cases. It's necessary to answer whether the severity of the disease was more than the specified one (yes or no).
From the book, I took the following sequence of layers of a neural network (but there's a completely different task):
Embedding (64 dimensions)
LSTM, unidirectional (64 -> 32)
Linear (32 -> 16)
Sigmoid
There are few samples (~ 20 thousands), so I tried to reduce the complexity of the network as much as possible, so that it really learns, and not just remembers samples.
Is there any good literature that explains, which layer is best used in which situation (and in what sequence)? Can you give tips on the correct network architecture for this case?
Thank you in advance for participating in the conversation.
I have several text fields describing various aspects of a patient's treatment (filled in by the doctors). The entire samples is about 20 thousand cases. It's necessary to answer whether the severity of the disease was more than the specified one (yes or no).
From the book, I took the following sequence of layers of a neural network (but there's a completely different task):
Embedding (64 dimensions)
LSTM, unidirectional (64 -> 32)
Linear (32 -> 16)
Sigmoid
There are few samples (~ 20 thousands), so I tried to reduce the complexity of the network as much as possible, so that it really learns, and not just remembers samples.
Is there any good literature that explains, which layer is best used in which situation (and in what sequence)? Can you give tips on the correct network architecture for this case?
Thank you in advance for participating in the conversation.