Attentive-Neural-Process

Description

• A pytorch implementation of Attentive Neural Process.
• Simple code for generating samples with ANP.
• I will update the super-resolution experiments soon.

Requirements

• Install python 3
• Install pytorch == 0.4.0

File description

• preprocess.py includes all preprocessing codes when you loads data.
• module.py contains all methods, including attention, linear and so on.
• network.py contains whole structure of network.
• train.py is for training ANP model.
• generate.ipynb is for generating samples.

Results

test samples after 50 epoch training with random context selection.

• original
• 10 contexts
• 50 contexts
• 100 contexts
• half contexts

Reference

• https://github.com/deepmind/neural-processes

Comments

• Any comments for the codes are always welcome.