This project refers to the paper "Measuring abstract reasoning in neural networks" (https://arxiv.org/abs/1807.04225)
Data: https://console.cloud.google.com/storage/browser/ravens-matrices
description of database: https://github.com/deepmind/abstract-reasoning-matrices
Images are of size 160x160x16. The last dimension denotes the panel number for the matrix, with the first 8 panels being the "context", and the last 8 being the "choices".
NB: each matrix of context is built by randomly sampling of:
- relation types (R, with elements r): progression, XOR, OR, AND, consistent union 1
- object types (O, with elements o): shape, line
- attribute types (A, with elements a): size, type, colour, position, number
Programming source code: I try to write codes with reference to the following site: https://github.com/gitlimlab/Relation-Network-Tensorflow However, the code use h5py file as data source. To make life easier, I wrote a testing program "h5py_generator.ipynb" to create the required h5py file with data for Relation Network and a program convert_npz_hy.ipynb which convert npz files to hy file. However, the created hy file is too big in size even for just convert 625 npz files.