Introducing Self-Attention to Target Attentive Graph Neural Networks (arXiv)

[ Model Architecture ]	[ Task ]	[ Paper ]	[ arXiv ]

PyTorch implementation of the model TAGNN++, presented in the paper "Introducing Self-Attention to Target Attentive Graph Neural Networks"

Accepted at AISP '22

---

Usage

python3 train.py [-h] [--dataset DATASET] [--defaults DEFAULTS] [--batchSize BATCHSIZE]
               [--hiddenSize HIDDENSIZE] [--nhead NHEAD] [--layer LAYER]
               [--feedforward FEEDFORWARD] [--epoch EPOCH] [--lr LR]
               [--lr_dc LR_DC] [--lr_dc_step LR_DC_STEP] [--l2 L2]
               [--patience PATIENCE] [--validation] 
               [--valid_portion VALID_PORTION]

Arguments:
  -h, --help            Description and Help Message
  --dataset DATASET     Name of the Dataset:
                        diginetica | yoochoose1_64
  --defaults DEFAULTS   Use default configuration   
  --batchSize BATCHSIZE
                        Batch size
  --hiddenSize HIDDENSIZE
                        Hidden state dimensions
  --epoch EPOCH         The number of epochs to train
  --lr LR               Set the Learning Rate
  --lr_dc LR_DC         Set the decay rate used with Learning rate
  --lr_dc_step LR_DC_STEP
                        Steps in Learning rate decay
  --l2 L2               Assign L2 Penalty
  --patience PATIENCE   Early stopping criterion
  --validation          validation
  --valid_portion VALID_PORTION
                        Portion of train-set to split into val-set

---

Dependencies

This code was developed with python3.6

Python (3.6.x)
PyTorch (1.7.x)
CUDA (10.2)
cuDNN (7.6.5)
networkx (2.5.1)
numpy (1.19.5)     

For original source of AGC, for further tweaks:

git clone https://github.com/vballoli/nfnets-pytorch.git   

Problem Statement Formulation

For further details, contact Sai Mitheran via Linkedin, or via email by clicking the icon below.

---

Reference

To cite our paper:

@article{mitheran2021mproved,
    title={Improved Representation Learning for Session-based Recommendation},
    author={Sai Mitheran and Abhinav Java and Surya Kant Sahu and Arshad Shaikh},
    year={2021},
    journal={arXiv preprint arXiv:2107.01516}
}

---