Welcome to your exciting journey into the world of Multi-Hop Knowledge Graph Reasoning! In this article, we will guide you through the process of setting it up, running experiments, and troubleshooting the common issues you may encounter along the way. Let’s dive in!

Quick Start

Before we dive into the specifics, you need to set up your environment. There are two main methods to do this: using Docker or setting it up manually.

Using Docker

• Build the Docker image:

docker build -f Dockerfile -t multi_hop_kg:v1.0

• Spin up a Docker container and run your experiments inside it:

nvidia-docker run -v pwd:workspace MultiHopKG -it multi_hop_kg:v1.0

• The remaining instructions assume you are working inside a container.

Manually Set Up

• Install PyTorch (version 0.4.1) manually.
• Use the Makefile to set up the rest of the dependencies:

make setup

Process Data

To get started with data, unpack the data files and run the following command to preprocess the datasets:

tar xvzf data-release.tgz

.experiment.sh configs/dataset.sh --process_data gpu-ID

Here, dataset refers to one of the five datasets located in the .data directory: umls, kinship, fb15k-237, wn18rr, or nell-995. The gpu-ID represents the index of your GPU as a non-negative integer.

Train Models

Now, let’s train our models! Here are the commands to do that:

• Train embedding-based models:

.experiment-emb.sh configs/dataset-emb_model.sh --train gpu-ID

• Train Reinforcement Learning (RL) models (policy gradient):

.experiment.sh configs/dataset.sh --train gpu-ID

• Train RL models (policy gradient + reward shaping):

.experiment-rs.sh configs/dataset-rs.sh --train gpu-ID

Evaluate Pretrained Models

To evaluate a pretrained model, simply modify the --train flag to --inference in your command:

.experiment-rs.sh configs/dataset-rs.sh --inference gpu-ID

You can also print the inference paths generated by beam search during inference:

.experiment-rs.sh configs/dataset-rs.sh --inference gpu-ID --save_beam_search_paths

Change the Hyperparameters

If you want to modify hyperparameters or the experimental setup, simply start with the configuration files.

Implementation Details

For our experiments, we use mini-batch training. To avoid memory issues caused by large fan-outs, we group different nodes’ action spaces into buckets based on their sizes. You can check out the bucket implementation details here and here.

Troubleshooting

If you encounter any issues during your setup or execution, consider the following tips:

• Ensure that your GPU is properly configured and recognized.
• Check the version of PyTorch to ensure compatibility.
• Make sure all dependencies are correctly installed as per your setup method.

For more insights, updates, or to collaborate on AI development projects, stay connected with fxis.ai.

Conclusion

At fxis.ai, we believe that such advancements are crucial for the future of AI, as they enable more comprehensive and effective solutions. Our team is continually exploring new methodologies to push the envelope in artificial intelligence, ensuring that our clients benefit from the latest technological innovations.

That’s all for your introduction to Multi-Hop Knowledge Graph Reasoning with Reward Shaping! Happy experimenting!