CSCI 548 Group 8: Knowledge-Graph-Completion (ditk.graph_completion)

Deliverables

1. graph_completion.py - abstract class with abstract methods
2. <module_name>.py - implementation of KG completion model (1 per selected KG completion repository) -- module hierarchy: ditk.graph_completion.<module_name>

Benchmarks

• WN18
• FB15
• CORA

Try to run model on all 3 datasets. If the dataset does not fit the format of your model, you will have to preprocess it so it does. If you are unable to run on a dataset, provide an explaination why so.

Evaluation Metrics

• AUC
• Average Precision
• MRR
• HITS@10

Try to train and evaluate on each of the above benchmark datasets. Do not worry about the score. If you are unable to run on a specific benchmark, provide explaination why. If you are unable to use an evaluation metric, specify why.

FAQ

1. My python version is different

• Separate packages for python version on gitbub (how?)

2. My dependency versions are different (e.g. tensorflow)

• For tensorflow there is package to make them compatible (prof will send more details)

3. Do we need to use pretrained weights

• No, just evaluation is fine

4. What is the pointer to parent class

• Just add url link to the parent class aka this repo