Integration of CEN model

[JuliaGast]

CEN

ongoing

Tasks:

• [x] restructure code s.t. it fits tgb
• [x] implement data loading and make sure it is in correct format for CEN
• [x] integrate test()
• [x] store and load models at correct location
• [x] compute scores for negative samples only
• [x] compute and log mrr
• [x] compute and log runtime
• [ ] hyperparameter tuning

[JuliaGast]

• [x] decoder.py: do I need the tanh for samples_of_interest_embd?

[JuliaGast]

• [x] check if I can reproduce CEN results from apples-oranges-paper

[JuliaGast]

First Results:

YAGO:

[
    {
        "model": "CEN",
        "data": "tkgl-yago",
        "run": 1,
        "seed": 1,
        "test_history_len": 3,
        "val mrr": 0.7986021041870117,
        "test mrr": 0.8283807039260864,
        "test_time": 208.7344170063734,
        "tot_train_val_time": 5281.509197265841
    },
    {
        "model": "CEN",
        "data": "tkgl-yago",
        "run": 2,
        "seed": 2,
        "test_history_len": 3,
        "val mrr": 0.7964026927947998,
        "test mrr": 0.823512852191925,
        "test_time": 205.0702550476417,
        "tot_train_val_time": 4963.789887539111
    }
]

POLECAT mini:

{
    "model": "CEN",
    "data": "tkgl-polecat",
    "run": 1,
    "seed": 1,
    "test_history_len": 3,
    "val mrr": 0.17525655031204224,
    "test mrr": 0.15257658064365387,
    "test_time": 42.989435222931206,
    "tot_train_val_time": 1481.4672054741532
}

[JuliaGast]

done