Add embeddings test

[mcollardanuy]

Test from other repository:

def test_compare_embeddings():
    """
    Compare our embeddings database with the old database.
    """

    # Test 1: Check old glove embedding is the same as new glove
    # embedding:
    mentions = ["apple"]
    embs = rel_utils.get_db_emb(
        "resources/rel_db/embedding_database.db", mentions, "snd"
    )
    with sqlite3.connect("resources/rel_db/generic/common_drawl.db") as conn2:
        c2 = conn2.cursor()
        result = c2.execute(
            "SELECT emb FROM embeddings WHERE word=?", ("apple",)
        ).fetchone()
        result = result if result is None else array("f", result[0]).tolist()

    assert embs[0] == result

    # Test 2: Check old glove embedding is different from new wiki2vec
    # word embedding:
    mentions = ["apple"]
    embs = rel_utils.get_db_emb(
        "resources/rel_db/embedding_database.db", mentions, "snd"
    )
    with sqlite3.connect("resources/rel_db/generic/entity_word_embedding.db") as conn2:
        c2 = conn2.cursor()
        result = c2.execute(
            "SELECT emb FROM embeddings WHERE word=?", ("apple",)
        ).fetchone()
        result = result if result is None else array("f", result[0]).tolist()

    assert embs[0] != result

    # Test 3: Check old wiki2vec word embedding is the same as new wiki2vec
    # word embedding:
    mentions = ["apple"]
    embs = rel_utils.get_db_emb(
        "resources/rel_db/embedding_database.db", mentions, "word"
    )
    with sqlite3.connect("resources/rel_db/generic/entity_word_embedding.db") as conn2:
        c2 = conn2.cursor()
        result = c2.execute(
            "SELECT emb FROM embeddings WHERE word=?", ("apple",)
        ).fetchone()
        result = result if result is None else array("f", result[0]).tolist()

    assert embs[0] == result

    # Test 4: Check old wiki2vec entity embedding is the same as new wiki2vec
    # entity embedding (old requires wikipedia input, new requires wikidata
    # input):
    mentions = ["Q84"]
    embs = rel_utils.get_db_emb(
        "resources/rel_db/embedding_database.db", mentions, "entity"
    )
    with sqlite3.connect("resources/rel_db/generic/entity_word_embedding.db") as conn2:
        c2 = conn2.cursor()
        result = c2.execute(
            "SELECT emb FROM embeddings WHERE word=?", ("ENTITY/London",)
        ).fetchone()
        result = result if result is None else array("f", result[0]).tolist()

    assert embs[0] == result

    # Test 5: Check old wiki2vec entity embedding is the same as new wiki2vec
    # entity embedding (old requires wikipedia input, new requires wikidata
    # input):
    mentions = ["Q84"]
    embs = rel_utils.get_db_emb(
        "resources/rel_db/embedding_database.db", mentions, "entity"
    )
    with sqlite3.connect("resources/rel_db/generic/entity_word_embedding.db") as conn2:
        c2 = conn2.cursor()
        result = c2.execute(
            "SELECT emb FROM embeddings WHERE word=?", ("London",)
        ).fetchone()
        result = result if result is None else array("f", result[0]).tolist()

    assert embs[0] != result