def test_bagofwords_id2vec(tmpdir, dummy_index):
benchmark = DummyBenchmark({})
tok_cfg = {"name": "anserini", "keepstops": True, "stemmer": "none"}
tokenizer = AnseriniTokenizer(tok_cfg)
extractor = BagOfWords(
{
"name": "bagofwords",
"datamode": "unigram",
"maxqlen": 4,
"maxdoclen": 800,
"usecache": False
},
provide={
"index": dummy_index,
"tokenizer": tokenizer,
"benchmark": benchmark
},
)
extractor.stoi = {extractor.pad_tok: extractor.pad}
extractor.itos = {extractor.pad: extractor.pad_tok}
extractor.idf = defaultdict(lambda: 0)
# extractor.preprocess(["301"], ["LA010189-0001", "LA010189-0002"], benchmark.topics["title"])
extractor.qid2toks = {"301": ["dummy", "doc"]}
extractor.stoi["dummy"] = 1
extractor.stoi["doc"] = 2
extractor.itos[1] = "dummy"
extractor.itos[2] = "doc"
extractor.docid2toks = {
"LA010189-0001": [
"dummy", "dummy", "dummy", "hello", "world", "greetings", "from",
"outer", "space"
],
"LA010189-0002": [
"dummy", "dummy", "dummy", "hello", "world", "greetings", "from",
"outer", "space"
],
}
transformed = extractor.id2vec("301", "LA010189-0001", "LA010189-0001")
# stoi only knows about the word 'dummy' and 'doc'. So the transformation of every other word is set as 0
assert transformed["qid"] == "301"
assert transformed["posdocid"] == "LA010189-0001"
assert transformed["negdocid"] == "LA010189-0001"
assert np.array_equal(transformed["query"], [0, 1, 1])
assert np.array_equal(transformed["posdoc"], [6, 3, 0])
assert np.array_equal(transformed["negdoc"], [6, 3, 0])
assert np.array_equal(transformed["query_idf"], [0, 0, 0])
def test_bagofwords_id2vec_trigram(tmpdir, dummy_index):
benchmark = DummyBenchmark({})
tok_cfg = {"name": "anserini", "keepstops": True, "stemmer": "none"}
tokenizer = AnseriniTokenizer(tok_cfg)
extractor = BagOfWords(
{
"name": "bagofwords",
"datamode": "trigram",
"maxqlen": 4,
"maxdoclen": 800,
"usecache": False
},
provide={
"index": dummy_index,
"tokenizer": tokenizer,
"benchmark": benchmark
},
)
extractor.stoi = {extractor.pad_tok: extractor.pad}
extractor.itos = {extractor.pad: extractor.pad_tok}
extractor.idf = defaultdict(lambda: 0)
# extractor.preprocess(["301"], ["LA010189-0001", "LA010189-0002"], benchmark.topics["title"])
extractor.qid2toks = {"301": ["dummy", "doc"]}
extractor.docid2toks = {
"LA010189-0001": [
"dummy", "dummy", "dummy", "hello", "world", "greetings", "from",
"outer", "space"
],
"LA010189-0002": [
"dummy", "dummy", "dummy", "hello", "world", "greetings", "from",
"outer", "space"
],
}
extractor.stoi["#du"] = 1
extractor.stoi["dum"] = 2
extractor.stoi["umm"] = 3
extractor.itos[1] = "#du"
extractor.itos[2] = "dum"
extractor.itos[3] = "umm"
transformed = extractor.id2vec("301", "LA010189-0001")
# stoi only knows about the word 'dummy'. So the transformation of every other word is set as 0
assert transformed["qid"] == "301"
assert transformed["posdocid"] == "LA010189-0001"
assert transformed.get("negdocid") is None
# Right now we have only 3 words in the vocabular - "<pad>", "dummy" and "doc"
assert np.array_equal(transformed["query"], [5, 1, 1, 1])
assert np.array_equal(transformed["posdoc"], [
39, 3, 3, 3
]) # There are 6 unknown words in the doc, so all of them is encoded as 0
assert np.array_equal(transformed["query_idf"], [0, 0, 0, 0])
# Learn another word
extractor.stoi["mmy"] = 4
extractor.stoi["my#"] = 5
extractor.stoi["#he"] = 6
extractor.itos[4] = "mmy"
extractor.itos[5] = "my#"
extractor.itos[6] = "#he"
transformed = extractor.id2vec("301", "LA010189-0001")
# The posdoc transformation changes to reflect the new word
assert np.array_equal(transformed["posdoc"], [32, 3, 3, 3, 3, 3, 1])
