class ThreeRankTask(Task):
""" A strange rank task that runs two searchers on benchmark #1 (via TwoRank) and the third searcher on benchmark #2 """
module_name = "threerank"
dependencies = [
Dependency(key="tworank", module="task", name="tworank"),
Dependency(key="rank3", module="task", name="rank"),
]
class TwoRankTask(Task):
""" A rank tasks two runs two searchers on the same benchmark """
module_name = "tworank"
dependencies = [
Dependency(key="benchmark", module="benchmark", name="rob04yang", provide_this=True, provide_children=["collection"]),
Dependency(key="rank1a", module="task", name="rank"),
Dependency(key="rank1b", module="task", name="rank"),
]
class RankTask(Task):
module_name = "rank"
dependencies = [
Dependency(key="benchmark",
module="benchmark",
name="rob04yang",
provide_this=True,
provide_children=["collection"]),
Dependency(key="searcher", module="searcher", name="bm25"),
]
class RerankTask(Task):
module_name = "rerank"
config_spec = [
ConfigOption("fold", "s1", "fold to run"),
ConfigOption(
"optimize", "map", "metric to maximize on the dev set"
), # affects train() because we check to save weights
]
dependencies = [
Dependency(key="benchmark", module="benchmark", name="rob04yang", provide_this=True, provide_children=["collection"]),
Dependency(key="rank", module="task", name="rank"),
Dependency(key="reranker", module="reranker", name="DRMM"),
]
class RerankerDRMM(ModuleBase):
module_type = "reranker"
module_name = "DRMM"
dependencies = [
Dependency(key="extractor", module="extractor", name="embedtext"),
Dependency(key="trainer", module="trainer", name="pytorch"),
]
config_spec = [
ConfigOption("nbins", 29, "number of bins in matching histogram"),
ConfigOption("nodes", 5, "hidden layer dimension for matching network"),
ConfigOption("histType", "LCH", "histogram type: CH, NH, LCH"),
ConfigOption("gateType", "IDF", "term gate type: TV or IDF"),
]
class AParent(ModuleTypeA):
module_name = "AParent"
config_spec = [
ConfigOption(key="key1",
default_value="val1",
description="test option")
]
dependencies = [
Dependency(key="myfoo",
module="Atype",
name="AFoo",
default_config_overrides={"changethis": 42}),
Dependency(key="bar", module="Atype", name="ABar"),
]
class SearcherBM25(ModuleBase):
module_type = "searcher"
module_name = "bm25"
dependencies = [Dependency(key="index", module="index", name="anserini")]
config_spec = [ConfigOption(key="k1", default_value=1.0, description="k1 parameter")]
# Searchers are unlikely to actually need a seed, but we require it for testing
requires_random_seed = True
class ExtractorEmbedtext(ModuleBase):
module_type = "extractor"
module_name = "embedtext"
dependencies = [
Dependency(key="index", module="index", name="anserini", default_config_overrides={"stemmer": "none"}),
Dependency(key="tokenizer", module="tokenizer", name="anserini"),
]
config_spec = [
ConfigOption("embeddings", "glove6b"),
ConfigOption("zerounk", False),
ConfigOption("calcidf", True),
ConfigOption("maxqlen", 4),
ConfigOption("maxdoclen", 800),
ConfigOption("usecache", False),
]
class IndexAnserini(ModuleBase):
module_type = "index"
module_name = "anserini"
dependencies = [
Dependency(key="collection", module="collection", name="robust04")
]
config_spec = [
ConfigOption(key="stemmer",
default_value="porter",
description="stemming")
]
class AFoo(ModuleTypeA):
module_name = "AFoo"
config_spec = [
ConfigOption(key="foo1",
default_value="val1",
description="test option"),
ConfigOption(key="changethis",
default_value=0,
description="something to override"),
]
dependencies = [
Dependency(key="myfoobar", module="Atype", name="AFooBar")
]
class BenchmarkTRECDL(ModuleBase):
module_type = "benchmark"
module_name = "trecdl"
dependencies = [
Dependency(key="collection", module="collection", name="msmarco")
]
class BenchmarkRob04(ModuleBase):
module_type = "benchmark"
module_name = "rob04yang"
dependencies = [
Dependency(key="collection", module="collection", name="robust04")
]