def to_df_by_assay(self, assay: str) -> pd.DataFrame:
raw = cache("raw",
Path("tox21") / f"{assay}.zip",
lambda: self.download(assay))
raw_fd = BytesIO(raw)
with ZipFile(raw_fd) as zip_fd:
for inner_filename in zip_fd.namelist():
if inner_filename.endswith("aggregrated.txt"):
with zip_fd.open(inner_filename) as inner_fd:
return pd.read_csv(inner_fd, sep="\t", index_col=False)
def __init__(self, mode="bow"):
model_file = f"{self.filename}_{mode}.pkz"
if mode == "default":
self.model = cache("model", model_file, self.train)
elif mode == "bow":
self.model = cache(
"model",
model_file,
lambda: self.train(
ChemicalLanguageHyperparameters.from_dict({
"vector_algo": "bow",
"max_vocab": 5000,
"max_ngram": 4
})),
)
elif mode == "lda":
self.model = cache(
"model",
model_file,
lambda: self.train(
ChemicalLanguageHyperparameters.from_dict({
"vector_algo": "lda",
"topics": 1000
})),
)
elif mode == "doc2vec":
self.model = cache(
"model",
model_file,
lambda: self.train(
ChemicalLanguageHyperparameters.from_dict({
"doc_epochs": 300,
"vec_dims": 512
})),
)
else:
raise ValueError("Invalid mode: " + mode)
def __init__(self, mode="prod"):
model_file = f"{self.filename}_{mode}.pkz"
if "doc2vec" in mode:
self.preprocessor = ChemicalLanguage("doc2vec")
elif "lda" in mode:
self.preprocessor = ChemicalLanguage("lda")
else:
self.preprocessor = ChemicalLanguage("bow")
if "test" in mode:
self.safety, self.feasibility, self.bbbp = self.train(
score=True, task_duration=12000
)
else:
self.safety, self.feasibility, self.bbbp = cache(
"model", model_file, self.train
)
def __init__(self):
self.preprocessor = ChemicalLanguage("bow")
self.model = cache(MODEL_ROOT / self.filename, self.train)
def to_df(self) -> pd.DataFrame:
name = self.filename + ".parquet"
return cache("raw", name, self.download)
def to_df(self) -> pd.DataFrame:
name = self.filename + ".parquet"
return cache("constructed", name, self.construct)