Katon Minhas
c4e519894b
Refactor/one time io * Delete client work * Streamline imports * Fix list class * Update tests * Clear code funcs test * Fix unit tests * Single read code funcs * Single-read model * Single-load embeddings * Successful E2E test * update unit tests * Update importsg * Reload poetry.lock * remove old exhibit header function * Remove aarete_derived generic function * remove align and format tables * Remove strings to dict * references * Clear code * Move preprocessing_funcs * remove keywords * refactor postprocessing_funcs * Pass unit test - remove qa_qc directory * Black and isort * remove print Approved-by: Alex Galarce
31 lines
1.5 KiB
Python
31 lines
1.5 KiB
Python
|
|
import src.utils.embedding_utils as embedding_utils
|
|
from src.utils.crosswalk_utils import CrosswalkBuilder
|
|
import src.codes.code_funcs as code_funcs
|
|
|
|
import os
|
|
import pandas as pd
|
|
|
|
from sentence_transformers import SentenceTransformer
|
|
|
|
# Load model
|
|
roberta_model = SentenceTransformer("all-roberta-large-v1")
|
|
|
|
mapping_dir = "crosswalk/mapping_csvs/proc_cd"
|
|
pkl_dir = "embeddings"
|
|
for filename in os.listdir(mapping_dir):
|
|
if filename.endswith(".csv") and "level" in filename:
|
|
print(f"Creating embeddings for {filename}")
|
|
stripped_filename = filename.replace(".csv", "")
|
|
mapping_df = pd.read_csv(os.path.join(mapping_dir, filename))
|
|
proc_crosswalk = CrosswalkBuilder().from_df(mapping_df, from_col="Code", to_col="Description")
|
|
proc_choices = [x for x in proc_crosswalk.mapping.values()]
|
|
os.makedirs(os.path.join(pkl_dir, stripped_filename), exist_ok=True)
|
|
proc_index = embedding_utils.create_faiss_index(choices=proc_choices,
|
|
model=roberta_model,
|
|
save_path=os.path.join(pkl_dir, stripped_filename, "faiss_index.bin"),
|
|
embedding_path=os.path.join(pkl_dir, stripped_filename, "embeddings.npy"),
|
|
choices_path=os.path.join(pkl_dir, stripped_filename, "choices.pkl")
|
|
)
|
|
|