Katon Minhas
c2d5db19d4
Feature/doczy codes * Ignore service and procedure generic terms * UPdate PT/OT/ST codes * Standardize stop-loss * UPdate for multiple identical values * Modify code funcs * Run highest match Implicit Level first * Fix implicit code bug * Add PT individually to special case * change revenue code conditions * Merge branch 'main' into feature/doczy-codes * Merged main into feature/doczy-codes * Update main * Relocate create_code_embeddings back to scripts * Remove investment code_funcs (use codes.code_funcs now) * Add docstrings * stop-loss * Update unit tests * Remove test * Remove reimb_dates * fix dtypes * Remove prints * update prompt * Update code breakout unit test * Merge branch 'main' into feature/doczy-codes * Merge branch 'main' into feature/doczy-codes 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 crosswalk.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")
|
|
)
|
|
|