paperswithcode_nbow / app_implementation.py
lambdaofgod's picture
test metrics for best models
c5a2694
raw
history blame
4.41 kB
from typing import Dict, List
import torch
import pandas as pd
import streamlit as st
from findkit import retrieval_pipeline
import config
from search_utils import (
RetrievalPipelineWrapper,
get_doc_cols,
get_repos_with_descriptions,
get_retrieval_df,
merge_cols,
)
class RetrievalApp:
def is_cuda_available(self):
try:
t = torch.Tensor([1]).cuda()
except:
return False
finally:
return True
def get_device_options(self):
if self.is_cuda_available():
return ["cuda", "cpu"]
else:
return ["cpu"]
@st.cache(allow_output_mutation=True)
def get_retrieval_df(self):
return get_retrieval_df(self.data_path, config.text_list_cols)
def __init__(self, data_path="lambdaofgod/pwc_repositories_with_dependencies"):
self.data_path = data_path
self.device = st.sidebar.selectbox("device", self.get_device_options())
print("loading data")
self.retrieval_df = self.get_retrieval_df().copy()
model_name = st.sidebar.selectbox("model", config.model_names)
self.query_encoder_name = "lambdaofgod/query-" + model_name
self.document_encoder_name = "lambdaofgod/document-" + model_name
doc_cols = get_doc_cols(model_name)
st.sidebar.text("using models")
st.sidebar.text("https://huggingface.co/" + self.query_encoder_name)
st.sidebar.text("HTTP://huggingface.co/" + self.document_encoder_name)
self.additional_shown_cols = st.sidebar.multiselect(
label="used text features", options=config.text_cols, default=doc_cols
)
@staticmethod
def show_retrieval_results(
retrieval_pipe: RetrievalPipelineWrapper,
query: str,
k: int,
all_queries: List[str],
description_length: int,
repos_by_query: Dict[str, pd.DataFrame],
additional_shown_cols: List[str],
):
print("started retrieval")
if query in all_queries:
with st.expander(
"query is in gold standard set queries. Toggle viewing gold standard results?"
):
st.write("gold standard results")
task_repos = repos_by_query.get_group(query)
st.table(get_repos_with_descriptions(retrieval_pipe.X_df, task_repos))
with st.spinner(text="fetching results"):
st.write(
retrieval_pipe.search(
query, k, description_length, additional_shown_cols
).to_html(escape=False, index=False),
unsafe_allow_html=True,
)
print("finished retrieval")
def run_app(self, retrieval_pipeline):
retrieved_results = st.sidebar.number_input("number of results", value=10)
description_length = st.sidebar.number_input(
"number of used description words", value=10
)
tasks_deduped = (
self.retrieval_df["tasks"].explode().value_counts().reset_index()
) # drop_duplicates().sort_values().reset_index(drop=True)
tasks_deduped.columns = ["task", "documents per task"]
with st.sidebar.expander("View test set queries"):
st.table(tasks_deduped.explode("task"))
repos_by_query = self.retrieval_df.explode("tasks").groupby("tasks")
query = st.text_input("input query", value="metric learning")
RetrievalApp.show_retrieval_results(
retrieval_pipeline,
query,
retrieved_results,
tasks_deduped["task"].to_list(),
description_length,
repos_by_query,
self.additional_shown_cols,
)
@st.cache(allow_output_mutation=True)
def get_retrieval_pipeline(self, displayed_retrieval_df):
return RetrievalPipelineWrapper.setup_from_encoder_names(
self.query_encoder_name,
self.document_encoder_name,
displayed_retrieval_df["document"],
displayed_retrieval_df,
device=self.device,
)
def main(self):
print("setting up retrieval_pipe")
displayed_retrieval_df = merge_cols(
self.retrieval_df.copy(), self.additional_shown_cols
)
retrieval_pipeline = self.get_retrieval_pipeline(displayed_retrieval_df)
self.run_app(retrieval_pipeline)