Spaces:
Running
Running
File size: 3,741 Bytes
efecb81 8b91e86 6608d38 d3cd55f 6608d38 c9fad14 6608d38 c9fad14 6608d38 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 |
import gradio as gr
import sentence_transformers
from sentence_transformers import SentenceTransformer
import torch
from sentence_transformers.util import semantic_search
import pandas as pd
import requests
#Import corpus embeddings
corpus_ger = pd.read_json('corpus_embed_ger_model5.json')
corpus_eng = pd.read_json('corpus_embed_eng_mboth.json')
corpus_fr = pd.read_json('corpus_embed_fr_Sahajtomar.json')
#Import models
model_ger = SentenceTransformer('JoBeer/model5')
model_eng = SentenceTransformer('mboth/distil-eng-quora-sentence')
model_fr = SentenceTransformer('Sahajtomar/french_semantic')
#Definition of search function
def predict(name, description, language, classCode='nofilter', top_k=10):
#language detection
if language == 'german':
model = model_ger
corpus = corpus_ger
if language == 'english':
model = model_eng
corpus = corpus_eng
if language == 'french':
model = model_fr
corpus = corpus_fr
text = name + '; ' + description #Verkettung name und description
query_embedding = model.encode(text, convert_to_tensor=True) #Erzeugung Query Embedding
#Filterung ECLASS Corpus
if classCode == 'nofilter':
corpus_filtered = corpus
else:
url = f'https://bcon2-api.azurewebsites.net/api/eclass?codedname={classCode}'
response = requests.get(url) #http request um auf Filter API zuzugreifen
lines = response.text.split('\n')
properties_filtered_list = [line[-21:-1] for line in lines]
corpus_filtered = corpus[corpus['irdi'].isin(properties_filtered_list)] #Der ECLASS Corpus wird mit Hilfe der Liste gefiltert
corpus_filtered.reset_index(drop=True,inplace=True) #Index wird zurückgesetzt
#Umwandlung corpus Embeddings in Tensor
corpus_embeddings = torch.Tensor(corpus_filtered["Embeddings"])
#Einspeisung Modell
output = sentence_transformers.util.semantic_search(query_embedding, corpus_embeddings, top_k = top_k)
#Auslesen der Modellausgabe
preferedNames = []
definitions = []
irdis = []
scores = []
for i in range(0,top_k):
preferedNames.append(corpus_filtered.iloc[output[0][i].get('corpus_id'),1])
definitions.append(corpus_filtered.iloc[output[0][i].get('corpus_id'),2])
irdis.append(corpus_filtered.iloc[output[0][i].get('corpus_id'),0])
scores.append(output[0][i].get('score'))
predictions = pd.DataFrame({'preferedName' : preferedNames, 'irdi' : irdis,'score' : scores,'definition' : definitions})
return predictions
#gradio user interface
with gr.Blocks() as demo:
with gr.Row():
gr.Markdown("""# ECLASS-Search-Demo
This is a semantic search application that maps unknown product properties to the ECLASS standard. It is created by ECLASS e.V. in collaboration with the GART-labortory of the cologne university of applied science.""") #This demo provides a semantic search application for ECLASS features.
with gr.Row():
#inputs
name_tx = gr.Textbox(label="Name:",placeholder="Name of the Property", lines=1)
description_tx = gr.Textbox(label="Description:", placeholder="Description of the Property", lines=1)
with gr.Row():
#inputs
classcode_tx = gr.Textbox(value='nofilter', label="Filter with ECLASS ClassCode", placeholder="type nofilter", lines=1)
top_k_nu = gr.Number(value=10, label="Number of Matches")
language_drop = gr.Dropdown(["german", "english","french"], value='german', label="Select language")
#button
search = gr.Button("search")
#output
prediction_df = gr.Dataframe(headers = ['preferedName', 'irdi', 'score', 'definition'])
#Hinterlegt Search-Function für button "search"
search.click(fn=predict, inputs=[name_tx,description_tx,language_drop,classcode_tx,top_k_nu], outputs=prediction_df)
demo.launch(debug=True) |