File size: 3,692 Bytes
b02596a 0fd593d b02596a 0fd593d b02596a 9e07383 b02596a aab011e b02596a |
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 94 95 96 97 98 |
import gradio as gr
import onnxruntime as ort
import numpy as np
from tokenizers import Tokenizer
from huggingface_hub import hf_hub_download
# Configuration parameters
MAX_LEN = 256
# Hardcoded description value
DESCRIPTION_TEXT = (
"I am raising this case to report a severe and ongoing issue of vermin infestation, "
"specifically rats and mice, in my residential area. The problem appears to be directly linked "
)
# Define possible choices for each field
status_choices = ["Assess & Assign", "Generate Letter", "Site Inspection"]
category_choices = ["Litter and Nuisance"]
request_reason_choices = ["Nuisance"]
request_sub_reason_choices = ["Animals"]
additional_reason_choices = ["Vermin, Rats and Mice", "Dog", "Cat", "Horse"]
notification_method_choices = ["No Notification", "Email", "Phone"]
inspection_performed_choices = ["Yes", "No"]
letter_sent_choices = ["Yes", "No"]
# Download the ONNX model and tokenizer from Hugging Face Hub using the correct file name "nba.onnx"
onnx_model_path = hf_hub_download(
repo_id="iimran/Case-Next-Best-Action-Classifier", filename="nba.onnx"
)
tokenizer_path = hf_hub_download(
repo_id="iimran/Case-Next-Best-Action-Classifier", filename="train_bpe_tokenizer.json"
)
# Load the tokenizer and ONNX model once outside the function for efficiency.
tokenizer = Tokenizer.from_file(tokenizer_path)
session = ort.InferenceSession(onnx_model_path)
input_name = session.get_inputs()[0].name
output_name = session.get_outputs()[0].name
def predict_action(status, category, request_reason, request_sub_reason,
additional_reason, notification_method, inspection_performed, letter_sent):
# Combine fields into one input string.
fields = [
status,
category,
request_reason,
request_sub_reason,
additional_reason,
notification_method,
DESCRIPTION_TEXT,
inspection_performed,
letter_sent
]
sample_text = " ".join(fields)
# Tokenize and pad the input
encoding = tokenizer.encode(sample_text)
ids = encoding.ids[:MAX_LEN]
padding = [0] * (MAX_LEN - len(ids))
input_ids = np.array([ids + padding], dtype=np.int64)
# Run inference
outputs = session.run([output_name], {input_name: input_ids})
predicted_class = np.argmax(outputs[0], axis=1)[0]
# Map predicted index to the actual action labels
label_names = [
"Assign Case Officer",
"Generate Letter and Send By Post",
"Generate Letter and Send Email",
"Generate Letter and Send SMS",
"Schedule Inspection",
"Send Feedback Survey"
]
predicted_label = label_names[predicted_class]
return predicted_label
# Create the Gradio Interface using the updated API.
demo = gr.Interface(
fn=predict_action,
inputs=[
gr.Dropdown(choices=status_choices, label="Status"),
gr.Dropdown(choices=category_choices, label="Category"),
gr.Dropdown(choices=request_reason_choices, label="Request Reason"),
gr.Dropdown(choices=request_sub_reason_choices, label="Request Sub Reason"),
gr.Dropdown(choices=additional_reason_choices, label="Additional Reason"),
gr.Dropdown(choices=notification_method_choices, label="Notification Method"),
gr.Dropdown(choices=inspection_performed_choices, label="Inspection Performed", value="No"),
gr.Dropdown(choices=letter_sent_choices, label="Letter Sent", value="No")
],
outputs=gr.Textbox(label="Predicted Action"),
title="Council - Case Next Best Action Predictor",
description="Select values from the dropdowns. The description field is fixed."
)
demo.launch()
|