File size: 3,659 Bytes
cb3a755 |
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 |
import gradio as gr
import torch
import joblib
import numpy as np
import pandas as pd
from transformers import AutoTokenizer, AutoModel
# Load IndoBERT tokenizer
tokenizer = AutoTokenizer.from_pretrained("indolem/indobert-base-uncased")
# Load IndoBERT model
model = AutoModel.from_pretrained("indolem/indobert-base-uncased")
# Mapping dictionaries for labels
priority_score_mapping = {1: "LOW", 2: "MEDIUM", 3: "HIGH"}
problem_domain_mapping = {0: "OPERATIONAL", 1: "TECHNICAL"}
# Load the trained Random Forest models
best_classifier1 = joblib.load('best_classifier1_optimized.pkl')
best_classifier2 = joblib.load('best_classifier2_optimized.pkl')
markdown_text = '''
## Label Description
### Priority Score
* **Low** label, means that the temporary/corrective solution can solve the problem. A permanent solution will be provided later because the impact on the business can still be handled.
* **Medium** label, means that there's a need to determine the time constraint to solve the problem. If it remains too long, it will impact the business side.
* **High** label, means that the problem is urgent and must be solved immediately.
### Problem Domain
* **Operational** label, means that the scope of the problem is on the business or daily operational.
* **Technical** label, means that the scope of the problem is on the technical (technology) side like the mobile/web application.
'''
description="Write the feedback about the capsule hotel that you've ever visited or stayed there. The machine learning model will predict the priority score and problem domain of the feedback."
# Function to perform predictions
def predict(text):
# Convert the sentences into input features
encoded_inputs = tokenizer(text, padding=True, truncation=True, return_tensors="pt", max_length=128)
# Perform word embedding using IndoBERT model
with torch.no_grad():
outputs = model(**encoded_inputs)
embeddings = outputs.last_hidden_state
# Convert the embeddings to numpy array
embeddings = embeddings.numpy()
embeddings_custom_flat = embeddings.reshape(embeddings.shape[0], -1)
# Ensure mean_pooled_embeddings has exactly 768 features
num_features_expected = 768
if embeddings_custom_flat.shape[1] < num_features_expected:
# If the number of features is less than 768, pad the embeddings
pad_width = num_features_expected - embeddings_custom_flat.shape[1]
embeddings_custom_flat = np.pad(embeddings_custom_flat, ((0, 0), (0, pad_width)), mode='constant')
elif embeddings_custom_flat.shape[1] > num_features_expected:
# If the number of features is more than 768, truncate the embeddings
embeddings_custom_flat = embeddings_custom_flat[:, :num_features_expected]
# Predict the priority_score for the custom input
custom_priority_score = best_classifier1.predict(embeddings_custom_flat)
# Predict the problem_domain for the custom input
custom_problem_domain = best_classifier2.predict(embeddings_custom_flat)
# Map numerical labels to human-readable labels
mapped_priority_score = priority_score_mapping.get(custom_priority_score[0], "unknown")
mapped_problem_domain = problem_domain_mapping.get(custom_problem_domain[0], "unknown")
return f"Predicted Priority Score: {mapped_priority_score}, Predicted Problem Domain: {mapped_problem_domain}"
# Create a Gradio interface
gr.Interface(fn=predict, inputs="text", outputs="text", title="Simple Risk Classifier Demo (Case Study: Capsule Hotel)", description=description, article=markdown_text).launch(debug=True)
|