app.py

# app.py

import gradio as gr
import pandas as pd
import numpy as np
from transformers import pipeline, AutoTokenizer, AutoModelForSequenceClassification
from sklearn.ensemble import RandomForestClassifier
import joblib

# Using the Hugging Face model "huggingface-course/distilbert-base-uncased-finetuned-imdb"
# Load the tokenizer and model separately
tokenizer = AutoTokenizer.from_pretrained("huggingface-course/distilbert-base-uncased-finetuned-imdb")
model = AutoModelForSequenceClassification.from_pretrained("huggingface-course/distilbert-base-uncased-finetuned-imdb")

# Define pipeline for anomaly detection using the loaded model and tokenizer
anomaly_detection = pipeline("text-classification", model=model, tokenizer=tokenizer)

# Load the Random Forest model for failure prediction (this should be trained and saved as failure_prediction_model.pkl)
failure_prediction_model = joblib.load('failure_prediction_model.pkl')

# Function to preprocess logs for anomaly detection
def preprocess_logs(logs):
    logs['timestamp'] = pd.to_datetime(logs['timestamp'])
    logs['log_message'] = logs['log_message'].str.lower()
    return logs

# Function to detect anomalies
def detect_anomaly(logs):
    preprocessed_logs = preprocess_logs(logs)
    results = []
    for log in preprocessed_logs['log_message']:
        anomaly_result = anomaly_detection(log)
        results.append(anomaly_result[0]['label'])  # Assuming the Hugging Face model outputs label as result
    return results

# Function to predict failures based on historical data and metrics
def predict_failure(device_metrics):
    metrics_array = np.array([device_metrics['cpu_usage'], device_metrics['memory_usage'], device_metrics['error_rate']]).reshape(1, -1)
    failure_prediction = failure_prediction_model.predict(metrics_array)
    return failure_prediction

# Gradio interface to upload log files and check anomaly detection and failure prediction
def process_logs_and_predict(log_file, metrics):
    logs = pd.read_json(log_file)
    anomalies = detect_anomaly(logs)
    failure_pred = predict_failure(metrics)
    
    return f"Anomalies Detected: {anomalies}, Failure Prediction: {failure_pred}"

# Set up Gradio interface for the dashboard
iface = gr.Interface(fn=process_logs_and_predict, 
                     inputs=["file", "json"], 
                     outputs="text", 
                     title="Cisco Device Monitoring",
                     description="Upload log files to detect anomalies and predict potential device failures.")

iface.launch()