model.py

import pandas as pd
from datetime import datetime
from transformers import pipeline

# --- Constants ---
ALERT_THRESHOLD = 60  # Threshold for flagging low-performing vendors
DAYS_PER_MONTH = 30
# --- Helper Functions ---

def calculate_quality_score(incident_logs):
    """
    Calculates a quality score based on the number and severity of incident logs.

    Args:
        incident_logs (str): A string containing incident log details.

    Returns:
        float: A score between 0 and 100, where 100 is the highest quality.
    """
    if not incident_logs:
        return 100  # Perfect score if no incidents

    # Basic keyword matching for severity (can be expanded)
    high_severity_keywords = ['major', 'critical', 'severe', 'fatality']
    medium_severity_keywords = ['minor', 'moderate', 'injury']
    low_severity_keywords = ['near miss', 'warning', 'caution']

    high_count = sum(1 for keyword in high_severity_keywords if keyword in incident_logs.lower())
    medium_count = sum(1 for keyword in medium_severity_keywords if keyword in incident_logs.lower())
    low_count = sum(1 for keyword in low_severity_keywords if keyword in incident_logs.lower())

    # Weighted scoring (adjust weights as needed)
    score = 100 - (high_count * 20 + medium_count * 10 + low_count * 5)
    return max(0, score)  # Ensure score doesn't go below 0

def calculate_timeliness_score(work_completion_details, delay_reports, log_date):
    """
    Calculates a timeliness score based on work completion details, delay reports,
    and the log date.

    Args:
        work_completion_details (str): Details of work completion.
        delay_reports (str): Reports of delays.
        log_date (str): The date of the log (YYYY-MM-DD).

    Returns:
        float: A score between 0 and 100, where 100 is perfectly on time.
    """
    if not work_completion_details:
        return 100

    log_date_obj = datetime.strptime(log_date, '%Y-%m-%d')
    # Assume a 30-day window for "on time" (can be adjusted)
    completion_window_end = log_date_obj

    # Check for explicit "on time" completion
    if "on time" in work_completion_details.lower():
        return 100

    # Penalize for delay reports
    delay_penalty = 0
    if delay_reports:
        delay_penalty = len(delay_reports.split(',')) * 15  # 15 points per delay report (adjust as needed)

    # Very basic check for "late" or "delayed"
    if "late" in work_completion_details.lower() or "delayed" in work_completion_details.lower():
      return max(0, 50 - delay_penalty)

    return max(0, 100 - delay_penalty) # cap at 100

def calculate_safety_score(incident_logs):
    """
    Calculates a safety score based on the presence of incident logs.

    Args:
        incident_logs (str): A string containing incident log details.

    Returns:
        float: 100 if no incidents, otherwise a lower score.
    """
    if not incident_logs:
        return 100
    else:
        # Further logic can be added to differentiate severity
        return max(0, 80 - len(incident_logs.split(',')) * 10)  # Reduce score per incident

def calculate_communication_score(work_completion_details):
    """
    Calculates a communication score based on the work completion details.
     Uses a simple sentiment analysis.

    Args:
        work_completion_details (str): Details of work completion.

    Returns:
        float: A score between 0 and 100.
    """
    if not work_completion_details:
        return 100

    # Initialize sentiment analysis pipeline
    sentiment_analyzer = pipeline("sentiment-analysis-ssbert-large-en") # More robust model

    try:
        result = sentiment_analyzer(work_completion_details)
        sentiment = result[0]['label']  # Get the sentiment label
        confidence = result[0]['score']

        if sentiment == 'POSITIVE':
            return 100
        elif sentiment == 'NEGATIVE':
            return max(0, 60 * confidence)  # Scale the negative impact by confidence
        else:  # NEUTRAL
            return 80
    except Exception as e:
        print(f"Error in sentiment analysis: {e}")
        return 80  # Return a neutral score on error

def calculate_final_score(quality_score, timeliness_score, safety_score, communication_score):
    """
    Calculates a final score based on weighted averages of the individual scores.

    Args:
        quality_score (float): The quality score.
        timeliness_score (float): The timeliness score.
        safety_score (float): The safety score.
        communication_score (float): The communication score.

    Returns:
        float: The final score, between 0 and 100.
    """
    # Weights (can be adjusted)
    quality_weight = 0.4
    timeliness_weight = 0.3
    safety_weight = 0.2
    communication_weight = 0.1

    final_score = (
        quality_weight * quality_score +
        timeliness_weight * timeliness_score +
        safety_weight * safety_score +
        communication_weight * communication_score
    )
    return final_score

def generate_performance_report(vendor_id, scores, month, trend_data=None):
    """
    Generates a performance report (as a dictionary).  Includes a placeholder for
    certificate generation.

    Args:
        vendor_id (str): The ID of the vendor.
        scores (dict): A dictionary containing the vendor's scores.
        month (str): The month for the report (e.g., "2024-01").
        trend_data (dict, optional):  Trend data for the vendor.  Defaults to None.

    Returns:
        dict: A dictionary containing the performance report.
    """
    report = {
        'vendor_id': vendor_id,
        'month': month,
        'quality': scores['quality'],
        'timeliness': scores['timeliness'],
        'safety': scores['safety'],
        'communication': scores['communication'],
        'final_score': scores['final_score'],
        'alert_flag': scores['final_score'] < ALERT_THRESHOLD,
        'certificate_url': f"/certificates/{vendor_id}_{month}.pdf",  # Placeholder URL
    }
    if trend_data:
        report['trend_deviation'] = trend_data.get('trend_deviation', 0)
    else:
        report['trend_deviation'] = 0
    return report

def process_vendor_logs(vendor_logs):
    """
    Processes a list of vendor logs, calculates scores, and generates performance reports.

    Args:
        vendor_logs (list): A list of dictionaries, where each dictionary represents
            a vendor log and contains the keys 'vendor_id',
            'work_completion_details', 'delay_reports', 'incident_logs', and 'log_date'.

    Returns:
        list: A list of performance report dictionaries, ready for Salesforce.
    """
    reports = []
    for log in vendor_logs:
        try:
            vendor_id = log['vendor_id']
            work_completion_details = log['work_completion_details']
            delay_reports = log['delay_reports']
            incident_logs = log['incident_logs']
            log_date = log['log_date']  # Assuming YYYY-MM-DD format

            quality_score = calculate_quality_score(incident_logs)
            timeliness_score = calculate_timeliness_score(work_completion_details, delay_reports, log_date)
            safety_score = calculate_safety_score(incident_logs)
            communication_score = calculate_communication_score(work_completion_details)
            final_score = calculate_final_score(quality_score, timeliness_score, safety_score, communication_score)

            scores = {
                'quality': quality_score,
                'timeliness': timeliness_score,
                'safety': safety_score,
                'communication': communication_score,
                'final_score': final_score,
            }
            #  Basic Trend Detection (Example)
            #  In a real scenario, you'd fetch previous months' scores from Salesforce
            #  and calculate a trend.  This is a placeholder.
            trend_data = None
            #  Placeholder logic:  If current score is more than 10 points lower
            #  than a hypothetical previous month, we have a negative trend.
            # previous_month_score = get_previous_month_score(vendor_id, log_date) #from salesforce
            # if previous_month_score and (final_score < previous_month_score - 10):
            #     trend_data = {'trend_deviation': -1}  #  Negative trend
            # elif previous_month_score and (final_score > previous_month_score + 10):
            #      trend_data = {'trend_deviation': 1}
            # else:
            #     trend_data = {'trend_deviation': 0}
            report = generate_performance_report(vendor_id, scores, log_date[:7], trend_data) # Use YYYY-MM
            reports.append(report)
        except Exception as e:
            print(f"Error processing log for vendor {log.get('vendor_id', 'Unknown')}: {e}")
            #  Consider logging the error to a file or database for further analysis
            #  You might also want to raise the exception if it's critical
            #  to stop processing.  For now, we'll just continue to the next log.
    return reports