import json
from typing import Dict, Any

from transformers import pipeline
from ai.sarvam_client import generate_response, extract_json


emotion_classifier = pipeline(
    "text-classification",
    model="j-hartmann/emotion-english-distilroberta-base",
    top_k=None
)


def get_emotion_scores(text: str):

    predictions = emotion_classifier(text)[0]

    scores = {}

    for item in predictions:
        scores[item["label"]] = round(item["score"], 4)

    return scores


def create_emotion_prompt(text: str):

    scores = get_emotion_scores(text)

    return f"""
Analyze this news article for emotional manipulation.

Emotion scores:
{scores}

Article:
{text}

Return ONLY this JSON.

{{
"fear_detected":false,
"urgency_detected":false,
"outrage_detected":false,
"sensationalism_detected":false,
"emotion_intensity":0.0,
"explanation":""
}}

Output ONLY JSON.
Do NOT explain.
Do NOT use markdown.
"""


def get_emotion_from_model(text):

    return generate_response(
        create_emotion_prompt(text)
    )


def parse_model_response(response):

    default = {

        "fear_detected": False,
        "urgency_detected": False,
        "outrage_detected": False,
        "sensationalism_detected": False,
        "emotion_intensity": 0.0,
        "explanation": "No response received from Sarvam AI."

    }

    if response is None:
        return default

    parsed = extract_json(response)

    if parsed is None:
        return {
            **default,
            "explanation": "Could not parse model response.",
        }

    for key in default:
        parsed.setdefault(key, default[key])

    return parsed


def analyze_emotion(text):

    if not text.strip():

        return {

            "fear_detected": False,
            "urgency_detected": False,
            "outrage_detected": False,
            "sensationalism_detected": False,
            "emotion_intensity": 0.0,
            "explanation": "Empty input."

        }

    raw = get_emotion_from_model(text)

    return parse_model_response(raw)