from fastapi import APIRouter
from datetime import datetime
from datasets import load_dataset
from sklearn.metrics import accuracy_score
import random

from .utils.evaluation import TextEvaluationRequest
from .utils.emissions import tracker, clean_emissions_data, get_space_info

## add-on imports
import numpy as np

# Logistic REG reqs
from sentence_transformers import SentenceTransformer
from sklearn.preprocessing import MinMaxScaler
import skops.io as sio

# BERT reqs
from transformers import AutoTokenizer,BertForSequenceClassification,AutoModelForSequenceClassification,Trainer, TrainingArguments,DataCollatorWithPadding
from datasets import Dataset
import torch

router = APIRouter()

DESCRIPTION = "Simple BERT classif"
ROUTE = "/text"

@router.post(ROUTE, tags=["Text Task"], 
             description=DESCRIPTION)
async def evaluate_text(request: TextEvaluationRequest):
    """
    Evaluate text classification for climate disinformation detection.
    
    Current Model: Random Baseline
    - Makes random predictions from the label space (0-7)
    - Used as a baseline for comparison
    """
    # Get space info
    username, space_url = get_space_info()

    # Define the label mapping
    LABEL_MAPPING = {
        "0_not_relevant": 0,
        "1_not_happening": 1,
        "2_not_human": 2,
        "3_not_bad": 3,
        "4_solutions_harmful_unnecessary": 4,
        "5_science_unreliable": 5,
        "6_proponents_biased": 6,
        "7_fossil_fuels_needed": 7
    }

    # Load and prepare the dataset
    dataset = load_dataset(request.dataset_name)

    # Convert string labels to integers
    dataset = dataset.map(lambda x: {"label": LABEL_MAPPING[x["label"]]})

    # Split dataset
    train_test = dataset["train"].train_test_split(test_size=request.test_size, seed=request.test_seed)
    test_dataset = train_test["test"]
    
    # Start tracking emissions
    tracker.start()
    tracker.start_task("inference")

    #--------------------------------------------------------------------------------------------
    # YOUR MODEL INFERENCE CODE HERE
    # Update the code below to replace the random baseline by your model inference within the inference pass where the energy consumption and emissions are tracked.
    #--------------------------------------------------------------------------------------------   
    
    ######################## LOG REG  ########################
    # ## Models loading
    # # Embedding model
    # query_prompt_name = "s2s_query"
    # model = SentenceTransformer("dunzhang/stella_en_400M_v5",trust_remote_code=True).cuda()

    # # Pre-trained Logistic Regression model
    # trusted_types = ['sklearn.feature_selection._univariate_selection.f_classif']
    # disp = sio.load('./tasks/logistic_regression_model.skops',trusted=trusted_types)
    
    # ## Data prep
    # embeddings = model.encode(list(test_dataset['quote']), prompt_name=query_prompt_name)
    # scaler = MinMaxScaler()
    # X_scaled = scaler.fit_transform(embeddings)
    
    # ## Predictions
    # predictions = disp.predict(X_scaled)


    ######################## BERT  ########################
    ## Model loading
    model = BertForSequenceClassification.from_pretrained("Oriaz/climate_change_bert_classif")
    tokenizer = AutoTokenizer.from_pretrained("Oriaz/climate_change_bert_classif")

    ## Data prep
    def preprocess_function(df):
        return tokenizer(df["quote"], truncation=True)
    tokenized_test = test_dataset.map(preprocess_function, batched=True)

    ## Modify inference model
    training_args = torch.load("./tasks/utils/training_args.bin")
    training_args.eval_strategy='no'

    trainer = Trainer(
        model=model,
        args=training_args,
        tokenizer=tokenizer
    )

    ## prediction
    preds = trainer.predict(tokenized_test)
    predictions = np.array([np.argmax(x) for x in preds[0]])

    
    #--------------------------------------------------------------------------------------------
    # YOUR MODEL INFERENCE STOPS HERE
    #--------------------------------------------------------------------------------------------   

    # Stop tracking emissions
    emissions_data = tracker.stop_task()
    
    # Calculate accuracy
    true_labels = test_dataset["label"]
    accuracy = accuracy_score(true_labels, predictions)
    
    # Prepare results dictionary
    results = {
        "username": username,
        "space_url": space_url,
        "submission_timestamp": datetime.now().isoformat(),
        "model_description": DESCRIPTION,
        "accuracy": float(accuracy),
        "energy_consumed_wh": emissions_data.energy_consumed * 1000,
        "emissions_gco2eq": emissions_data.emissions * 1000,
        "emissions_data": clean_emissions_data(emissions_data),
        "api_route": ROUTE,
        "dataset_config": {
            "dataset_name": request.dataset_name,
            "test_size": request.test_size,
            "test_seed": request.test_seed
        }
    }
    
    return results