from fastapi import APIRouter
from datetime import datetime
from datasets import load_dataset
from sklearn.metrics import accuracy_score
import os
import torch
from torch.utils.data import DataLoader, TensorDataset
from torchaudio import transforms
#from torchvision import models
#import onnxruntime as ort  # Add ONNX Runtime
from openvino.runtime import Core
import numpy as np
from .utils.evaluation import AudioEvaluationRequest
from .utils.emissions import tracker, clean_emissions_data, get_space_info

from dotenv import load_dotenv
load_dotenv()

router = APIRouter()

DESCRIPTION = "Tiny_DNN_new_split_remove_empty_audio"
ROUTE = "/audio"

torch.set_num_threads(4)
torch.set_num_interop_threads(2)

@router.post(ROUTE, tags=["Audio Task"], description=DESCRIPTION)
async def evaluate_audio(request: AudioEvaluationRequest):
    # Get space info
    username, space_url = get_space_info()

    # Define the label mapping
    LABEL_MAPPING = {
        "chainsaw": 0,
        "environment": 1
    }

    # Load and prepare the dataset
    dataset = load_dataset(request.dataset_name, token=os.getenv("HF_TOKEN"))
    
    ## old split
    # train_test = dataset["train"].train_test_split(test_size=request.test_size, seed=request.test_seed)
    # test_dataset = train_test["test"]

    ## new split
    # Split dataset
    test_dataset = dataset["test"]
    
    # find a way to clean datas (empty arrays) by remplace or remove or nothing
    way = "remove" # "remove" or "remplace" or "nothing"
    if way == "remplace": 
        def replace_empty_audio_array(batch): # fill empty samples like mashpi_2020_8d4d84bd-3b24-4af1-82e2-168b678774f3_64-67.wav with one 0
            for audio in batch["audio"]:
                if audio["array"].shape[0] == 0:
                    audio["array"] = np.array([0])
            return batch
        test_dataset = test_dataset.map(replace_empty_audio_array, batched=True)
    elif way == "remove": 
        def audio_array_is_not_empty(sample): # remove empty samples like mashpi_2020_8d4d84bd-3b24-4af1-82e2-168b678774f3_64-67.wav
            return sample["audio"]["array"].shape[0] != 0
        test_dataset = test_dataset.filter(audio_array_is_not_empty)
    else:
        print("no clean method to clean datas")

    true_labels = test_dataset["label"]

    resampler = transforms.Resample(orig_freq=12000, new_freq=16000)
    mel_transform = transforms.MelSpectrogram(sample_rate=16000, n_mels=64)
    amplitude_to_db = transforms.AmplitudeToDB()

    def resize_audio(_waveform, target_length):
        num_frames = _waveform.shape[-1]
        if num_frames != target_length:
            _resampler = transforms.Resample(orig_freq=num_frames, new_freq=target_length)
            _waveform = _resampler(_waveform)
        return _waveform

    resized_waveforms = [
        resize_audio(torch.tensor(sample['audio']['array'], dtype=torch.float32).unsqueeze(0), target_length=72000)
        for sample in test_dataset
    ]

    waveforms, labels = [], []
    for waveform, label in zip(resized_waveforms, true_labels):
        waveforms.append(amplitude_to_db(mel_transform(resampler(waveform))))
        labels.append(label)

    waveforms = torch.stack(waveforms)
    labels = torch.tensor(labels)

    test_loader = DataLoader(
        TensorDataset(waveforms, labels),
        batch_size=128,
        shuffle=False
        #pin_memory=True,
        #num_workers=4
    )

    # Load Openvino model
    core = Core()
    model_path = "./openvino_model/model.xml"
    compiled_model = core.compile_model(model=model_path, device_name="CPU")
    input_layer = compiled_model.input(0)
    output_layer = compiled_model.output(0)
    # Start tracking emissions
    tracker.start()
    tracker.start_task("inference")

    # Openvino inference
    predictions = []
    for data, target in test_loader:
        inputs = data.numpy()  # Convert tensor to numpy
        inputs = inputs.reshape((-1, 1, 64, 481))
        output = compiled_model([inputs])[output_layer]
        predicted = np.argmax(output, axis=1)
        predictions.extend(predicted.tolist())

    # Stop tracking emissions
    emissions_data = tracker.stop_task()

    # Calculate accuracy
    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