from dataclasses import dataclass, field

from os import getcwd, pardir

from os.path import join

from typing import Dict, Any, Tuple, Optional

import numpy as np

from openpyxl.compat.singleton import Singleton

@dataclass

class Config(metaclass=Singleton):

##############

# PREPROCESS #

##############

N_FFT: int = 2 ** 8

HOP_LENGTH: int = 2 ** 7

DURATION: float = 1.0

SAMPLE_RATE: int = 16_000

EPS: float = 1e-10

SPECTROGRAMS_SAVE_DIR: str = "../datasets/one_shot_percussive_sounds/spectrograms_db/"

FILES_DIR: str = "../datasets/one_shot_percussive_sounds/audio/"

TOP_DB: float = 80.

NORM_RANGE: Tuple[float, float] = (0., 1.)

DB_RANGE: Tuple[float, float] = (-TOP_DB, 0.)

SPECTROGRAM_SHAPE: Tuple[int, int, int] = (128, 128, 1)

EMPTY_SPEC: np.ndarray = np.zeros(SPECTROGRAM_SHAPE)

###############

# AUTOENCODER #

###############

IS_VARIATIONAL: bool = True

RECONSTRUCTION_LOSS_WEIGHT: float = 1e-7

NUM_HIDDEN_CONV_LAYERS: int = 7

FILTER_SIZE_CONV_LAYERS: int = 4

LATENT_SPACE_DIM: int = 4

CONV_PARAMS_MIDDLE: Dict[str, Any] = field(default_factory=lambda: {

'kernel_size': (6, 3),

'strides': (2, 1),

'padding': 'same',

'activation': 'elu',

'kernel_initializer': 'glorot_normal',

})

CONV_PARAMS_END: Dict[str, Any] = field(default_factory=lambda: {

'kernel_size': 6,

'strides': 1,

'padding': 'same',

'activation': 'softplus',

'kernel_initializer': 'glorot_normal',

})

USE_MOCK_ENCODER: bool = False

USE_MOCK_DECODER: bool = False

ARCH_PLOT_ARGS: Dict[str, bool] = field(default_factory=lambda: {

'show_layer_activations': True,

'show_shapes': True,

})

LOGS_DIR: str = join(getcwd(), pardir, "logs")

CLIP_PREDICTION: bool = True

LOSS_SQUARED: bool = False

LOSS_MSE_LINEAR_WEIGHT: Optional[Tuple[float, float]] = (2., 1.)

USE_LSTM: bool = True

#########

# TRAIN #

#########

VERBOSE: bool = True

LEARNING_RATE: float = 1e-3

BATCH_SIZE: int = 2 ** 6

PRETRAIN_EPOCHS: int = 500

EPOCHS: int = 2000

SPECTROGRAMS_PATH: str = "../datasets/one_shot_percussive_sounds/spectrograms_db/"

PRETRAIN_DATA: np.ndarray = \

np.tile(np.linspace(NORM_RANGE[1], NORM_RANGE[0], SPECTROGRAM_SHAPE[0])[..., None], SPECTROGRAM_SHAPE[1]).T[None, ..., None]

############

# GENERATE #

############

SAVE_DIR_ORIGINAL: str = "../output/samples/original/"

SAVE_DIR_GENERATED: str = "../output/samples/generated/"

NORMALIZATION_FACTOR: float = 2 ** 16

cfg = Config()