PPGVAE_ModelArgs.py

import torch
from dataclasses import dataclass
from typing import Callable
#from PPG_Gauss_Helperfunctions import mae, unit_0_1

device = (
    "cuda"
    if torch.cuda.is_available()
    else "mps"
    if torch.backends.mps.is_available()
    else "cpu"
)


@dataclass
class ModelArgs:
    γ_discriminator: float = 50.0
    λₖₚ: float = 1                  # regularization paramater for kp should be ~0.1 as we are using MAE
    λᵣₑ: float = 0.1                # regularization paramater for reg
    toprint_plots: int = 400
    epochs: int = 200           # number of epochs
    input_dim: int = None          # image size
    latent_dim: int = 4          # latent dimension
    hidden_dim: int = 32         # hidden dimension
    use_dense: bool = True
    dy_scale: float = 20  #default was 20 which is about +/-20* angled slope max, changing it to 5* and seeing what happens
    shift_left_pad: int = 3
    shift_right_pad: int = 3

    use_slope:bool =  True
    use_slope_latdim_in_klloss: bool = False
    use_slope_in_discrim: bool = False
    
    use_outscaling: bool = True# nothing
    device = device
    
    nn_modeltype = "PPGVAE"
    theta = 2/90
    
    def get_config(self):
        config = dict() # super(self).get_config().copy()
        config.update({
            'γ_discriminator': self.γ_discriminator,
            'λₖₚ': self.λₖₚ,
            'λᵣₑ': self.λᵣₑ,
            'λₘ': self.λₘ,
            'toprint_plots': self.toprint_plots,
            'epochs': self.epochs,
            'input_dim': self.input_dim,
            'latent_dim': self.latent_dim,
            'max_channels': self.max_channels,
            'hidden_dim': self.hidden_dim,
            'use_dense': self.use_dense,
            'shift_left_pad': self.shift_left_pad,
            'shift_right_pad': self.shift_right_pad,
            'use_slope_latdim_in_klloss': self.use_slope_latdim_in_klloss,
            'use_slope_in_discrim': self.use_slope_in_discrim,
            'use_outscaling': self.use_outscaling,
            'nn_modeltype': self.nn_modeltype,
            'theta': self.theta,
        })
        return config



if __name__ == "__main__":
    a = ModelArgs()