"""Convolutional autoencoder embedding.
Importing this module maps torch -- and so torch's copy of ``libomp`` -- into the
process. On macOS that must not happen before xgboost's copy is initialised or
XGBoost model fitting segfaults; see ``qbiocode.utils._openmp`` for the full
diagnosis. The preload below makes this module safe to import on its own, in any
order, rather than relying on the caller having imported ``qbiocode`` first.
"""
from qbiocode.utils._openmp import preload_openmp_libraries
preload_openmp_libraries()
import torch
import torch.nn as nn
import torch.optim as optim
# Define the Autoencoder Model
[docs]
class ConvAutoencoder(nn.Module):
def __init__(self):
super(ConvAutoencoder, self).__init__()
# Encoder
self.encoder = nn.Sequential(
nn.Conv2d(7, 64, kernel_size=3, stride=2, padding=1), # (64, 192, 192)
nn.ReLU(),
nn.Conv2d(64, 128, kernel_size=3, stride=2, padding=1), # (128, 96, 96)
nn.ReLU(),
nn.Conv2d(128, 256, kernel_size=3, stride=2, padding=1), # (256, 48, 48)
nn.ReLU(),
nn.Conv2d(256, 512, kernel_size=3, stride=2, padding=1), # (512, 24, 24)
nn.ReLU(),
nn.Conv2d(512, 7, kernel_size=3, stride=2, padding=1), # (7, 16, 16)
nn.ReLU(),
)
# Decoder
self.decoder = nn.Sequential(
nn.ConvTranspose2d(
7, 512, kernel_size=3, stride=2, padding=1, output_padding=1
), # (512, 24, 24)
nn.ReLU(),
nn.ConvTranspose2d(
512, 256, kernel_size=3, stride=2, padding=1, output_padding=1
), # (256, 48, 48)
nn.ReLU(),
nn.ConvTranspose2d(
256, 128, kernel_size=3, stride=2, padding=1, output_padding=1
), # (128, 96, 96)
nn.ReLU(),
nn.ConvTranspose2d(
128, 64, kernel_size=3, stride=2, padding=1, output_padding=1
), # (64, 192, 192)
nn.ReLU(),
nn.ConvTranspose2d(
64, 7, kernel_size=3, stride=2, padding=1, output_padding=1
), # (7, 384, 384)
nn.Sigmoid(),
)
[docs]
def forward(self, x):
latent = self.encoder(x)
reconstructed = self.decoder(latent)
return reconstructed
