def test_on_batch(self, batch_ind, batch):
logSoftmax = nn.LogSoftmax()
outdir = os.path.join(self.output_dir, 'batch%04d' % batch_ind)
os.makedirs(outdir, exist_ok=True)
self.net_q.eval()
self.net_c.eval()
# read in X
X_list = [None] * self.n_modality
for i in range(self.n_modality):
X_list[i] = batch[self.modality[i]]
X_list[i] = X_list[i].view(X_list[i].shape[0], -1)
X_list = [tmp.cuda() for tmp in X_list]
# Concatenate X before feeding into Autoencoder
X = torch.cat(X_list, dim=1).cuda()
y = batch['labels'].cuda()
with torch.no_grad():
z = self.net_q(X)
pred_logits = self.net_c(z)
pred_prob = logSoftmax(pred_logits)
eva = reportMetricsMultiClass(y, pred_prob)
formatTable(eva, outpath=os.path.join(outdir, 'eva.csv'))
output = self.pack_output(pred_prob, batch)
np.savez(os.path.join(outdir, 'batch%04d.npz' % batch_ind), **output)
def test_on_batch(self, batch_ind, batch):
logSoftmax = nn.LogSoftmax()
outdir = os.path.join(self.output_dir, 'batch%04d' % batch_ind)
os.makedirs(outdir, exist_ok=True)
for i in range(self.n_modality):
self.net_q[i].eval()
self.net_c.eval()
X_list = [None] * self.n_modality
z_list = [None] * self.n_modality
# read in training data
for i in range(self.n_modality):
X_list[i] = batch[self.modality[i]]
X_list[i] = X_list[i].view(X_list[i].shape[0], -1)
X_list = [tmp.cuda() for tmp in X_list]
y = batch['labels'].cuda()
with torch.no_grad():
for i in range(self.n_modality):
z_list[i] = self.net_q[i](X_list[i])
z_combined = torch.cat(z_list, dim=1)
pred_logits = self.net_c(z_combined)
pred_prob = logSoftmax(pred_logits)
eva = reportMetricsMultiClass(y, pred_prob)
formatTable(eva, outpath=os.path.join(outdir, 'eva.csv'))
output = self.pack_output(pred_prob, batch, z_list)
np.savez(os.path.join(outdir, 'batch%04d.npz' % batch_ind), **output)
def test_on_batch(self, batch_ind, batch):
logSoftmax = nn.LogSoftmax()
outdir = os.path.join(self.output_dir, 'batch%04d' % batch_ind)
os.makedirs(outdir, exist_ok=True)
self.net_q.eval()
self.net_c.eval()
X = batch[self.modality].cuda()
X = X.view(X.shape[0], -1)
y = batch['labels'].cuda()
with torch.no_grad():
z = self.net_q(X)
pred_logits = self.net_c(z)
pred_prob = logSoftmax(pred_logits)
eva = reportMetricsMultiClass(y, pred_prob)
formatTable(eva, outpath=os.path.join(outdir, 'eva.csv'))
output = self.pack_output(pred_prob, batch)
np.savez(os.path.join(outdir, 'batch%04d.npz' % batch_ind), **output)