andrea.pedrotti
/
gFun
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Implementing inference functions

This commit is contained in:
andrea 2021-01-22 16:54:56 +01:00
parent 4d3ef41a07
commit 9af9347531
2 changed files with 29 additions and 25 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
refactor/models/pl_gru.py
View File

@ -110,9 +110,13 @@ class RecurrentModel(pl.LightningModule):
def encode(self, lX, l_pad, batch_size=128): def encode(self, lX, l_pad, batch_size=128):
""" """
Returns encoded data (i.e, RNN hidden state at second feed-forward layer - linear1). Dimensionality is 512. Returns encoded data (i.e, RNN hidden state at second feed-forward layer - linear1). Dimensionality is 512.
# TODO: does not run on gpu..
:param lX: :param lX:
:param l_pad:
:param batch_size:
:return: :return:
""" """
with torch.no_grad():
l_embed = {lang: [] for lang in lX.keys()} l_embed = {lang: [] for lang in lX.keys()}
for lang in sorted(lX.keys()): for lang in sorted(lX.keys()):
for i in range(0, len(lX[lang]), batch_size): for i in range(0, len(lX[lang]), batch_size):
@ -135,7 +139,7 @@ class RecurrentModel(pl.LightningModule):
output = self.dropout(F.relu(self.linear1(output))) output = self.dropout(F.relu(self.linear1(output)))
l_embed[lang].append(output) l_embed[lang].append(output)
for k, v in l_embed.items(): for k, v in l_embed.items():
l_embed[k] = torch.cat(v, dim=0) l_embed[k] = torch.cat(v, dim=0).cpu().numpy()
return l_embed return l_embed
def training_step(self, train_batch, batch_idx): def training_step(self, train_batch, batch_idx):

2
refactor/view_generators.py
View File

@ -229,7 +229,7 @@ class RecurrentGen(ViewGen):
l_pad = self.multilingualIndex.l_pad() l_pad = self.multilingualIndex.l_pad()
data = self.multilingualIndex.l_devel_index() data = self.multilingualIndex.l_devel_index()
# trainer = Trainer(gpus=self.gpus) # trainer = Trainer(gpus=self.gpus)
# self.model.eval() self.model.eval()
time_init = time() time_init = time()
l_embeds = self.model.encode(data, l_pad, batch_size=256) l_embeds = self.model.encode(data, l_pad, batch_size=256)
transform_time = round(time() - time_init, 3) transform_time = round(time() - time_init, 3)