elif mode == 'inference':
seq_logProb, seq_predictions = self.decoder.infer(
encoder_last_hidden_state=encoder_last_hidden_state,
encoder_output=encoder_outputs,
)
else:
raise KeyError('mode is not valid')
return seq_logProb, seq_predictions
if __name__ == '__main__':
import logging
logger.setLevel(logging.INFO)
from vocabulary import Vocabulary
json_file = 'data/testing_label.json'
numpy_file = 'data/testing_data/feat'
helper = Vocabulary(json_file, min_word_count=5)
input_data = Variable(torch.randn(3, 80, 4096).view(-1, 80, 4096))
encoder = EncoderRNN(input_size=4096, hidden_size=1000)
decoder = DecoderRNN(hidden_size=1000,
output_size=1700,
vocab_size=1700,
word_dim=128,
helper=helper)
return x
def parms_n(self):
model_parameters = filter(lambda p: p.requires_grad, self.parameters())
params = sum([np.prod(p.size()) for p in model_parameters])
return params
# get unique string repr for each different net
def get_name(self):
net_name = self.__class__.__name__
return "{}_{}".format(net_name, self.depth)
def summary(self):
model_parameters = filter(lambda p: p.requires_grad, self.parameters())
params = sum([np.prod(p.size()) for p in model_parameters])
print("Model Parameters")
print(self.parameters)
print("Trainable parameters: {}".format(params))
if __name__ == "__main__":
logger.setLevel(logging.DEBUG)
t = MnistCNN(depth=128)
a = Variable(torch.rand(2, 3, 32, 32))
print(t.parms_n())
print(t(a).shape)
print(t(a))