DNC-tensorflow issues

New task for adding scalar values (0 or 1)

6

### Common Settings The model is trained on 2-layer feedforward controller (with hidden sizes 128 and 256 respectively) with the following set of hyperparameters: - RMSProp Optimizer with learning rate...

Zeta36

in-progress

Training is too slow on GPU (batch size = 1)

Hi, I am training the network on NVIDIA Tesla GPU with a batch size of 1. I am training for 500K iterations, it's taking huge time, around some 600 hours....

nrc53

why is ' self.lstm_cell(X, state) ' ?

I successfully run the program.However, I found threre is something seen abnormal. class RecurrentController(BaseController): def network_vars(self): self.lstm_cell = tf.nn.rnn_cell.BasicLSTMCell(256) self.state = self.lstm_cell.zero_state(self.batch_size, tf.float32) def network_op(self, X, state): X = tf.convert_to_tensor(X)...

fanyuzeng

I am trying to fallow Project Structure and Usage but complete running examples would help

I am trying to fallow Project Structure and Usage but complete running examples would help I ran the code with no output to Initial Transformation Weights that I skipped the...

josheeg

Questions about FeedforwardController

1

Hello. This is a very good code. But I am very curious about why use FeedforwardController instead of the original BaseController. Is there small specific function?

tangbohu

TypeError: 'numpy.float64' object cannot be interpreted as an index

python tasks/copy/train.py --iterations=50000 Building Computational Graph ... Traceback (most recent call last): File "tasks/copy/train.py", line 82, in batch_size File "/Users/yike.ke/yike_prj/DNC-tensorflow/tasks/copy/dnc/dnc.py", line 46, in __init__ self.controller = controller_class(self.input_size, self.output_size, self.read_heads, self.word_size,...

SeekPoint