ValueError: too many values to unpack (expected 2)

[developeralgo8888]

Error occurs after the first run (50 epochs) is completed on the first training & signal files. This is the commit before the current commit. I am going to do a quick run of the current commit to see if i run into any issues.

Traceback (most recent call last):
File "UFCNN1.py", line 596, in
case_tc = train_and_predict_classification(UFCNN_TC, features=features, output_dim=output_dim, sequence_length=sequence_length, epochs=50, training_count=10, testing_count = 6 )
File "UFCNN1.py", line 511, in train_and_predict_classification
xdim, ydim = yp.shape
ValueError: too many values to unpack (expected 2)

[lukovkin]

train_and_predict_classification is yet not stable, use 'python UFCNN1.py
tradcom_simple' or in notebook accordingly.
You can try with more epochs for example.

Best regards,

Dmitry Lukovkin
Deep Gnosis
http://stocksneural.net
Tel. +7 903 750 2976
Skype: dmitry.lukovkin

On 16 March 2016 at 20:21, DeveloperAlgo [email protected] wrote:

Error occurs after the first run (50 epochs) is completed on the first
training & signal files. This the commit before the last one. I am going to
do a quick run of the current commit to see if i run into any issues.

Traceback (most recent call last):
File "UFCNN1_REPO_V6_TESTMODE.py", line 596, in
case_tc = train_and_predict_classification(UFCNN_TC, features=features,
output_dim=output_dim, sequence_length=sequence_length, epochs=50,
training_count=10, testing_count = 6 )
File "UFCNN1_REPO_V6_TESTMODE.py", line 511, in
train_and_predict_classification
xdim, ydim = yp.shape
ValueError: too many values to unpack (expected 2)
root@DEEPGPU:/home/allcode/ufcnn-keras-master/models#

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub
#3

[developeralgo8888]

Hi Dmitry,
FYI,
i tested the tradcom_simple but the optimizer is not stable if you increase the epochs. I played around with Epochs = 50 , 100, 200, 300 , 400 but as you increase the epochs and start a fresh . Making sure to delete all the training files created in the previous tests. I get NAN as it goes to converge after 150 epochs but then when i do same test it converges without NAN or it stays at 0.4137 with no change. so it is a little unstable.

[ErnstTmp]

Hi Developer,
I have seen the same behaviour. I think it has to do with the RELU
activations, but the others do not work because of a bug. I hope it is
getting fixed.

Cheers,
Ernst
On 03/17/2016 05:58 AM, DeveloperAlgo wrote:

Hi Dmitry,
FYI,
i tested the tradcom_simple but the optimizer is not stable if you
increase the epochs. I played around with Epochs = 50 , 100, 200, 300
, 400 but as you increase the epochs and start a fresh . Making sure
to delete all the training files created in the previous tests. I get
NAN as it goes to converge after 150 epochs but then when i do same
test it converges without NAN or it stays at 0.4137 with no change. so
it is a little unstable.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub
#3 (comment)

[lukovkin]

OK, we are not 100% sure about signals yet, list of features, etc. And multiday training started working just now.
So it's OK for now, I think.