Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
  • Jun 21 2015 23:06

    TNick on master

    unroll iterator buffer for debu… (compare)

  • Jun 20 2015 20:29

    TNick on master

    fix issue 1172 CudaNdarray error fix add doc string for test file and 33 more (compare)

  • Jun 20 2015 20:23

    TNick on master

    ordered dict in yaml csv remove y_labels make training_algorithms/tests … (compare)

  • Jun 10 2015 00:00

    TNick on predict_page

    A page about using models to ge… (compare)

  • Jun 09 2015 16:41

    TNick on predict_page

    Make ZCA test use fix seed Bugfixes in SVHN and DenseDesig… SVHN/DenseDesignMatrixPyTables … and 16 more (compare)

  • May 21 2015 21:57

    TNick on master

    Added X_labels and y_labels to … Redefined _check_labels for Den… Added unit test and 34 more (compare)

  • May 21 2015 21:54

    TNick on master

    Ignore documentation (compare)

  • May 13 2015 12:48

    TNick on fix_1496

    Added X_labels and y_labels to … Redefined _check_labels for Den… Added unit test and 19 more (compare)

  • May 13 2015 12:07

    TNick on fix_1496

    (compare)

  • May 05 2015 22:25

    TNick on master

    OneHotFormatter: include dtype … Make sure dtype of formatted da… Added test for ZCA_Dataset. and 47 more (compare)

  • Apr 30 2015 21:00

    TNick on train_multi

    OneHotFormatter: include dtype … Make sure dtype of formatted da… Added test for ZCA_Dataset. and 42 more (compare)

  • Apr 30 2015 20:50

    TNick on train_multi

    Allow downloading even if the d… download caltech dataset (compare)

  • Apr 27 2015 14:51

    TNick on fix_1496

    AdaGrad accepts zero-filled par… (compare)

  • Apr 27 2015 13:51

    TNick on fix_1496

    (compare)

  • Apr 26 2015 23:26

    TNick on fix_1468

    run print_monitor_cv.py on a pk… (compare)

  • Apr 26 2015 23:00

    TNick on fix_1468

    (compare)

  • Apr 26 2015 22:35

    TNick on fix_1467

    tear_down for train extensions;… (compare)

  • Apr 26 2015 21:40

    TNick on fix_1467

    (compare)

  • Apr 26 2015 21:32

    TNick on fix_1496

    AdaGrad accepts zero-filled par… (compare)

  • Apr 26 2015 19:04

    TNick on fix_1496

    (compare)

HuniyaArif
@HuniyaArif
http://deeplearning.net/tutorial/lstm.html I run the following lstm code for my own data, but I don't understand the output. Train 0.281.. Valid 0.0556 Test 0.5. This statement is printed multiple times. How can I find out the total overall accuracy of lstm based on this result?
Nicu Tofan
@TNick
The inner loop prints the accuracy for train sub-set, for valid subset and for test subset
After the training ended these values are computed one last time
HuniyaArif
@HuniyaArif
So the last time is the overall accuracy? For accuracy I look at the test value?
Nicu Tofan
@TNick
The value you're interested in, I think, is the value after the test in the last printed line. That tells you how well your model performs on an input that the model never saw before / is not used in any way in training that model.
HuniyaArif
@HuniyaArif
@TNick This is my result http://i.stack.imgur.com/Cp78v.png
Nicu Tofan
@TNick
Right; the relevant lines for that line with results are:
train_err = pred_error(f_pred, prepare_data, train, kf_train_sorted)
valid_err = pred_error(f_pred, prepare_data, valid, kf_valid)
test_err = pred_error(f_pred, prepare_data, test, kf_test)
As you can see, the code simply does a forward propagation and computes the result
Interpretation: on train dataset the model predicts the wrong thing 28% of the time (and gets it right 72%)
on validation dataset the error goes down to 5%
on test dataset the error is 50%
I should mention that I only say general things here as I don't have the time to read the code in http://deeplearning.net/tutorial/code/lstm.py and understand what it's doing
I'm not that cursive in this matters to understand at a glance the details
HuniyaArif
@HuniyaArif
Thank you so much!
Nicu Tofan
@TNick
sure, welcome
HuniyaArif
@HuniyaArif
I switched to lstm code by keras library and got better results (86%!)
Nicu Tofan
@TNick
for train subset?
HuniyaArif
@HuniyaArif
Test
Nicu Tofan
@TNick
oh. By the way, do you plan to give TensorFlow a try? I don't have the time to do that myself these days.
Punita Ojha
@punitaojha
This message was deleted
Renato Marinho
@renatomarinho
This message was deleted
Renato Marinho
@renatomarinho
This message was deleted