Kaggle Plankton example problem

User 1902 | 5/9/2015, 7:58:42 PM

Hi,

I have the exact same problem as described by https://github.com/dato-code/PowerGraph/issues/174 below:

Can you please help? My email is: pembek at gmail.com.tr

I run the make_submission.py for the kaggle competition for plankton under Ubuntu It runs smoothly up to a point and then crushes. Any idea?

Python 2.7.9 |Anaconda 2.2.0 (64-bit)| (default, Mar 9 2015, 16:20:48) Type "copyright", "credits" or "license" for more information.

IPython 3.0.0 -- An enhanced Interactive Python. Anaconda is brought to you by Continuum Analytics. Please check out: http://continuum.io/thanks and https://binstar.org ? -> Introduction and overview of IPython's features. %quickref -> Quick reference. help -> Python's own help system. object? -> Details about 'object', use 'object??' for extra details. %guiref -> A brief reference about the graphical user interface.

IIn [2]: run makesubmission.py A newer version of GraphLab Create (v1.3) is available! Your current version is v1.3.0.gpu. You can use pip to upgrade the graphlab-create package. For more information see https://dato.com/products/create/upgrade.[INFO] Start server at: ipc:///tmp/graphlabserver-23031 - Server binary: /home/midas/anaconda2/lib/python2.7/site-packages/graphlab/unityserver - Server log: /tmp/graphlabserver1428320002.log [INFO] GraphLab Server Version: 1.3.0.gpu Loading images... Resizing images... Data Augmentation... Training Model... Using network:

network layers

layer[0]: ConvolutionLayer initrandom = gaussian padding = 2 stride = 4 numchannels = 96 numgroups = 1 kernelsize = 5 layer[1]: RectifiedLinearLayer layer[2]: MaxPoolingLayer padding = 0 stride = 2 kernelsize = 3 layer[3]: ConvolutionLayer initrandom = gaussian padding = 2 stride = 1 numchannels = 128 numgroups = 1 kernelsize = 3 layer[4]: RectifiedLinearLayer layer[5]: ConvolutionLayer initrandom = gaussian padding = 1 stride = 1 numchannels = 128 numgroups = 1 kernelsize = 3 layer[6]: RectifiedLinearLayer layer[7]: MaxPoolingLayer padding = 0 stride = 2 kernelsize = 3 layer[8]: FlattenLayer layer[9]: FullConnectionLayer initsigma = 0.01 initrandom = gaussian initbias = 0 numhiddenunits = 512 layer[10]: RectifiedLinearLayer layer[11]: DropoutLayer threshold = 0.5 layer[12]: FullConnectionLayer initsigma = 0.01 initrandom = gaussian initbias = 0 numhiddenunits = 512 layer[13]: RectifiedLinearLayer layer[14]: DropoutLayer threshold = 0.5 layer[15]: FullConnectionLayer initsigma = 0.01 initrandom = gaussian initbias = 0 numhidden_units = 121 layer[16]: SoftmaxLayer

end network layers

network parameters

initrandom = xavier randommirror = 1 learningrate = 0.01 inputshape = 3,48,48 learningrategamma = 0.1 biaslearningrate = 0.02 initsigma = 0.01 batchsize = 256 learningratestep = 10000 randomcrop = 1 learningrateschedule = exponentialdecay l2regularization = 0.0005 momentum = 0.9 biasl2_regularization = 0.0

end network parameters

Traceback (most recent call last):

File "/home/midas/kaggle/bowl4/makesubmission.py", line 140, in <module> randomcrop=True)

File "/home/midas/anaconda2/lib/python2.7/site-packages/graphlab/toolkits/classifier/neuralnetclassifier.py", line 976, in create **(_network.params))

File "/home/midas/anaconda2/lib/python2.7/site-packages/graphlab/toolkits/supervisedlearning.py", line 359, in create validationset.head().append(dataset.head())

File "/home/midas/anaconda2/lib/python2.7/site-packages/graphlab/datastructures/sframe.py", line 3350, in append raise RuntimeError("Two SFrames have to have the same number of columns")

RuntimeError: Two SFrames have to have the same number of columns

Comments

User 91 | 5/11/2015, 6:51:25 PM

Which version of Python are you using? Is it python 3? Currently, we do not support Python 3.


User 1902 | 5/12/2015, 8:44:19 AM

I think it was Python 2.3, but I am not sure. I had to format my OS last night. But I was setting up your reccomendaditons: requirements-gpu.pip and requirements.pip
from the address https://github.com/srikris/datascience-bowl .


User 91 | 5/12/2015, 3:16:05 PM

Our requirements are python 2.7+. Can you try with that?