Warning message while training the Model

[geetsagar pagare]

Hello Team,

I am trying to build a model which will predict the digits having labels 0-9.The train data set contains 42000 obs.

The images are of size 28*28.

Below is my model code and Waring message.please let me known if someone came across similar warning.Where  i am going wrong.

> res.dl <- h2o.deeplearning(x = 2:785, y = 1, trData, activation = "Tanh", hidden=rep(160,3),epochs = 10)

  |===================================================================================================| 100%

Warning message:

In .h2o.startModelJob(algo, params, h2oRestApiVersion) :

  Dropping bad and constant columns: [pixel729, pixel644, pixel645, pixel448, pixel727, pixel728, pixel560, pixel52, pixel760, pixel10, pixel54, pixel53, pixel168, pixel56, pixel11, pixel55, pixel57, pixel16, pixel18, pixel17, pixel19, pixel754, pixel755, pixel756, pixel757, pixel758, pixel759, pixel83, pixel196, pixel82, pixel85, pixel671, pixel84, pixel111, pixel672, pixel112, pixel673, pixel476, pixel392, pixel700, pixel701, pixel141, pixel780, pixel30, pixel781, pixel782, pixel420, pixel783, pixel31, pixel421, pixel140, pixel699, pixel139, pixel8, pixel9, pixel6, pixel7, pixel4, pixel5, pixel2, pixel3, pixel0, pixel21, pixel1, pixel20, pixel23, pixel532, pixel730, pixel22, pixel731, pixel25, pixel24, pixel27, pixel26, pixel29, pixel28].

[Darren Cook]

> I am trying to build a model which will predict the digits having labels 0-9.The
> train data set contains 42000 obs.
>
> The images are of size 28*28.
>
> Below is my model code and Waring message.please let me known if someone came
> across similar warning.

Some of the pixels are zero in all of your 42,000 observations. (It
could be any constant value, to get this message, but this comes up with
mnist data set for e.g. the pixels in the 4 corners.)

It is usually fine to ignore this warning message, as it is doing
something sensible (if a pixel always has the same value, your model
will never learn anything useful from it).

But, if you want to include those pixels (and thus get rid of this
warning), you can set:
ignore_const_cols = FALSE

You might do this if intending to make visualizations from your model,
for instance.

Darren

> > res.dl <- h2o.deeplearning(x = 2:785, y = 1, trData, activation = "Tanh",
> hidden=rep(160,3),epochs = 10)
>
> |===================================================================================================|
> 100%
> Warning message:
> In .h2o.startModelJob(algo, params, h2oRestApiVersion) :
> Dropping bad and constant columns: [pixel729, pixel644, pixel645, pixel448,
> pixel727, pixel728, pixel560, pixel52, pixel760, pixel10, pixel54, pixel53,
> pixel168, pixel56, pixel11, pixel55, pixel57, pixel16, pixel18, pixel17,
> pixel19, pixel754, pixel755, pixel756, pixel757, pixel758, pixel759, pixel83,
> pixel196, pixel82, pixel85, pixel671, pixel84, pixel111, pixel672, pixel112,
> pixel673, pixel476, pixel392, pixel700, pixel701, pixel141, pixel780, pixel30,
> pixel781, pixel782, pixel420, pixel783, pixel31, pixel421, pixel140, pixel699,
> pixel139, pixel8, pixel9, pixel6, pixel7, pixel4, pixel5, pixel2, pixel3,
> pixel0, pixel21, pixel1, pixel20, pixel23, pixel532, pixel730, pixel22,
> pixel731, pixel25, pixel24, pixel27, pixel26, pixel29, pixel28].

--
Darren Cook, Software Researcher/Developer
My New Book: Practical Machine Learning with H2O:
http://shop.oreilly.com/product/0636920053170.do