# Deep Neural Network visualization on multi dimension datasets [closed]

I have seen the PlayGround of Tensorflow. But it is using only 2D values which means only 2 inputs are taken.
I have 7 inputs and the output is only 1.

See the datasets sample:

input1,input2,input3,input4,input5,input6,input7,output
1.20047,1.20047,1.20045,1.20045,1.200059606178815,0.0003585714285714223,-3.791789292925229e-06,-1.0
1.20088,1.20088,1.20087,1.20087,1.200329737452543,0.0003835714285714117,6.180045764568489e-07,-1.0
1.20007,1.20023,1.20007,1.20007,1.200243158301695,0.0004278571428571306,4.129903631236368e-06,0.0
1.20022,1.20023,1.20022,1.20023,1.2002387722011298,0.0004271428571428402,7.19294850670973e-06,1.0
1.20023,1.20029,1.20023,1.20029,1.2002558481340868,0.0004192857142857093,9.916424565845627e-06,1.0
1.20046,1.20046,1.20046,1.20046,1.2003238987560578,0.00042499999999998485,1.259884292315837e-05,0.0
1.20053,1.20053,1.20053,1.20053,1.2003925991707052,0.0004049999999999966,1.523257396764079e-05,0.0
1.20045,1.20072,1.20045,1.20067,1.200485066113804,0.0004164285714285794,1.794705969746539e-05,1.0
1.20073,1.20092,1.20073,1.20092,1.2006300440758693,0.000424285714285742,2.103349972108536e-05,1.0
1.2009299999999998,1.2009299999999998,1.20075,1.2008,1.2006866960505802,0.0004228571428571772,2.3925052012490654e-05,-1.0
1.20094,1.20094,1.20094,1.20094,1.200771130700387,0.000396428571428607,2.676900242219225e-05,0.0
1.20094,1.20094,1.20094,1.20094,1.200827420466924,0.0002628571428571757,2.9380206159665192e-05,0.0
1.2009299999999998,1.2009299999999998,1.20082,1.2009,1.200851613644616,0.0002557142857143352,3.156425194537858e-05,-1.0
1.2009,1.2009,1.2009,1.2009,1.2008677424297445,0.00019428571428575792,3.3289202180031105e-05,0.0


Kindly, suggest me how I can visualize the datasets similar to Playground of Tensorflow. If there is any edited version of the Tensorflow Playground please do suggest if any.

## closed as unclear what you're asking by Firebug, Heteroskedastic Jim, Sycorax, kjetil b halvorsen, mdeweyDec 15 '18 at 10:27

Please clarify your specific problem or add additional details to highlight exactly what you need. As it's currently written, it’s hard to tell exactly what you're asking. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.

• The simple answer is you can't, unless you could visualize 7 dimensions (most of us can't from my experience). – Firebug Dec 14 '18 at 11:55
• Similar to TensorFlow playground in what terms? What kind of visualization you expect? – Tim Dec 14 '18 at 14:58
• I disagree with my answer being transformed into a comment. The asker specifically asked if it was possible to visualize like in the PlayGround, which simple is not possible. My answer was to the point, sure, but it's not wrong and it's complete. – Firebug Dec 14 '18 at 18:06

There is no plug and play package like Tensorflow Playground to visualize higher dimensional data.

First you need to specify what you want to achieve by your representation: is it exploratory analysis of the data? Or is it representation of the decision boundaries (I'm assuming your problem is a classification, not a regression)?

If your purpose is to represent the decision boundary, it's a challenging topic and there's no established method to do so, but research in that direction has been performed, see here and here.

If your purpose is simply to analyze the data, prior to classification, then a pairwise scatter plot is one the more common approaches, see for example the below image of a pairwise scatter plot of the Titanic data from Kaggle. This can be done in 3 lines of code using Matplot lib (in the below example I also use Seaborn - but that's optional):

• As OP specifically mentioned the TF Playground and an output variable, I suppose he meant to visualize decision surfaces and activation spaces, like it's possible in the Playground. – Firebug Dec 14 '18 at 23:06
• Yes this is what I mean Sir. @Firebug You have suggested the rightful meaning of the question. I do not understand what was difficulty in understanding as my question is put on hold. – Jaffer Wilson Dec 17 '18 at 5:18
• @JafferWilson Ok. Then check the paper I linked to and the references therein. – Skander H. Dec 17 '18 at 7:41
• @SkanderH. I have check it. But I was looking for interactive one which similar to Playground. But the above is not exactly but gives meaning – Jaffer Wilson Dec 17 '18 at 7:44
• @JafferWilson as this is relatively recent research, you're unlikely to find an ready made package like Tensorflow, you will need to code it yourself. – Skander H. Dec 17 '18 at 8:16

The simple answer is you can't, unless you could visualize 7 dimensions (most of us can't from my experience).

• Is there no way out to visualize it using a machine? Please sir help me in making i feasible if any possibility exists. – Jaffer Wilson Dec 14 '18 at 12:05
• This is being automatically flagged as low quality, probably because it is so short. At present it is more of a comment than an answer by our standards. Can you expand on it? Why wouldn't methods to visualize multidimensional datasets (scatterplot matrices, conditional plots, interactive plots, etc) work in this context? You can also turn it into a comment. – gung Dec 14 '18 at 14:31
• This answer seems to rely on an extremely limited concept of "visualization" as "the psychological construction of a geometrically accurate seven-dimensional world." That, however, is too limited to be useful in most applications, because--as you say--it is impossible to do. Thus, over the centuries statisticians and scientists have developed alternative ways to see high-dimensional data. These use 2D or 3D graphics, animated or not, that employ visual symbols (geometric and graphic) to represent aspects of the data. That is what most people nowadays mean by "visualization." – whuber Dec 14 '18 at 15:27
• @whuber I meant visualization as in the TF playground, where all features are visible in a 2D space since it's a two-dimensional problem. I think that's exactly what the question was asking for. – Firebug Jan 25 at 20:39
• I think I understand what you mean (I lack experience with the TF playground). Would you mind clarifying your answer to help your readers interpret it as you intended? – whuber Jan 25 at 21:44