[TriPython] Prediction Model. Data Visualization.
Art
artem.nesterenko at gmail.com
Wed Oct 11 00:22:32 EDT 2017
Thank you for the prompt replies!
Sorry for the picture, didn't expect to have this issue.
So, basically what I did is I built a confusion matrix with correct and
incorrect predictions:
For example:
*Predicted*
target_1 target_2
*Real* target_1 120 23
target_2 37 44
and then I built a donut that represents the *number* of true pos, true
neg, false pos, false neg predictions from the confusion matrix. I've
attached the image. The attached donut doesn't represent the numbers in "%"
from the above matrix. This matrix is just an example.
And now I'm thinking of the best graph type to represent the number of pos
and neg predictions if the model predicts 7 targets.
Art Nestsiarenka
email: artem.nesterenko at gmail.com
Cell: (919) 455-5055
On Tue, Oct 10, 2017 at 8:52 PM, Jeff Heard <jefferson.r.heard at gmail.com>
wrote:
> I'd suggest you look at Seaborn. If you're dead-set on visualizing data
> in
> javascript, it might well give you some inspiration to draw from,
> otherwise it's a great Python solution for data
> vis.**[1]https://seaborn.pydata.org/ . I'm not 100% sure what I'd
> recommend given that much information. Also, it looks like the mailer
> scrubbed the images. If you could host the originals somewhere
> (GitHub?),
> it'd help understand what you did better to give a recommendation.
> Are your model outputs continuous or discrete? I understand that you're
> looking at "positive/negative" values, but decisions/classifications and
> actual model output can be different. Plus the choice of a donut chart
> to
> represent boolean values doesn't make a lot of sense to me, so I thought
> that I maybe understood you wrong.
> **
> On Tue, Oct 10, 2017 at 3:53 PM, Art <[2]artem.nesterenko at gmail.com>
> wrote:
>
> ** **Good afternoon!
> ** **I'm reaching out to you guys for a suggestion on the data
> visualization.
> ** **I'm wondering if anyone has an experience or an idea of
> visualizing
> the
> ** **volume of multiple model prediction targets.
> ** **Here is an example of a model with just 2 prediction targets I've
> recently
> ** **been working on:
> ** **1. This model provides 2 predictions: target_1 and target_2.**
> ** **2. I also know the actual result to compare against, so I could
> see
> the
> ** **model accuracy.**
> ** **3. I've built a confusion matrix to calculate true/false pos/neg
> (see
> ** **below). So, there are 4 values.
> ** **4. I chose donut**chart to visualize these values (see below).
> ** **[1]Inline image 2**
>
> ** **My question is what if the model provides more than 2
> predictions?
> Our
> ** **next model has 7 targets, which is 49 pos and neg values as far
> as
> I
> ** **understand. And I think a**donut**graph is not going to work in
> this
> ** **case.**Maybe a bar graph or something else fits better.**
>
> ** **I'd appreciate any ideas or examples of visualizations that easy
> to
> look
> ** **at and understand.
> ** **I'm using d3/c3.js for data visualization.
> ** **Thank you!
> ** **Art Nestsiarenka
> ** **email: [2][3]artem.nesterenko at gmail.com
>
> References
>
> ** **Visible links
> ** **2. mailto:[4]artem.nesterenko at gmail.com
>
> _______________________________________________
> TriZPUG mailing list
> [5]TriZPUG at python.org
> [6]https://mail.python.org/mailman/listinfo/trizpug
> [7]http://tripython.org is the Triangle Python Users Group
>
> References
>
> Visible links
> 1. https://seaborn.pydata.org/
> 2. mailto:artem.nesterenko at gmail.com
> 3. mailto:artem.nesterenko at gmail.com
> 4. mailto:artem.nesterenko at gmail.com
> 5. mailto:TriZPUG at python.org
> 6. https://mail.python.org/mailman/listinfo/trizpug
> 7. http://tripython.org/
>
> _______________________________________________
> TriZPUG mailing list
> TriZPUG at python.org
> https://mail.python.org/mailman/listinfo/trizpug
> http://tripython.org is the Triangle Python Users Group
>
>
-------------- next part --------------
Thank you for the**prompt replies!
Sorry for the picture, didn't expect to have this issue.
So, basically what I did is I built a confusion matrix with correct and
incorrect predictions:
For example:
** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **Predicted ** ** ** **
** ** ** ** ** ** ** ** ** ** ** ** ** **target_1 ** target_2
Real ** ** target_1 ** **120 ** ** ** ** ** 23
** ** ** ** ** ** target_2 ** ** 37 ** ** ** ** ** ** 44**
and then I built a donut that represents the number of true pos, true neg,
false pos, false neg predictions from the confusion matrix. I've attached
the image. The attached donut doesn't represent the numbers in "%" from
the above matrix. This matrix is just an example.
And now I'm thinking of the**best graph type to represent the number of
pos and neg predictions if the**model predicts 7 targets.
Art Nestsiarenka
email: [1]artem.nesterenko at gmail.com
Cell: (919) 455-5055
On Tue, Oct 10, 2017 at 8:52 PM, Jeff Heard
<[2]jefferson.r.heard at gmail.com> wrote:
** **I'd suggest you look at Seaborn. If you're dead-set on visualizing
data in
** **javascript, it might well give you some inspiration to draw from,
** **otherwise it's a great Python solution for data
** **vis.**[1][3]https://seaborn.pydata.org/ . I'm not 100% sure what
I'd
** **recommend given that much information. Also, it looks like the
mailer
** **scrubbed the images. If you could host the originals somewhere
(GitHub?),
** **it'd help understand what you did better to give a recommendation.
** **Are your model outputs continuous or discrete? I understand that
you're
** **looking at "positive/negative" values, but
decisions/classifications and
** **actual model output can be different. Plus the choice of a donut
chart to
** **represent boolean values doesn't make a lot of sense to me, so I
thought
** **that I maybe understood you wrong.
** ****
** **On Tue, Oct 10, 2017 at 3:53 PM, Art
<[2][4]artem.nesterenko at gmail.com>
** **wrote:
** ** **** **Good afternoon!
** ** **** **I'm reaching out to you guys for a suggestion on the data
** ** **visualization.
** ** **** **I'm wondering if anyone has an experience or an idea of
visualizing
** ** **the
** ** **** **volume of multiple model prediction targets.
** ** **** **Here is an example of a model with just 2 prediction
targets I've
** ** **recently
** ** **** **been working on:
** ** **** **1. This model provides 2 predictions: target_1 and
target_2.**
** ** **** **2. I also know the actual result to compare against, so I
could see
** ** **the
** ** **** **model accuracy.**
** ** **** **3. I've built a confusion matrix to calculate true/false
pos/neg
** ** **(see
** ** **** **below). So, there are 4 values.
** ** **** **4. I chose donut**chart to visualize these values (see
below).
** ** **** **[1]Inline image 2**
** ** **** **My question is what if the model provides more than 2
predictions?
** ** **Our
** ** **** **next model has 7 targets, which is 49 pos and neg values as
far as
** ** **I
** ** **** **understand. And I think a**donut**graph is not going to
work in
** ** **this
** ** **** **case.**Maybe a bar graph or something else fits better.**
** ** **** **I'd appreciate any ideas or examples of visualizations that
easy to
** ** **look
** ** **** **at and understand.
** ** **** **I'm using d3/c3.js for data visualization.
** ** **** **Thank you!
** ** **** **Art Nestsiarenka
** ** **** **email: [2][3][5]artem.nesterenko at gmail.com
** ** **References
** ** **** **Visible links
** ** **** **2. mailto:[4][6]artem.nesterenko at gmail.com
** ** **_______________________________________________
** ** **TriZPUG mailing list
** ** **[5][7]TriZPUG at python.org
** ** **[6][8]https://mail.python.org/mailman/listinfo/trizpug
** ** **[7][9]http://tripython.org is the Triangle Python Users Group
References
** **Visible links
** **1. [10]https://seaborn.pydata.org/
** **2. mailto:[11]artem.nesterenko at gmail.com
** **3. mailto:[12]artem.nesterenko at gmail.com
** **4. mailto:[13]artem.nesterenko at gmail.com
** **5. mailto:[14]TriZPUG at python.org
** **6. [15]https://mail.python.org/mailman/listinfo/trizpug
** **7. [16]http://tripython.org/
_______________________________________________
TriZPUG mailing list
[17]TriZPUG at python.org
[18]https://mail.python.org/mailman/listinfo/trizpug
[19]http://tripython.org is the Triangle Python Users Group
References
Visible links
1. mailto:artem.nesterenko at gmail.com
2. mailto:jefferson.r.heard at gmail.com
3. https://seaborn.pydata.org/
4. mailto:artem.nesterenko at gmail.com
5. mailto:artem.nesterenko at gmail.com
6. mailto:artem.nesterenko at gmail.com
7. mailto:TriZPUG at python.org
8. https://mail.python.org/mailman/listinfo/trizpug
9. http://tripython.org/
10. https://seaborn.pydata.org/
11. mailto:artem.nesterenko at gmail.com
12. mailto:artem.nesterenko at gmail.com
13. mailto:artem.nesterenko at gmail.com
14. mailto:TriZPUG at python.org
15. https://mail.python.org/mailman/listinfo/trizpug
16. http://tripython.org/
17. mailto:TriZPUG at python.org
18. https://mail.python.org/mailman/listinfo/trizpug
19. http://tripython.org/
More information about the TriZPUG
mailing list