# "Neural Net interpret result"

Member Posts: 4 Contributor I
edited June 2019 in Help
Hi,

im working with neural nets. The problem is, i dont understand how to interpret the result from rapid miner???
Im no mathematican. How should i multiplicate the input value with the weights to get the predicted value?  I dont know what bias and threshold measns.

I hope you can give me hint.
Tagged:

• Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 2,463 RM Data Scientist
Hi Immo,

why do you want to understand this result? For me their is no reason to recalculate it. Just use the model.
If you really want to understand it, you will have to understand the math of neural nets.

Cheers,
Martin
- Head of Data Science Services at RapidMiner -
Dortmund, Germany
• Member Posts: 4 Contributor I

I want to predict the turnover of a grocery strore. I have data (turnover and influences on the turnover) of 145 stores.
If i want to establish a new store how can i use the result from rapid miner to predict the turnover of the new store?

Im sure it is possible with rapid miner. Can you help me?

How can i use the model? Can rapid miner show me the equation?

I understand how to recalculate but get a different result.

Greetings

Immo001

• Member Posts: 321 Unicorn
Hi,
It might be useful to provide us more information: what data are you using, which process, what is the result, which conclusions did you take based on the results. How does your result differ from what RM produces. Comparing requires exactly the same parameters, etc.
Good luck!
Sven
• Member Posts: 4 Contributor I
My data is normalized (between 0 and 1)

I use for recalculating and understanding 27 datasets (row). Two attributes, one label (column)

My process:
"Data" --> "x-validation"
In the "x-validation" is the "neural net" on the left side (training), "apply model" and "performance" is on the right side (testing).
In the end of x-validation is again "apply model" connected with multiply to the data.

Rapid miner shows in the input layer attribute 1, attribute 2 and a threshold node.
In the hidden layer are three noddes and a threshold node.
There is only one output node.

Descripted Result:

Node 1
1 att. -1.763
2 att. -1.144
Bias

Node 2
1 att. -1.776
2 att. -1.103
Bias -1.178

Node 3
1 att. -1.937
2 att. -1.937
Bias -0.996

Output
1 Node -1.389
2 Node -1.376
3 Node -1.495
Threshold 0.112

Rapid miner gives me also a predicted value for every row.

Now I want to use the first row of my dataset (att. 1 and 2) to recalculate the predicted result.
The Idea is, if i know how to recalculate I can calculate new data. Or use the formula in a Excel sheet. In my exapmle it is the turnover of a neu grocery store.

My calculations (1att. 0.532, 2. att 0.089, the predicted result is 0.341)

Node 1
0.532 * (-1.763) + 0.089 * (-1.144) + (-1.173) = -2,212
after the sigmoid transformation I get 0,098

Node 2
0.532 * (-1.776) + 0.089 * (-1.103) + (-1.178) = -2,220
after the sigmoid transformation I get 0,097

Node 3
0.532 * (-1.937) + 0.089 * (-1.254) + -0.996 = -2,137
after the sigmoid transformation I get 0,105

Then I do the linear regression. Im not sure if it is right. I dont know how to use the threshold value.

0,098  (-1.389)
0,097  (-1.376)
0,105  (-1.495)

I get a correlation coefficient of 0,999
but I expect 0.341 ????

Do you understand my problem?

I hope somebody can help me. If you need detail please ask me.

How can rapid miner predict the turnover of a new grocery store (att. 1+ 2, but without a label/turnover)?