Instructions on Using the to ( Building a prediction Mod Step 1: Enter Your Data Usually one builds prediction model with 1 output only. If you have say, 2 output variables Y1 and Y2, both of which depend on the sa you may be better off, building 2 separate models - One with Y1 as Output, an Make sure that the number of Input (Cat & Cont) columns exactly match with t Application will replace it by the column mean Application will reaplce it by the most frequently occuring category If one of the category of a Cat column has only 1 observation, you Remove that observation OR Rename the category to any other categories of that Cat col Step 2: Fill up Model Inputs Step 3: Results of Modeling At the end of the run, the final set of weights are saved in the Calc sheet. as the training of the model progresses. Two charts showing training and Vali have been already provided in the Output sheet. will be created containing the model inputs, your data, and the fitted model You will be able to use this file as a calculator to do prediction, given any Step 4: Study Profiles Profile plot is the next best way to visualize this fitted surface. (A) Enter your data in The Data worksheet, starting from the cell AC105 (B) The observations should be in rows and the variables should be in columns (C) Above each column, choose appropriate Type (Omit, Output, Cont, Cat) To drop a column from model - set the type = Omit To treat a column as categorical Input, set type = Cat To treat a column as continuous Input, set type = Cont To treat a column as Output, set type = Output You can have atmost 10 output variables. Application will automatically treat You can have at most 50 input variables, out of which atmost 40 could be cate (D) Please make sure that your data does not have blank rows or blank columns (E) Continuous Inputs: Any non-number in Cont column will be treated as missing value. (E) Categorical Inputs: Any blank cell or cells containing Excel error in Cat column will be Category labels are case insensitive - lables good, Good, GoOd, GOOD There should be at least 2 observations in each category of a Cat co (A) Fill up the model inputs in the User Input Page. (B) Make sure that your inputs are within the range of values allowed by the (C) Click the 'Build Model' button to start modeling. (A) A Neural Network model is basically a set of weights between the layers o (B) The output page of this file will show you the values of MSE and ARE on t (C) In UserInput page if you have asked to save the model in a separate file, Fitted model is a surface in p-dimension where the number of your inputs is p Unless p is 2 or less, it is not possible to show the surface graphically.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Instructions on Using the tool ( Building a prediction Model)
Step 1: Enter Your Data
Usually one builds prediction model with 1 output only. If you have say, 2 output variables Y1 and Y2, both of which depend on the same set of Input variables, you may be better off, building 2 separate models - One with Y1 as Output, another one with Y2 as output.
Make sure that the number of Input (Cat & Cont) columns exactly match with the number entered in UserInput sheet.
Application will replace it by the column mean
Application will reaplce it by the most frequently occuring category.
If one of the category of a Cat column has only 1 observation, you should do one of the following - Remove that observation ORRename the category to any other categories of that Cat column.
Step 2: Fill up Model Inputs
Step 3: Results of Modeling
At the end of the run, the final set of weights are saved in the Calc sheet.
as the training of the model progresses. Two charts showing training and Validation MSE'shave been already provided in the Output sheet.
will be created containing the model inputs, your data, and the fitted model ( i.e. the weights)You will be able to use this file as a calculator to do prediction, given any new input.
Step 4: Study Profiles
Profile plot is the next best way to visualize this fitted surface.
(A) Enter your data in The Data worksheet, starting from the cell AC105(B) The observations should be in rows and the variables should be in columns.(C) Above each column, choose appropriate Type (Omit, Output, Cont, Cat)
To drop a column from model - set the type = OmitTo treat a column as categorical Input, set type = CatTo treat a column as continuous Input, set type = ContTo treat a column as Output, set type = Output
You can have atmost 10 output variables. Application will automatically treat them all as continuous variables.
You can have at most 50 input variables, out of which atmost 40 could be categorical.
(D) Please make sure that your data does not have blank rows or blank columns.(E) Continuous Inputs:
Any non-number in Cont column will be treated as missing value.
(E) Categorical Inputs: Any blank cell or cells containing Excel error in Cat column will be treated as missing value
Category labels are case insensitive - lables good, Good, GoOd, GOOD will all be treated as the same categoryThere should be at least 2 observations in each category of a Cat column.
(A) Fill up the model inputs in the User Input Page.(B) Make sure that your inputs are within the range of values allowed by the application.(C) Click the 'Build Model' button to start modeling.
(A) A Neural Network model is basically a set of weights between the layers of the net.
(B) The output page of this file will show you the values of MSE and ARE on the training and validation set
(C) In UserInput page if you have asked to save the model in a separate file, then a new file
Fitted model is a surface in p-dimension where the number of your inputs is p.Unless p is 2 or less, it is not possible to show the surface graphically.
we get the profile plot - which is really a one dimensional cross section of the high dimensional surface.In the Profile sheet you can specify which predictor to vary and the values at which the other predictors should be held fixed.
If the predictor you choose to vary is categorical then the other info ( #points to be generated, start and end values)will be ignored and the graph will show you the predicted response for each category of the predictor you have chosen to vary.
Profile plot lets you study the following things:
( E.g. Y increases as X increases OR Y decreases as X increases OR the relationship is non-linear - Y first increases and then decreases with X etc etc.
Suppose there are two predictors X and Z and we are studying the profile of Y as X variesSuppose we look at the profile by keeping Z fixed at 1 and varying X between -10 and 10.Now keep Z fixed at 2 instead of 1 and vary X between -10 and 10. If the shape of the profiles in these two scenarios are drastically different (e.g. one is increasing and the other is decreasing) then that says thay X and Z has interaction. In other words, the effect of X on the Response is not same at all levels of ZTo study the effect of X, it matters where Z is set.
A few more points …
Initial weightsFor the training of the model, we need to start with an initial set of values of the network weights.
Next time you want to train a model with same architecture and same data, the application will ask you whether to start with the weights already saved in Calc sheet.
Specifying your choice of starting weights is a bit non-trivial for this application. Here is how you do it.
This will just setup the Calc page without doing any training.Now go to Calc sheet and write down your choice of weights in the appropriate places of the weight matrices.
Now come back to UserInput sheet and specify the number of trining cycles you want and click on the Buil Model button.When the application asks whether to use the already saved weights, click on the YES button.Now your network will be trained with the starting weights specified by you.
By varying only one predictor between two values and keeping all the others fixed at some pre-specified values
Click Create Profile button to generate the profile.
(1) Nature of relationship bettween a particular predictor X and the response Y
(2) Profile plots also lets you study the interaction between predictors.
By default, the weights are initialized with random values between -w and w.where w is a number between 0 and 1, specified by you in the UserInput page. (A) Once you build a model, the final weights are stored in Calc page.
If you say YES, these wights are used. If you say NO, the weights are re-initialized with random values.(B) Instead of starting with ramdom weights, you may want to start with our own choice of weights.
Specify the inputs in the UserInput page and specify the number of training cycle as 0
# Missing ValueMinMaxAverage
If you have say, 2 output variables Y1 and Y2, both of which depend on the same set of Input variables, sdyou may be better off, building 2 separate models - One with Y1 as Output, another one with Y2 as output. Intercept
Slope
Make sure that the number of Input (Cat & Cont) columns exactly match with the number entered in UserInput sheet.
If one of the category of a Cat column has only 1 observation, you should do one of the following -
Application will automatically treat them all as continuous variables.
column will be treated as missing value
GOOD will all be treated as the same category
The output page of this file will show you the values of MSE and ARE on the training and validation set
we get the profile plot - which is really a one dimensional cross section of the high dimensional surface.In the Profile sheet you can specify which predictor to vary and the values at which the other predictors should be held fixed.
If the predictor you choose to vary is categorical then the other info ( #points to be generated, start and end values)will be ignored and the graph will show you the predicted response for each category of the predictor you have chosen to vary.
OR the relationship is non-linear - Y first increases and then decreases with X etc etc.
Suppose there are two predictors X and Z and we are studying the profile of Y as X variesSuppose we look at the profile by keeping Z fixed at 1 and varying X between -10 and 10.
(e.g. one is increasing and the other is decreasing) then that says thay X and Z has interaction.
For the training of the model, we need to start with an initial set of values of the network weights.
Specifying your choice of starting weights is a bit non-trivial for this application. Here is how you do it.
Now go to Calc sheet and write down your choice of weights in the appropriate places of the weight matrices.Now come back to UserInput sheet and specify the number of trining cycles you want and click on the Buil Model button.
predictor between two values and keeping all the others fixed at some pre-specified values
, the weights are re-initialized with random values.Instead of starting with ramdom weights, you may want to start with our own choice of weights.
specify the number of training cycle as 0.
Cont. Var. Cat. Var. Values Dummy# Missing Value #Levels
Lables
From very last cycle 1 2With least Training Error 2
With least Validation Error 3 1
Partition data into Training / Validation set 1Use whole data as training set 2
Network ArchitectureOptions
2
2
0.7 Initial Wt Range ( 0 +/- w): w =
0.5
Training Options13
Present Inputs in Random order while Training ? NO
Save Network weights With least Training Error
Training / Validation Set Partition data into Training / Validation set
If you want to partition, how do you want to select the Validation set ? Please choose one option 1 Option 1 : Randomly selectPlease fill up the input necessary for the selected option Option 2: Use last
Save model in a separate workbook? NO
Number of Inputs ( bewtween 2 and 50) Number of Outputs
Number of Hidden Layers ( 1 or 2 ) Hidden Layer sizes
Learning parameter (between 0 and 1)
Momentum (between 0 and 1)
Total #rows in your data ( Minimum 10 ) No. of Training cycles
Training Mode
1
Hidden 1 Hidden 26 3
Initial Wt Range ( 0 +/- w): w = 0.99
500
Sequential
Partition data into Training / Validation set
Option 1 : Randomly select 10%Option 2: Use last 6 rows of the data as validation set
Number of Outputs ( between 1 and 10 )
Hidden Layer sizes ( Maximum 20 )
No. of Training cycles ( Maximum 500 )
Training Mode (Batch or Sequential )
of data as Validation set (between 1% and 50%)
Enter your Data in this sheetInstructions:
Make sure that the row 104 is blank.Specify variable type in row 102.
For each continuous Input, there will be 1 neuron in Input Layer.
Var Type Cont Cont Output Omit
Var Name X1 X2 Y0 1 2.651 -2 14.12 3 30.853 5 76.75
Cont - for continuous Input, Cat - for Categorical Input, Output -for Output var. Omit - if you don't want to usethe variable in the model
For Each categorical Input with K levels, there will be K neurons in Input LayerPlease make sure that there are no more than 50 neurons in Input Layer.There should be at most 10 Output variables - application will treat them all as Continuous.There should be no more than 40 Categorical Input Variables.
Specify variable name in row 103.
Omit Omit Omit Omit Omit
- application will treat them all as Continuous.
Omit Omit Omit Omit Omit
Omit Omit Omit Omit Omit
X18 X19
Omit Omit Omit Omit Omit
X20 X21 X22 X23 X24
Omit Omit Omit Omit Omit
X25 X26 X27 X28 X29
Omit Omit Omit Omit Omit
X30 X31 X32 X33 X34
Omit Omit Omit Omit Omit
X35 X36 X37 X38 X39
Omit Omit Omit Omit Omit
X40 X41 X42 X43 X44
Omit Omit Omit Omit Omit
X45 X46 X47 X48 X49
Omit Omit Omit Omit Omit
X50 X51 X52 X53 X54
Omit Omit Omit Omit Omit
X55 X56 X57 X58 X59
Omit
X60
Neural Network Model for Prediction Created On : 6-Nov-07
Generate profile for Generateby varying keeping the other predictors fixed at the specified values
Outputs Predictors X1 Predicted YY X1 0 0.825191
X2 0.01 0.8389260.02 0.851391 Predictor X10.03 0.862673 Fixed Value 0.6920.04 0.872862 Min / Max in Original Data (for user's reference only)0.05 0.882046 Min -4.000.06 0.890312 Max 5.000.07 0.8977420.08 0.9044130.09 0.9103980.1 0.915763