Model prototyping produces an initial version of machine learning to evaluate its viability and collect input before completing the design and constructing the final model. The prototype is often a reduced version of the final model, with a smaller dataset and fewer features. It is used to rapidly test the fundamental functionality and performance of the model and discover problems that must be resolved before completing the design. This enables quick iteration and testing of several techniques, ensuring the final model is well-suited to the job.
An example of a prototype model is the K-nearest neighbors (or KNN). KNN uses the distance between the neighbors to determine which other data set(s) are the closest to the set in question and assigns the response variable based on the neighbors. The idea is that all nearby points should be similar. This algorithm is also known as supervised learning since the destination is known. The algorithm calculates the distance of the point in question to all the stored data points, the distance values are sorted, and the k-nearest neighbors are determined. The labels of these neighbors are gathered, and a majority vote is used for classification or regression purposes. Selecting the value of k (the number of neighbors to use) will determine how well the data can be utilized to generalize the results of the kNN algorithm. A large k value reduces the variance due to noisy data but could be biased and may ignore useful patterns within the data set.
Overall, the kNN algorithm is highly unbiased and makes no a priori assumptions of the underlying data. It is simple to easy to understand and is quite popular. The disadvantage is that it is too simple and does not create any models for the data. It does not deeply understand the data's nature and might sometimes miss useful insights.
Note that this functionality requires that RScript software be installed on your computer and linked to the Sigma Magic software within the options menu. Prototype Models worksheet can be added to your active workbook by clicking on Analytics and then selecting Prototype Models.
Inputs
Setup
Click on Analysis Setup to open the menu options for this tool. A sample screenshot of the menu is shown below.
1
Response Type:
Specify the response data type you want to use to fit this model. The typical options for the response type are shown below. However, not all of these options may be applicable or relevant in all cases. Only those options that apply to your particular case are shown in the dropdown box.
Option
Description
Classification
The response variable can contain two or more classes. For example, Grades A, B, C, D, E, and F.
Regression
The response variable is usually continuous. For example, temperatures, pressures, etc., can take on any value.
2
Model:
Specify which library you want to use to fit your model. Usually, several different libraries are available, and more are added regularly. Currently, the following libraries are available depending on the response variable you selected.
Num
Response
Model
Name
Library
Num Params
1
Classification
protoclass
Greedy Prototype Selection
proxy, protoclass
2
2
Classification
kknn
k-Nearest Neighbors
kknn
3
3
Classification
lvq
Learning Vector Quantization
class
2
4
Classification
pam
Nearest Shrunken Centroids
pamr
1
5
Classification
ownn
Optimal Weighted Nearest Neighbor Classifier
snn
1
6
Classification
snn
Stabilized Nearest Neighbor Classifier
snn
1
7
Regression
cubist
Cubist Model
cubist
2
8
Regression
kknn
k-Nearest Neighbors
kknn
3
3
Pre-Process:
Specify if you want to pre-process your data before you fit a model. The available options are:
Option
Description
None
No pre-processing of data is performed before model fitting. The data is used as-is.
Center
Subtract a mean value from each of the data points so that the average factor value is zero.
Center, Scale
Subtract a mean value from each of the data points and divide by its standard deviation so that the average factor value is zero and the standard deviation is one.
Range
Adjust the values such that the minimum value is mapped to 0 and the maximum value is mapped to 1.
Scale
Divide each of the values by the standard deviation so that the standard deviation of the factor is 1.
4
Model Selection:
Specify the metric to use for tuning the model. This metric selects the best model from an available list of models. The available options depend on the type of response.
Option
Response Type
Description
RMSE
Regression
Uses the smallest value of the Root Mean Square Error (RMSE) to pick the best model.
MAE
Regression
Uses the smallest value of the Mean Absolute Error (MAE) to pick the best model.
Accuracy
Binary, Classification
Uses the largest value of the percentage of items that are matched correctly to pick the best model.
Kappa
Binary, Classification
Uses the largest value of the Kappa statistic to pick the best model.
ROC
Binary
Uses the largest value of the Area Under the ROC curve to pick the best model.
5
Selected Model:
The details of the selected model are displayed in this box based on the response type and the model you have chosen. You can review the model description to determine if this is the model you want to fit for your data.
6
Help Button:
Click on the Help Button to view the help documentation for this tool.
7
Cancel Button:
Click on the Cancel Button to discard your changes and exit this menu.
8
OK Button:
Click on the OK Button to save your changes and try to execute the program. Note that we do not control the algorithms for fitting these models, and there is no guarantee that the model will properly fit your selected data set. You can contact the R community for any possible resolution if there are model errors.
Data
You will see the following dialog box if you click the Data button. Here, you can specify the data required for this analysis.
1
Search Data:
The available data displays all the columns of data that are available for analysis. You can use the search bar to filter this list and speed up finding the right data for analysis. Enter a few characters in the search field, and the software will filter and display the filtered data in the Available Data box.
2
Available Data:
The available data box contains the list of data available for analysis. If your workbook has no data in tabular format, this box will display "No Data Found." The information displayed in this box includes the row number, whether the data is Numeric (N) or Text (T), and the name of the column variable. Note that the software displays data from all the tables in the current workbook. Even though data within the same table have unique column names, columns across different tables can have similar names. Hence, it is crucial that you not only specify the column name but also the table name.
3
Add or View Data:
Click on this button to add more data to your workbook for analysis or to view more details about the data listed in the available data box. When you click on this button, it opens the Data Editor dialog box, where you can import more data into your workbook. You can also switch from the list view to a table view to see the individual data values for each column.
4
Required Data:
The code for the required data specifies what data can be specified for that box. An example code is N: 2-4. If the code starts with an N, you must select only numeric columns. If the code begins with a T, you can select numeric and text columns. The numbers to the right of the colon specify the min-max values. For example, if the min-max values are 2-4, you must select a minimum of 2 columns of data and a maximum of 4 columns in this box. If the minimum value is 0, then no data is required to be specified for this box.
5
Select Button:
Click on this button to select the data for analysis. Any data you choose for the analysis is moved to the right. To select a column, click on the columns in the Available Databox to highlight them and then click on the Select Button. A second method to choose the data is to double-click on the columns in the list of Available Data. Finally, you can drag and drop the columns you are interested in by holding down the select columns using your left mouse key and dragging and dropping them in one of the boxes on the right.
6
Selected Data:
You will need to select one column for the response data and one or more columns as the factor/features for the given model. If the right number of data columns has been specified, the list box header will be displayed in black. If sufficient data has not been specified, the list box header will be displayed in red. Note that you can double-click on any of the columns in this box to remove them from the box.
7
View Selection:
Click on this button to view the data specified for this analysis. The data can be viewed in a tabular format or a graphical summary of the selected data.
Train
If you click the Train button, the software will let you pick the options for training the given model. Training is a step where we split the data into groups - a train data set and a test data set. The train data set is used to build the model, and the test data set is used to evaluate how good of a model we have built. We don't want to use the same data set for both test and train since we can have overfitting, and the model may perform excellently on the train data set but may perform poorly on data sets it has not seen.
The train function can be used for the following:
Evaluate the effect of model tuning parameters on performance
Choose the "optimal" model across these paramters
Estimate model performance from a training set
A sample screenshot of the training page is shown in this figure.
1
Method:
Specify the method to use to determine the test/train data sets. The available options are:
Option
Description
None
Do not split the data into test and train data sets. Use the entered data for training and testing purposes. Note that using this method, you cannot tune the model using tuning length or grid methods.
Boot
Use the bootstrap method for determining the test and train data sets. The bootstrap method involves repeatedly drawing samples from the data set with replacement. If you use this method, you must specify the number of resamples.
Cross Validation (CV)
Cross-validation involves dividing the available data into multiple folds or subsets and using one of these folds for validation and the remaining folds for training. This process is repeated multiple times, using a different fold for validation. Finally, the results from each validation step are averaged to produce a more robust estimate of the model's performance. If you use this method, you must specify the number of folds.
Leave One Out Cross Validation (LOOCV)
It is a method of cross-validation where each data point is used for validation, and the rest of the data is used for training. This method is very computationally expensive if you have large data sets but is simple to use and requires no configuration.
Leave Group Out Cross Validation (LGOCV)
It is a method of cross-validation where a set of data points is used for validation, and the rest of the data is used for training. You must specify this method's holdout percentage (0-100)%.
Repeated Cross Validation (RepeatedCV)
In this method the cross-validation is repeated multiple times. You will need to specify the number of folds and the number of repeats. For example, if you specify five repeats of 10-fold cross-validation, it will perform 10-fold cross-validation on the training data 5 times, using a different set of folds for each cross-validation. The rationale is to come up with better accuracy and robust results.
2
Holdout Percent:
Specify the % of data to hold out for performing validation testing. The % specified should be between 0 - 100%. Note that this value is only enabled for the LGOCV method.
3
Num K-Folds:
Specify the number of K-folds for cross-validation. If you specify 5, it means that the data is split into five groups. Hence, each group contains 20% of the data points. The first 20% of the data points are used for the first validation and the remaining 80% for training. The second 20% of the data is used for the second validation, and so on. This value is to be specified for CV and RepeatedCV methods. For the bootstrap sampling method, select the number of resamples of the data to use.
4
Num Repeats:
Specify the number of repeats for repeated cross-validations. This textbox is disabled for other methods.
5
Random Seed:
Specify the random seed value. Use a value of 0 for truly random numbers. If you want to replicate your results between runs, specify the seed value for the random number generator.
6
Sub Sampling:
Specify the sampling strategy to use when selecting the samples. In classification problems, a disparity in the frequencies of the observed classes can significantly impact model fitting. Class imbalances can be mitigated using a sampling strategy.
Option
Description
None
Do not use any sampling strategy. Select the samples at random.
Up
Use the up-sampling strategy. Randomly sample with replacement the minority class to be the same size as the majority class.
Down
Use the down-sampling strategy. Randomly subset all the classes in the training set so that their class frequencies match the least prevalent class.
Tuning
If you click the Tuning button, the software will let you pick the options for tuning the given model. Tuning refers to identifying the best set of hyperparameters fitting the given model. This method only applies to those models that contain one or more hyperparameters.
The main steps of tuning are:
A sample screenshot of the tuning page is shown in this figure.
1
Tuning Method:
Specify the method to use to determine the best set of hyperparameters. The following options are available:
Option
Description
Fixed
Do not tune the hyperparameters. Use the values specified below for the hyperparameters. It is your responsibility to provide the right values for these parameters.
Grid
Tune the hyperparameters using a grid search method. Using this method, the entire range of the hyperparameters is searched based on the parameters you provided. Make sure you provide the right range for each of the given hyperparameters.
Random
Specify the tuning length. The system will randomly sample the hyperparameters based on the tuning length. For example, if the tuning length is 3, the system will try three different settings for the hyperparameters and report the best of what it finds among these three searches.
2
Tuning Length:
This textbox is only applicable if your model has hyperparameters and if you are using the random search method to select the tuning parameters. Otherwise, this textbox is disabled.
3
Model Parameters:
Depending on the model you select for fitting to your data, you may need to specify hyperparameters for that model. Some models may not have hyperparameters, while others may have one or more parameters. For some situations where you specify the tuning method is Random, you can specify the hyperparameter as Auto and let the system select this for you. For other cases where the tuning method is Fixed, you must specify the hyperparameters for the model.
Verify
If you click the Verify button, the software will perform some checks on the data you entered. A sample screenshot of the data is shown in this figure.
1
Verify Checks:
The objective of this analysis and any checks performed are listed in this dialog box. For example, the software may check if you have correctly specified the input options and if you have specified the data correctly for analysis.
2
Check Status:
The results of the analysis checks are listed here. If the checks are passed, they are shown as a green checkmark. If the verification checks fail, they are shown as a red cross. If the verification checks result in a warning, they are shown in the orange exclamation mark. Finally, any checks that are required to be performed by the user are shown as blue info icons.
3
Overall Status:
The overall status of all the checks for the given analysis is shown here. The overall status check shows a green thumps-up sign if everything is okay and a red thumps-down sign if any checks have not passed. Note that you cannot proceed with generating analysis results for some analyses if the overall status is not okay.
Outputs
Click on Compute Outputs to update the results on the worksheet. A sample screenshot of the worksheet is shown below.
1
Notes Section:
The notes section summarizes the input data, and the analysis results section shows the response variable and input variables' names. To make predictions using the developed model, click the Make Prediction button on the main menu bar and enter the values for the input variables. When you click Compute Outputs, the model predictions are updated in the output column.
2
Graph Section:
The graph section shows a chart for the model accuracy, a plot of hyperparameter tuning optimization if applicable, and the variable importance plot showing which input factors impact the output response most.
If you want to use this model to make predictions, you must click on the Make Predictions button on the menu bar. Refer to the help file on this topic for more details.
Notes
Here are a few pointers regarding this analysis:
This analysis requires that the R software needs to be installed on your computer. Further, you will need to provide a link to the RScript executable file under Sigma Magic Options so that the software can use the R software to generate analysis results.
Examples
The following examples are in the Examples folder.
For the data in the file, perform the prototype models analysis for predicting the gear based on other variable inputs. (Prototype Analysis 1.xlsx)
References
For more information on this topic, please refer to the following articles. Do note that if any external links are mentioned below, they are for reference purposes only.