Klera Machine Learning Service

Overview 

Klera Machine Learning services enable systems to become accurate in predicting future events by providing out-of-the-box intelligence/ability to automatically learn from given datasets (training datasets) without any explicit programming or human intervention. 

Prerequisites 

Hardware Requirements: 

Number of records per user: 50K Records/Rows 

Parameter	Specifications (Minimum Recommended)
RAM	5 GB
CPU	2 Cores or 4 Hyper threads
Disk	100 GB

Note: The RAM requirements will depend on the number of input rows/records and size of the textual content in the records.

Software Requirements: 

Parameter	Specifications (Minimum Recommended)
Operating System	Windows Server 2012 R2 (64 Bit)

Features 

1. Create Model: Learns and Creates a model based on the input data set given. 
2. Classify: Classifies the data or Predicts events/class for the data based on the model created. 
3. Association Rules: Identifies frequent patterns in data. 
4. Forecast*: Forecasts time series data. 

Note: 

• *Forecast operation requires the Python module ‘fbprophet’ to be installed. Please install the package ‘fbprophet’ using “conda install”. 
• Create model does not accept DATE/TIME/DATETIME columns as Feature columns or Class column. 
•  Models will be created in the following path:
  "C:\AnalyticsModels\PythonCoreMLEngineModels". 

Create Model 

       Operation is exposed on DST 
                              On DST>Analytics>Machine Learning > Create Mode

Input Parameters 

• Model Name:  
               Name of the model to be created. 

       

• Attributes: (Figure: 2b)  

1. Class Column: The column to be predicted. 
2. Feature Columns: The columns that will be used for predicting the Class. 
3. Filter Column: Rows that have a filter column value as TRUE will be selected for Create Model as input data. Only Boolean columns can be selected as Filter column. 
4. Percentage of Input Data to be used for Training: The percentage of data from your input DST/filtered input DST (as per the selection) that you enter in this field will be considered for creating the model. The remaining percentage of your data will be used for validating the model. For example, if you enter 100 percent then your entire data will be used for creating the model as well as validating it. If you enter 70 percent, then 70 percent of your data will be used for creating the model and 30 percent will be used for validating it.  

• Preprocess: 

       

1. Tokenizer Type: Type of the tokenizer to be applied on the DST: 
               I. NGram Tokenizer: This is used to break the text down into ngrams. (list of Tokens) 
   

   Note: An ngram is a contiguous sequence of n items from a given sample of text or speech.

               II. Word Tokenizer: It is used to break down the text into words. (list of Tokens) Note: An ngram is a contiguous sequence of n items from a given sample of text or speech. 
2. Number of words in the bag: Total number of words to be considered in the bag from input data. This is considered while building the DTM (Document Term Matrix) for creating the model. 
         -Minimum number of words=100 
         -Maximum number of words=10000 
3. Use Stop words: When selected, Stop words of the selected language will be dropped while forming the DTM. 
4. Convert words to lower case: By default, text is preprocessed in lower case. Uncheck this box, if you require upper casing. 

• Classifier:
  
  
  • Select Algorithm: Algorithm to Create model and Classify 
    I. Naïve Bayes: is used when the class column is non-numeric type. 
    II. Regression: is used when the class column is numeric type. 
  
   

Output Parameters 

• Model Update History: 
  Shows the history of the model updates. 

• Model Meta Information: 
  Gives the meta information of the model. 

• Model Evaluation Metrics: 
  Provides the evaluation metrics of the model created. 

Let’s see how to Create Model: 

In the example, we have a dataset of Google search results. 

We are creating a model to predict the “Search term” based on the feature “Summary”. 

1. Model Name= Predict Search Term 
2. Class Column= Search Term 
3. Feature Columns= Summary 
4. Filter Column= Empty 
5. Percentage of Input Data for training= 70% 
6. Tokenizer Type= NGram Tokenizer 
7. Number of words in the Bag= 10000 
8. Use Stop Words= TRUE 
9. Convert Words to Lower Case= TRUE 
10. Select Algorithm= Naïve Bayes 

On clicking Submit, model with name “Predict Search Term” would be created. 

Classify 

          Enables the user to classify the input dataset, by applying a created model. 

          Operation would be exposed on DST 
                         On DST>Analytics>Machine Learning > Classify 

Input Parameters  

• Select Model:  
  Select the model to be applied from the model(s) created. 

      

• Select Feature Columns: 
  Select the feature columns from the DST that are to be used. 
  

  Note: Columns, based on the data type used while creating the model, would be shown in the list.

     

Output Parameters 

 DST with Predicted Class based on the model selected 

Note: For each row in the input DST,

1. Most probable predicted class is given as input, in case of regression model created using Linear Regression. 
2. 3 most probable predicted classes are given as input, in case of classification model created using Naïve Bayes.  

Let’s Classify using the model created: 

            Select Model= Predict Search Term 
            Select Feature Columns= Summary 

Show Models 

            Lists the models created by the user 

The operation would be exposed on Floor: 
              Floor > Analytics >Machine Learning>Show Models 

Delete Model 

            Deletes the selected model. 

Model Details 

             Provides the model meta information. 

Model History 

             Provides the model update History. 

Rename Model 

               Renames the model. 

Note: All the above-mentioned operations are exposed on the Model name Model Name>Analytics>Machine Learning> Delete Model  Model Name>Analytics>Machine Learning> Model Detail  Model Name>Analytics>Machine Learning> Model History  Model Name>Analytics>Machine Learning> Rename Model

Association Rules 

Provides the frequent if->then patterns in the data provided. 

Operation is exposed on DST: 
             On DST>Analytics>Machine Learning>Association Rules 

Input Parameters 

• Aggregate by Column 
              Column used to make a transaction (Set of items). 
• Items Column(s) 
               Columns used to prepare item sets. 
• Minimum Support* (0.1 to 1.0) 
• Maximum Support* (>= Minimum support and <= 1.0) 
  *Support is an indication of how frequently the if-then relationship appears in the dataset. 
• Minimum Confidence* (0.1 to 1.0) 
  *Confidence is an indication of how often the if-then relationship is found to be true. 
• Maximum Number of Rules (N) 
             Maximum number of rules to be shown in the output. 
  

  Note: If multiple rules are generated with same support value, then number of rules will exceed this value.

    

Output Parameters 

• Association Rules 
              Rules as if-then sequence for items 
• Frequent Item Sets 
             Sets as frequently co-occurring items 

For more information please follow the below links: 
https://en.wikipedia.org/wiki/Association_rule_learning 
https://www.saedsayad.com/association_rules.htm 

Forecast 

Forecasts time series data 

Note: Forecast requires the Python module ‘fbprophet ‘ 
             Please install the package by using ‘conda install’ from command prompt 
                                       - Go to .\Anaconda3\Scripts folder  
                                       - Open command prompt with Administrator rights 
                                       - Run the below command **- 
                                          conda install -c conda-forge fbprophet  

**Note: This requires active internet connection

 The operation is exposed on DST 
  On DST>Analytics>Machine Learning>Forecast 

Input Parameters 

             

• Date_Column:  
                The dates in the column should be periodic. 
• ForecastColumn: 
                Column to be forecasted. 
• Steps:  
                Number of future steps to be forecasted. 

Output Parameters 

• Forecasted DST