In shawngiese/classic.models: Classic Models from OpenText Analytics Designer

Although you can quickly use your data in R to get a statistical report or produce a visualization, you might want to benefit from features available in R packages. Many R packages enable building supervised, unsupervised and reinforcement learning models. Once these models are built they can be deployed for use in reports, web interfaces or through API queries.

About workflow

After you load your data into R a typical workflow for building a data model includes the following steps:

1. Load
2. Explore
3. Clean
4. Model
5. Validate
6. Deploy
7. Update

The methods you use to achieve these steps depends on your goals and resources.

Load\s\s Choosing the data loading process depends where your data is located. If your data is in CSV files you can load it with read.csv(), if it is in a report design you can use BIRT-R, if your data is in a Spark cluster you can use the R package SparkR to access content on the Hadoop File System.

Explore\s\s Explore your data One common R function is head() to see the top rows in your data, str() to display the data frame structure, and cor() and the R package corrplot to check for correlations. You can use this step to help identify columns that are relevant or if more data should be added.

Clean\s\s After exploring your available data, you might need to clean it up in preparation for training a model. Some common tasks include removing or replacing null values, creating factors from strings, and normalizing data values. Use functions like is.na() or specialized R packages such as Amelia. You can use R packages such as dplyr to organize data frames to use only the data you need. Depending on your model, you also may need to separate the data into a training and test set, such as with the R package caTools. Splitting data into a training and testing set enables you to later verify how well your model works.

Model\s\s When your data is ready, you can apply different machine learning models to find the one that best fits your data. R libraries exist to build models such as regression, nearest neighbor, random forests, clustering, neural nets and more.

Validate\s\s Validating your model is an important step to find the model that best meets your goals. Maybe the best K nearest model is still not accurate enough with your data and so you try linear regressions.

Deploy\s\s After selecting the best model for your needs, you can deploy it for use by others. Some of the possible deployment options include:

• Use the Rserve library to enable remote access to your R objects
• Export the model as an RData (rda) file for use in another R instance
• Export the model as a Predictive Model Markup Language pmml file for use in a OMML compliant application
• Save the steps to build the model in an Rscript file to reapply when needed

Update\s\s As a model is used, you may need to update it. For example, new data allows additional features, or new obeservations are available from another year of data collection.

shawngiese/classic.models documentation built on Dec. 31, 2020, 5:14 a.m.