On the Predictive Model list page in Analytics Builder, a column has been added to display the ROC metric. The column displays a value only when the model’s goal field is a Boolean data type. The ROC column is empty when the model’s goal field is any other data type. ROC values are displayed in the models list with four significant digits.

TA-3639

In certain scenarios, the current method for calculating ROC values can produce inaccurate results. These scenarios can occur when a model is trained on a dataset that has unbalanced classes (for example: only 1% of values are true). In these scenarios, models are likely to predict values very close to 0 in the majority of cases (or close to 1 in cases where only 1% of values are false). These type of results can distort the ROC curve which makes comparison between models difficult. This issue has been resolved by developing a new, more accurate method for computing ROC values.

TA-3958

The option to use standard deviation from the mean to identify profiles was removed in a previous release. Instead, sub-populations in a dataset were always calculated based on Z Scores, which are adjusted for sub-population size. Z Score calculations are more likely to find large, statistically distinct sub-populations. However, using a distance from the mean calculation is better able to identify smaller sub-populations of outliers.

In this release, the Calculation Method parameter is added to the profile options that are available for generating profiles. Two calculation methods are supported: Z Scores (distance from the mean, adjusted by sub-population size) and Distance from the Mean (not adjusted for sub-population size).

TA-4097

TA-4139

TA-4151

In this release, security was improved by closing several vulnerabilities.

TA-4140

When data is appended to a dataset using either AppendDataset or AppendDatasetWithDatasetRef, subsequent jobs that run on the appended dataset show a performance slowdown. The root cause for the slowdown is a limitation in the way that appended data is handled. Each append call results in creation of a new data frame in the data storage. When a job runs, it has to read through all of the data frames.

To help resolve the performance issue, an optimize flag has been added to the append dataset services. When set to True, it allows all the data frames to be combined. The flag is set to True by default but can optionally be set to False to allow multiple data frames to be created before performing the operation to combine them.

TA-3568

During Analytics Server installation, the ThingWorx Port parameter, on the ThingWorx Connection Information page, limits values to four digits. This limitation has been removed. Port numbers can now include any number from 1 to 65535.

TA-3995

When a filter is added to the Datasets tab or the Predictive Scoring sub-tab, the dataset is not displayed according to the filter parameters. This issue has been resolved so that all filter functionality is working correctly.

TA-4019

When a create model job is running in Analytics Builder, the Job Details dialog box does not display the expected information. This issue has been resolved so that, during a model creation job, the Model and Validation sides of the Job Details dialog box each show status information about the training and validation jobs, respectively. The validation job is launched automatically when the training job is complete.

TA-4021

When a non-Administrator user creates a new dataset, or appends data to an existing dataset, the expected status messages do not display. This issue has been resolved so that dataset status messages display for non-Administrator users.

TA-4062

During some Analytics Server upgrades, the Analytics Extension utility files are being unpackaged into a default directory, even when the user has selected an alternate location for installation. As a result, the upgrade fails. This issue has been resolved so that all Analytics Extension files are saved to the user-specified installation directory.

TA-4069

When running a synchronous analysis job using the Analytics Server Connector, no error message is displayed when a row of data contains an error. The job appears in a completed state but no meaningful error message from the Analytics Server appears in the analysis job Results or on the Information tab.

This issue has been resolved so that the job itself appears in a completed state and appropriate error messages are displayed for the failing rows.

TA-4106

When running a synchronous scoring job using the Analytics Server Connector, some models generate no scoring results, even though the same model has no problem generating results in an Analytics Server real time scoring job. This difference was caused by inconsistent encoding of field names that contain certain characters. The issue has been resolved so that both Analytics Server and the Analytics Server Connector encode field names consistently.

TA-4107

When user names contain a backslash character (\), such as names formatted from directory service mappings (doman\user), dataset creation in Analytics Builder does not always succeed. To resolve this issue, user names have been removed from the dataset uploading process for all new dataset creation jobs.

TW-64141

When running a request for Signals, and specifying maxAtATime = 2, Mutual Information (MI), in relation to the goal, is not returned for all field combinations as expected. Instead, individual (one-at-a-time) MI scores are returned for all fields and then are filtered down to the top 25% most relevant fields. Two-at-a-time signals are calculated only for those fields. There is currently no way to modify this filtering behavior.

TA-729

Analysis agents must be restarted after a ThingWorx server restart. Without restart of the agent, any new simulation jobs will fail to execute.

—

When a non-time series model is uploaded using the Analytics Server Connector, and all of the properties, including the key field are mapped, a time window event does not work.

AM-1289

When filtering is applied to a list page with multiple pages (such as the Models list page), the filtered results are not aggregated for display on a single page. Instead, the filtered results remain on their original pages and multiple pages must be clicked through to find all the filtered results.

To improve handling of list page displays, a configurable parameter has been added to set the number of table items that can be displayed per page. A more permanent resolution for the filtering with pagination issue will be provided in a future release.

TA-1596

TA-1618

TA-1928

There are currently some limitations on the size and number of columns a dataset can contain. These limitations depend on the mix of data types your data includes. For specific information, see the Data section of Prepare Data and Metadata.

When trying to upload a dataset with a large number of columns, if the upload fails and an error message is generated, try the following workarounds:

TA-2599

When a field that begins with a number is included in a dataset or in metadata, most Analytics Server jobs can run successfully, such as training a model, running or scoring. However, when a server query is executed, such as a BinnedDistribution, the query fails. As a work around, avoid using field names that begin with a number.

This issue will be resolved in a future release so that field names can begin with numbers.

TA-2795

TA-2799

ThingWatcher was discontinued as a standalone SDK in 8.3. However, ThingWatcher functionality is available in the Anomaly Detection features native to the ThingWorx platform. For more information, see Anomaly Detection.

ThingPredictor software media is no longer available for download as of 8.3. However, support will continue for 8.x users until the next major release. To replace ThingPredictor capabilities, an improved set of predictive services is being introduced in 8.3. For more information, see How Predictive Scoring Using the Analytics Server Connector Works.

DataConnect and the Data Analytics Definition are not available as of 8.3. Replacement functionality will be introduced in a future release.