Altair SmartWorks Analytics

 

Fixed and Known Issues

This help page describes fixed and known issues in the current release of SmartWorks Analytics

Fixed Issues

Issue No.

Description

SWA-44

Model Evaluation UI does not load if 'actual' column is missing from the input dataframe

SWA-46

Model Evaluation UI does not raise an issue and displays the wrong model if there are multiple champion models in the input dataframe

SWA-48

Column Change: Apply button is disabled when replacing null with decimal values (pyspark)

SWA-49

Model Evaluation- Metrics chart not loaded in full size

SWA-52

No data are plotted when calculating cumulative gains and lift when the 'actual' column has integer type

SWA-54

Can't remove Prediction Type after setting it

SWA-55

Error calculating probability-based classification metrics when at least one input record has a probability outside the interval [0.0, 1.0]

SWA-57

Mode evaluation node throw error if dataframe is changed

SWA-512

Import file: data preview takes a long time for unsupported delimiter files

SWA-304

Incorrect data point in Model Evaluation KS plot for reference binary classification dataset

SWA-856

GraphQL error when merging using "\" separator.

 

Known Issues

Issue No.

Description

SWA-36

Overlapping x-axis labels in object node viewer Chart section for numeric column (Pyspark only)

SWA-40

Pivot (pandas) - Boolean data type with nulls in columns to include returns error on unchecking Aggregate checkbox

SWA-45

EMR: session is not failed in case when lost connection with server

SWA-852

PySpark models generated or zipped in Windows environment fail to deploy

SWA-853

Revoke Access-Shared file is still available in D folder and can be used ,if the file was already used in any of the Knowledge Studio plugin

SWA-859

Aggregate Object Node- Incorrect error message in Chart for STD/Var columns with 'nan'

SWA-1071

Object Node- Chart x-axis & filter slide values chopped

SWA-1088

Bit Datatype column values are Nulls in Pandas

SWA-1095

Column Changes node fails to convert object column into datetime column

SWA-1102

Library: Data cannot be retrieved from the file from folder having name with “/”

SWA-1127

Temporary folder is getting created in Spark Export_Text to Library

SWA-1835

About section - Download link is not functional with timezone locale with integer value offset

SWA-1887 / SWA-1889

Deployments with autoscaling – autoscaling not being triggered unless resource usage is at least 5% different from the target (for both scale-out and scale-in).

 

Third-Party-Dependent Issues

Issue No.

Description

SWA-1133

Incorrect result with Nulls in dataset when using ‘Right Outer Join’ method (Pandas).

Root Cause: Nullable columns having Nulls

Current workaround: Replace Nulls in the dataset with a different value

SWA-1154

The data preview does not load if the column comprises of an Integer value with 13+ digits (Spark)

Root Cause: Older versions of PySpark does not support this

Current workaround: Version of deployed PySpark needs to be >= 2.4.8 or 3.0

SWA-1075

The First and Last values are incorrect if all functions are applied on a column with Date datatype. (Spark)

Root Cause: First and Last aggregation methods using F.countDistinct returns an incorrect value.

Current workaround: In the Column Change node, cast the column to String and then to Datetime and Run node. Attach the Aggregate node to output and all aggregation methods are properly computed.

SWA-2015

When sending a request to a model endpoint at a high rate and volume, the seldon model server can occasionally crash. (Note that Kubernetes will automatically revive it, though there may be some down time).

Root Cause: With the default settings (flask and gunicorn), the seldon model server can only handle a limited rate and volume of requests.

Current workaround: There are two workarounds: (1) Increase the number of pods to have some extra in case some of the pods fail. (2) Set default (Admin / DevOps task) environment variables for all containers deployed to seldon to increase the load the model servers can handle.

 

Other Issues

Creating New Sessions

  • Once a session has been created, you may need to wait ~1 minute for all of the jars to be distributed to the cluster nodes prior to usage in a workflow. Otherwise, the following error may display in the Node Viewer or when you attempt to run the node:

  • {"graphQLErrors":[{"message":"Unexpected error value: { message: \"Did not observe any item or terminal signal within 30000ms in 'map' (and no fallback has been configured)\", locations: [[Object]], path: [\"getRawPreview\"], extensions: { type: \"GENERAL\", code: \"UNEXPECTED_ERROR\", message: \"Did not observe any item or terminal signal within 30000ms in 'map' (and no fallback has been configured)\", traceId: null, requestId: null, details: [Object] }

 

COLUMN NAMES

  • Column names may include all alphanumeric (i.e., A-Z, a-z, 0-9) and some special characters (e.g., - _ | \ / @ & *). Whether a special character is valid for a column name is engine-dependent.

  • Valid special characters for Apache Spark: ~ ! @ # $ % ^ & * ( ) _ - + = { [ } ] | : ; < > ? . /

  • Valid special characters for Pandas: ~ ! @ # $ % ^ & * ( ) _ - + = [ ] | : ; < > ? . /

  • The characters ` (back tick), ' (single quote), " (double quotes), and \ (backslash) are similarly invalid characters for Apache Spark and Pandas. In addition, { } (curly brackets) are invalid for Pandas.

  • If the column name includes an invalid character, the Apply button is disabled.

Text Imports in Apache Spark

  • The data preview will not load if a column has integer values of 13 digits.

Filtering Columns In Apache Spark

  • When the columns of tables opened using the Apache Spark engine are filtered using the value =, some filter elements may not display in the Values field if these elements are not included in the rows selected for the data preview table. You can continue setting up your Filter node by manually typing in the filter element(s) you wish to apply in the Values field and then saving and running the node configuration. The resulting Data Frame node will display the correctly filtered table.

Viewing Folders In Remote DeskTop Sessions

  • Empty folders as well as folders that do not contain data files in the SmartWorks Analytics Library do not display in the D:/ drive during a remote desktop session. You can work around this issue by creating new folders in the Windows virtual machine (Altair Desktop Engine) that you intend to fill during your remote session.

Configuration Errors in the Node Viewer

  • When an invalid configuration in any of the Node Viewers is saved, an error message displays. This message persists until a valid configuration is provided and the “Save” button is clicked once more to save the updated settings.