Predictive Intelligence Common issuesIssue Solution Training 1. There are no rows available for the model. There are no records that satisfy the filter conditions on the solution definition. The model cannot be trained due to a lack of data. Verify that your filters are correct. 2. Too many rows for training The trainer may lose connection (times out) and not retrieve the full set of data. Adjust the filters to reduce the number of rows. 3. Cannot create a model There may not be enough data or the customer instance does not contain enough unique data. For example, customers may duplicate their data by changing 1-2 fields (i.e. category and assignment group), but keep the majority of the data the same. It makes it hard for PI to predict the right value. 4. Solution creation takes too long It takes various steps to create the final model. The time it takes to train a model depends on the framework and number of rows being processed. The trainers may be busy (not available to train), there could be some performance issues on the server or on the customer instance. Check the solution progress to determine the potential problem. 5. User authentication fails Trainer fails to authenticate to start training. Check if the sharedservice.worker is active. Solution Prediction 1. Prediction takes too long The first prediction call takes the most time. Otherwise, the problem could be caused because the model is not on the prediction server or it was not pushed to the server correctly, there is an issue with the cache, or the prediction server may be overloaded. 2. A number of predictions are not returned No values are returned. This problem can occur when the confidence value is less than the threshold setting for the solution for that outcome. The class precision/coverage change be changed to increase coverage or precision. 3. Prediction Failed There is no prediction (prediction returns will null pointer). This may be caused by an access issue or incorrect domain settings. Use the REST API Explorer to determine issues. 4. Inconsistent prediction results Different users are getting different outcomes. This is typically caused by an access issue to the Class Confidence [ml_class] and [ml_pc_lookup] tables. Instance Cloning 1. Training failure due to authentication Training may fail after a clone because of the sharedservice.worker user may have been inactivated, locked out, or the user ID has not been set. 2. Training failure due to timeout Training may timeout after a clone because the customer URL or scheduler URL for the instance is incorrect. 3. Prediction failing Predictions my fail after a clone because the artifacts stored in the Artifacts [ml_artifacts] and Attachment [sys_attachments] table were not included in the clone. Missing solutions or functionality - The required plugins have not been installed/activated. Solution training may fail due to other reasons. Please explore the list below to learn more about other potential errors you may encounter. 1. Data is skewed, a percentage of records are duplicate. 90% of the values for the input columns (i.e., short description, description) are the same. Solution: Use a dataset that has more unique data to train a solution. Note: This applies only to classification solutions. 2. Data is skewed and a percentage of records have the same output value. More than 60% of the values for the output column (i.e., category) are the same. Solution: Exclude this output value from the data set. Note: This applies only to classification solutions. 3. The output field for this classification solution contains only one value. The output column only contains one single value. Solution: Use a dataset with more values in the output column to train a solution. Note: This applies only to classification solutions. 4. The output field is empty in the training dataset. The values in the output column are empty. Solution: Use a dataset with values in the output column to train a solution. Note: This applies only to classification solutions. 5. Training completed but the solution artifacts could not be uploaded. The solution training is successful, but the solution artifacts are not pushed back to the customer instance. Solution: This issue usually occurs because of network issues. Try to train the solution again at a later time. If this issue keeps happening after two or three times of try, please contact support. Note: This applies only to classification and similarity solutions. 6. Data used is not sufficient or the input is not predictive of the output field. The data for solution training is not sufficient. Solution: Use more data for solution training. Note: This applies only to classification solutions. Ask Support to use log key [log_key] to investigate trainer logs furtherSome complex issues occur the customers cannot solve. Solution: Contact support and provide the log key and full log from the training server.