Troubleshooting & Error Messages

Lists possible errors and how to address them

Error message
Notes
Resolution

CSV file must not be empty

May occur if upload fails or is interrupted

Provide a CSV with content; try re-uploading

CSV file must contain headers

Headers are used to create your variable names and/or map CSV columns to your variables

Edit your CSV to include headers

Invalid header name

Requirements for header names are included in the Data-to-Model wizard

Address header name issues (such as special characters)

Found null/empty value

Currently, Genius does not handle empty cells/values for CSVs, there is a future feature planned to address this

Either remove rows which have empty values or provide values within those cells

Error parsing CSV file: {contents of CSV file}

Likely to occur if the CSV is malformed or does not conform to CSV data format

Ensure CSV adheres to the required format; if problem persists, report an issue via your Customer Portal

Column {number} appears to not have headers (header name same as data values)

Warning will not prevent additional actions; continuing is likely to result in incorrect inferences

Ensure all column headers are present and correct

Column {number} appears to contain dates; these will be treated as categorical variables and will likely be useless for inference

Warning will not prevent additional actions; continuing is likely to result in incorrect inferences; there is a future feature planned to make use of dates/times

Ensure column does not include dates

Column {number} contains both numeric and categorical data; this may lead to incorrect inference

Warning will not prevent additional actions; continuing is likely to result in incorrect inferences

Ensure values in the column are accurate the represent one set/type of observation data

Error message
Notes
Resolution

Invalid variable name: {variable}

Update variable names such that they do not contain any spaces or special characters

Variable {variable} must have at least 1 value

Ensure the variable's possible states are inputted

Factor {factor} must have at least one variable

Ensure the factor connects to at least one variable

Factor {factor} must have at least one probability value

Genius does provide assistance for initializing a factor with probabilities before any learning takes place if you do not have values to input manually

Ensure the factor contains some probability (initial or learned)

Factor variable {variable} is not defined in variables

When manually creating a model, it is possible to refer to a variable that has not yet been created

Ensure all variables named in your factors are represented as variables in your model

Factor {factor} 's tensor shape {actual_shape} is incompatible with its variable cardinalities {expected_shape}

Ensure the factor's tensor accounts for the correct number of variable state combinations

Tensor values for each target variable element must sum to 1.0 for categorical distributions. Found sum of {sum} for factor {factor}

Ensure use of the right kind of distribution, update parameter values such that they sum to 1 if correct, otherwise update the distribution type to conditional

There must be a distribution over variable {variable}

Connect the variable to a factor and ensure that factor has an accurate probability table

The graph representing a Bayesian Network must be acyclic

Variables cannot depend (even by an inheritance chain) on variables which depend on them (variable a cannot depend on variable b if variable b also depends on variable a)

Remove any bidirectional dependencies in your model

Variables {variables} are connected by a factor whose distribution must be conditional

Connect the variables via a conditioned factor

Factors involving variable(s) {variables} have role {role} which is undefined for this model type

Likely to occur if the model is specified as a Bayesian Network or Markov Random Field but the model includes an action variable which can currently only be used with POMDPs

Correct the roles associated with the variables to match the model type or update the model type to support the variable roles; likely will require additional updates to the model

Factor {factor} represents a conditional distribution. Markov random fields use symmetric potential functions (categorical distributions)

Specific to Markov Random Fields

Update factors to categorical distributions

Variable(s) {variables} lack transition factors

Add/Connect state-transition factors to variables

State variable(s) {variables} participate in no likelihood factors

Specific to POMDPs; May indicate that the variables are not strictly necessary for the model

Add/Connect likelihood factors to variables

Observation variable(s) {variables} lack likelihood factors

Specific to POMDPs; May indicate that the variables are not strictly necessary for the model

Add/Connect likelihood factors to variables

Role for variable(s) {variables} cannot be inferred from VFG information

Explicitly set the variable role as action, latent, or observed

POMDP contains no transition factors

Add a state-transition factor and connect it to an action

POMDP contains no likelihood factors

Add likelihood factor(s) and connect it/them to observed variables

State factors {variables} not connected to a control variable

Add/Connect a latent state variable to the state-transition factor

Errors occurred during model validation. Catch this exception and check the .errors field for details. VFG.apply_patches(errors) will automatically correct recoverable errors.

General error message when validating your model fails but none of the cases above are found

Varies, see .errors; if error persists, report an issue via your Customer Portal

Other possible issues

Error message
Notes
Resolution

Last updated