> For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.umnai.com/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.umnai.com/_mcp/server. # Comparison The Comparison view compares two trained models. Use it to understand how a baseline model differs from another trained model, especially after changing training data, schema settings, training configuration, constraints, or model version. ## When to use | Use case | Description | | ------------------------ | ----------------------------------------------------------------------------------------------- | | Compare training runs | Check whether a new training run changed the model structure. | | Review model changes | See which modules or partitions were added, deleted, or edited. | | Inspect structural drift | Compare module and partition changes across two models trained from different data or settings. | | Support model review | Use structural differences to guide review before selecting a model for deployment. | ## Generate the view Generate the Comparison view from the baseline model. The model in the path is treated as the baseline model. The request body must include `comparison_model_id`, which identifies the model to compare against the baseline. ```json { "data": [ { "view_type": "MODEL_COMPARISON", "comparison_model_id": "comparison-model-id" } ] } ``` A Fern example for `MODEL_COMPARISON` should be added before replacing this request body with endpoint snippets. ## Output The response follows the shared [Response structure](../response-structure) format. The Comparison view returns multiple named dataframes inside `view_data`. | Dataframe | Description | | ------------ | -------------------------------------------------------------------------------------------------------------------- | | `summary` | High-level differences between the baseline model and comparison model. | | `modules` | Module-level differences, including added, deleted, or edited modules and changes to partition and parameter counts. | | `partitions` | Partition-level differences, including changed rules, intercepts, coefficients, and transformed features. | ## Interpret the result The Comparison view is easiest to read from broad differences to detailed differences: start with `summary`, then inspect changed modules, then review changed partitions. ### Start with the summary Use `summary` to understand which high-level model attributes changed. The `attributes` column identifies the attribute being compared. `baseline_model` and `comparison_model` show the values on each side. `delta` shows the change, and `edit_operation` describes whether the attribute was added, deleted, or edited. This is the fastest way to see whether the two models differ in overall structure before inspecting the detailed dataframes. ### Inspect module changes Use `modules` to understand how the model organisation changed. The `module_operation` columns identify the module and whether it was added, deleted, or edited. The `baseline_model` and `comparison_model` columns show the module identifiers, indexes, partition counts, parameter counts, and non-zero parameter counts on each side. The `delta` columns show the size of the change. Focus first on added or deleted modules, then inspect edited modules with large changes in `delta.n_partitions`, `delta.n_params`, or `delta.non_zero_params`. ### Review partition changes Use `partitions` to inspect changes inside modules. The `partition_operation` columns identify the rule and edit operation. The `baseline_partition` and `comparison_partition` columns show the module, rule, intercept, coefficients, and transformed features on each side. The `delta` columns show the changes to the rule expression components. Partition changes are useful when the model appears similar at the module level but behaves differently because the rules inside a module changed. ### Compare models before deployment Use the Comparison view before deployment when you need to choose between two trained models or understand the impact of a training change. For prediction-level differences, deploy the relevant model and use deployed model views to compare predictions, attributions, diagnostics, or evaluation results. ## Field reference The Comparison view returns dataframes, so individual columns are not documented as standalone API schema properties. This section describes the main columns returned in each dataframe. ### `summary` The `summary` dataframe contains high-level differences between the baseline model and the comparison model. | Field | Description | | ------------------ | ------------------------------------------------------------------------------------------- | | `attributes` | Model attribute being compared. | | `baseline_model` | Value of the attribute in the baseline model. | | `comparison_model` | Value of the attribute in the comparison model. | | `delta` | Difference between the baseline and comparison values, when a difference can be calculated. | | `edit_operation` | Type of change detected for the attribute, such as added, deleted, or edited. | ### `modules` The `modules` dataframe contains module-level differences between the two models. | Field | Description | | -------------------------------------- | ------------------------------------------------------------------------------- | | `module_operation.module_stable_label` | Stable label of the module being compared. | | `module_operation.module_features` | Features used by the module. | | `module_operation.edit_operation` | Type of change detected for the module, such as added, deleted, or edited. | | `baseline_model.module_id` | Module ID in the baseline model. | | `baseline_model.module_index` | Module index in the baseline model. | | `baseline_model.n_partitions` | Number of partitions in the baseline model module. | | `baseline_model.n_params` | Number of parameters in the baseline model module. | | `baseline_model.non_zero_params` | Number of non-zero parameters in the baseline model module. | | `comparison_model.module_id` | Module ID in the comparison model. | | `comparison_model.module_index` | Module index in the comparison model. | | `comparison_model.n_partitions` | Number of partitions in the comparison model module. | | `comparison_model.n_params` | Number of parameters in the comparison model module. | | `comparison_model.non_zero_params` | Number of non-zero parameters in the comparison model module. | | `delta.n_partitions` | Change in partition count between the baseline and comparison modules. | | `delta.n_params` | Change in parameter count between the baseline and comparison modules. | | `delta.non_zero_params` | Change in non-zero parameter count between the baseline and comparison modules. | ### `partitions` The `partitions` dataframe contains partition-level differences between the two models. | Field | Description | | --------------------------------------------- | ------------------------------------------------------------------------------ | | `partition_operation.rule_canonical_label` | Canonical label of the rule being compared. | | `partition_operation.rule_friendly_condition` | Friendly condition expression for the rule, when available. | | `partition_operation.edit_operation` | Type of change detected for the partition, such as added, deleted, or edited. | | `baseline_partition.module_id` | Module ID for the partition in the baseline model. | | `baseline_partition.module_index` | Module index for the partition in the baseline model. | | `baseline_partition.rule_id` | Rule ID in the baseline model. | | `baseline_partition.intercept` | Intercept for the partition expression in the baseline model. | | `baseline_partition.coefficients` | Coefficients for the partition expression in the baseline model. | | `baseline_partition.transformed_features` | Transformed features used by the partition in the baseline model. | | `comparison_partition.module_id` | Module ID for the partition in the comparison model. | | `comparison_partition.module_index` | Module index for the partition in the comparison model. | | `comparison_partition.rule_id` | Rule ID in the comparison model. | | `comparison_partition.intercept` | Intercept for the partition expression in the comparison model. | | `comparison_partition.coefficients` | Coefficients for the partition expression in the comparison model. | | `comparison_partition.transformed_features` | Transformed features used by the partition in the comparison model. | | `delta.intercept` | Change in intercept between the baseline and comparison partitions. | | `delta.coefficients` | Change in coefficients between the baseline and comparison partitions. | | `delta.transformed_features` | Change in transformed features between the baseline and comparison partitions. |