📊 results.csv
What is a CSV?
A CSV (Comma-Separated Values) file is a simple, text-based file format used to store tabular data. Each line represents a row, and each value is separated by a comma. CSVs are widely used because they are human-readable and compatible with many tools.What is results.csv
?
In OmniOpt, the results.csv
file is used to store the outcomes of all executed trials. It contains:- The hyperparameter configurations tested during the optimization process.
- The results (objective values) produced by these configurations.
- Metadata about the optimization process such as the trial index, arm name, trial status, and generation method used.
Example content:
trial_index,arm_name,trial_status,generation_node,RESULT,int_param
0,0_0,COMPLETED,SOBOL,-41764.12,-59
1,1_0,COMPLETED,BOTORCH_MODULAR,-916.12,10
Column Explanations
trial_index
A unique, sequential number identifying the trial.
arm_name
A name identifying the specific configuration (or arm) tested during the trial.
In Ax (the optimization framework used by OmniOpt), an arm represents a single set of parameter values to be evaluated.
trial_status
Indicates the current or final state of a trial. Possible values are:- COMPLETED – The trial ran successfully and returned a result.
- FAILED – The evaluation failed due to an error (e.g., crash, invalid config).
- ABANDONED – The trial was skipped or terminated early, often due to constraints (post-generation-constraints) or poor performance.
- RUNNING – The trial is still in progress and has not yet returned a result.
generation_node
This shows which model or strategy was responsible for generating the configuration. It reflects the generation logic used for that trial.
Generation Models (generation_node
)
Below is a list of possible values in the generation_node
column and a brief description of each; they correspond to the models available:| Generation Node | Description |
|---|---|
| MANUAL | A configuration that was inserted by previous jobs. |
| SOBOL | Sobol sequence sampling – a quasi-random method for uniform space coverage. |
| SAASBO | SAASBO – Sparse Axis-Aligned Subspace Bayesian Optimization. |
| UNIFORM | Uniform random sampling – purely random selection of parameters. |
| LEGACY_GPEI | Gaussian Process Expected Improvement (legacy version). |
| BO_MIXED | Bayesian Optimization with mixed models – hybrid of multiple models. |
| TPE | TPE (Tree-structured Parzen Estimator) is a sequential model-based optimization algorithm used for hyperparameter tuning. |
| RANDOMFOREST | Random Forest model used for prediction and optimization. |
| EXTERNAL_GENERATOR | A trial configuration provided by an external generator (custom logic). |
| BOTORCH_MODULAR | A modular, flexible BoTorch strategy – highly customizable. |
These options are color-coded in the GUI for visual clarity.
Summary
Theresults.csv
file is central to understanding what was tried, how it was generated, and how it performed. It enables:- Tracking progress and failures.
- Analyzing performance across generations.
- Debugging model behavior and identifying patterns.
