Models
Diafunc supports automatic modeling through Ufinq, our Symbolic Regression technology. The platform analyzes your data fully automatically and creates appropriate models. A model provides deep insights into the underlying relationships and can also be used for predictions.
Symbolic regression is a model-discovery method that searches the universal function space for the underlying generative function of a given dataset. The resulting model is a mathematical expression which approximatively describes the relationship between dependent variables (aka labels) and independent variables (aka features).
It aims to find generalizable, simple and explainable models describing a set of labeled observations. The aim of the method is to find particularly generalizable models that also provide the user with insight. This method adds value by providing white box models rather than black box models, which are currently very common (e.g. deep learning).
Symbolic regression is a model-discovery method that searches the universal function space for the underlying generative function of a given dataset. The resulting model is a mathematical expression which approximatively describes the relationship between dependent variables (aka labels) and independent variables (aka features).
It aims to find generalizable, simple and explainable models describing a set of labeled observations. The aim of the method is to find particularly generalizable models that also provide the user with insight. This method adds value by providing white box models rather than black box models, which are currently very common (e.g. deep learning).
Ufinq Technology
Providing the best automatic modeling and symbolic regression technology (aka Ufinq) is a long-term goal of Diafunc as symbolic regression represents an essence of data science, to extract knowledge/insights/understanding from data.
Ufinq
Manual Modeling
You can build up your model from scratch by manually providing all observations.
Manual Modeling
Modeling based on Selections
Usually modeling is performed on selections as this works on all kinds of data sources (tables, element attributes, ...) and is more fluent.
Modeling based on Selections
Modeling based on Tables
Tables can be easily used as data source by selecting their rows.
Modeling based on Tables
Live Output
During the modeling, the current state is output for each generation. This includes the relevant metrics (quality, complexity & error) as well as the current top 10 solutions.
Result
After the successful completion of the modeling, the result is displayed. This includes the solution found in the form of a symbolic expression, as well as the relevant metrics (quality, complexity, error & time)
Held-out prediction
Modeling deterministically holds out the first rows of the dataset so the resulting model can be evaluated on data it has never seen. Applying the model to the held-out rows produces actual-vs-predicted pairs that reveal how well it generalises — far more honest than training error alone.
Automatic analyses use the held-out rows to render a scatter chart of actual vs. predicted values inside the per-target Prediction sub-chapter.
Automatic analyses use the held-out rows to render a scatter chart of actual vs. predicted values inside the per-target Prediction sub-chapter.
Feature importance
Each model exposes per-feature importance metrics that explain which inputs actually drive the output:
In automatic analyses, the top-5 features per model are surfaced in the Findings narrative; the full set is available from the FDK on the model entity.
| Metric | Description |
|---|---|
| Permutation importance | How much the model's accuracy drops when the values of a single feature are shuffled. |
| Sensitivity | How sharply the model's output changes as a single feature changes. |
| Occurrences | How many times a feature appears inside the symbolic expression. |
| Residual correlation | Correlation between a feature and the model's residual error — a non-zero value hints that the feature could still be better exploited. |
In automatic analyses, the top-5 features per model are surfaced in the Findings narrative; the full set is available from the FDK on the model entity.
Plain-English explanations
Models can also be restated in plain English for a non-technical audience. In automatic analyses this restatement is produced by the platform's language model and leads with the top features — so a reader who skips the formula still gets the headline insight.