Key Features
Dataset Management
Upload and manage datasets in multiple formats — CSV, Excel, and JSON. Each dataset shows a summary of rows, columns, data types, and distributions. From the dataset view you can directly launch EDA explorations and apply data converters.
Supported formats: .csv, .xlsx / .xls, .json
Data Exploration (EDA)
Analyze your data visually with built-in explorers before committing to a model. Available explorers include:
| Explorer | Purpose |
|---|---|
| BoxPlotExplorer | Distribution of values per column |
| CorrelationMatrixExplorer | Correlation between variables (Pearson, Kendall, Spearman) |
| DescribeExplorer | Descriptive statistics (mean, median, std dev, etc.) |
| ScatterPlotExplorer | Visual relationship between pairs of variables |
| HistogramPlotExplorer | Frequency distributions |
| WordcloudExplorer | Word frequency for text columns |
Model Training
Create experiments that associate a dataset with a task, then train one or more models in parallel. Configure hyperparameters through auto-generated forms — no schema files to write. Supported tasks:
- Tabular Classification — predict a category from structured data
- Regression — predict a continuous value
- Text Classification — assign categories to text
- Translation — sequence-to-sequence language translation
- Text-to-Image Generation — generative image synthesis
- Text-to-Text Generation — conversational and summarization models
Data Converters
Transform your data with a library of 30+ converters before training — normalization, encoding, dimensionality reduction, imputation, and more. Converters are composable into chains and powered by Scikit-learn and imbalanced-learn.
Predictions
Once a model is trained, run predictions on new data and download the results. The prediction interface works for all supported tasks.
Explainability
Understand model decisions with built-in explainability tools:
- Kernel SHAP — per-instance feature contributions
- Permutation Feature Importance — global feature ranking via permutation
- Partial Dependence Plots — relationship between a feature and the model's output
Hyperparameter Optimization
Automatically search for the best hyperparameter configuration using Optuna or HyperOpt optimizers. Optimization history, parallel coordinate plots, and feature importance plots are generated per run.
Plugin System
Extend DashAI with third-party plugins distributed as PyPI packages. Install directly from the UI — new components appear automatically in their respective sections.