at-gan

Create a data augmentation tool using the 'at-gan' package for generating synthetic tabular datasets. This tool will be particularly useful for researchers and data scientists who need to augment their datasets without compromising privacy or introducing biases from real-world data sources. The application should have the following functionalities:

1. **Data Input**: Users should be able to upload a CSV file containing their original dataset. The tool should handle basic preprocessing such as handling missing values, encoding categorical variables, and normalizing numerical columns.

2. **Model Training**: Using the 'at-gan' framework, train a generative adversarial network (GAN) on the uploaded dataset. The model should learn the underlying distribution of the input data and generate new samples that are statistically similar to the training set.

3. **Synthetic Data Generation**: After training, allow users to specify the number of synthetic rows they wish to generate. The application should then use the trained GAN to produce these synthetic samples.

4. **Evaluation Tools**: Implement basic evaluation metrics to assess the quality and diversity of the generated data compared to the original dataset. Metrics could include statistical tests like Kolmogorov-Smirnov test for numerical distributions and chi-square tests for categorical distributions.

5. **Export Functionality**: Provide an option for users to export the generated synthetic data as a CSV file for further analysis or integration into other applications.

To utilize the 'at-gan' package, you'll need to leverage its capabilities for training GANs on arbitrary tabular data. Specifically, focus on how 'at-gan' handles complex tabular structures, including mixed data types (categorical and numerical), and ensures that the synthetic data maintains the integrity of the original dataset's structure and distribution.