Digital Vineyard

Inspiration

Wine is not only a beverage in people's lives, but also a part of cultural, social, and economic life. It carries multiple meanings of tradition, celebration, communication, and enjoyment that enrich people's daily lives. More novel red wine varieties can help explore the diversity of red wine, improve product quality, satisfy the needs of different consumers, and gain a competitive advantage in the red wine market.

What we did

We generated new wine data sets using Generative Adversarial Networks (GAN). Afterward, we filtered the wine data based on outliers that potentially have novel combinations of chemical attributes. This data may help winemakers to explore different tastes and flavors and create more attractive products. This will help to improve existing products or develop new wine varieties.

How we built it

Our model is mainly created by using deep learning generative models, we first use Generative Adversarial Networks (GANs) to generate a new set of data, by creating generators, discriminators, and adversarial training to learn and train on the data we provide. We added K-Means clustering to evaluate the diversity and quality of the generated data. Finally, we use scatterplot, heatmap, box plots, and histograms for visualization and analysis.

Challenges we ran into

We initially spent a long time working on how to import wine CSV datasets. Generators, discriminators, and training models for GAN are very difficult to optimize (we need to constantly adjust the number of model layers, batch size, drop-out rate, learning rate, etc,.) We struggled with how to optimize the GAN model significantly. For example, we tried to optimize the model by improving the generator and discriminator structure, tuning the hyperparameters, and introducing regularization.

Accomplishments that we're proud of

We successfully handled the data preprocessing tasks for the wine dataset, which included addressing missing values, scaling the features, and ensuring data quality. We implemented a Generative Adversarial Network (GAN) to generate synthetic wine data. We conducted a thorough comparison between the original wine data and the generated data. We created various data visualizations, including pair plots and heatmaps. We successfully screened the wine data for potential special new varieties (available for further analysis by winemakers). We successfully analyzed the data generated.

What we learned

We observed the power of deep learning, particularly Generative Adversarial Networks (GANs), in data generation. Data generated using GANs can be used to explore new possibilities and patterns and foster innovation. For example, in materials science, drug discovery, or agricultural variety improvement, synthetic data can reveal unknown combinations of properties or behaviors. We also learned more techniques for data visualization.

What's next for Digital Vineyard

We plan to further fine-tune the code to improve its performance, including tweaking hyperparameters, trying different algorithms or optimizing the neural network structure, experimenting with feature engineering techniques to create new relevant features from existing data, etc.

Built With

• python