AI Sales Astrologer

• 

  Changing loss with increasing batches

• 

  Final Accuracy

AI Sales Astrologer

This project is based on sales data from an apparel industry, which is a multivariate time series dataset. The aim is to create a predictive model using a custom architecture for forecasting sales.

Inspiration

The inspiration for this project is to create a tool that can forecast sales data for the apparel industry using AI technology.

What it does

AI Sales Astrologer is a predictive model architecture that forecasts sales for the apparel industry. It is based on a multivariate time series dataset containing sales data from 2014 to 2023. The model uses a custom multi-head CNN LSTM architecture built using Keras and TensorFlow. The model takes into account features such as Discounts and Demand to predict Gross sales for the upcoming months.

How we built it

We built AI Sales Astrologer using a combination of Python libraries such as Keras, TensorFlow, and scikit-learn. We used a multi-head CNN LSTM architecture to build the model. We also used Docker to create a custom image with determined AI base image, keras_multi_head and scikit-learn. The model was trained using Adam optimizer and different hyperparameters were tuned to find the optimal settings. We used the Determined AI platform to train and optimize the model.

Challenges we ran into

One of the challenges we faced was finding the optimal hyperparameters for the model. We experimented with different batch sizes, learning rates, and number of epochs to find the best settings. We also faced some issues with dependencies such as keras_multi_head and scikit-learn which were not included in default AI packages. We overcame these challenges by building a custom Docker image and using the Determined AI platform.

Accomplishments that we're proud of

We are proud of building a custom multi-head CNN LSTM architecture that outperforms traditional time series models. The model was able to accurately forecast sales for the apparel industry using a multivariate time series dataset and achieving the accuracy of 90.08% using the best hyperparameters on the test set, which demonstrates the effectiveness of our model. We are also proud of overcoming the challenges we faced during the development of the project and successfully training and optimizing the model using the Determined AI platform.

What we learned

Through this project, we learned how to build a custom multi-head CNN LSTM architecture for time series forecasting. We also learned how to use Docker to create a custom image with dependencies and how to use the Determined AI platform for model training and optimization. We gained experience in tuning hyperparameters and using different optimization techniques.

What's next for AI Sales Astrologer

In the future, we plan to improve the model's accuracy by incorporating more features and experimenting with different architectures. We aim to make AI Sales Astrologer a reliable tool for businesses in the apparel industry to make informed decisions about their sales strategy.

Built With

• determined
• keras
• python
• scikit-learn
• tensorflow