AI-Driven Sales Forecasting & LSTM Performance Visualization

Showcase
1

Hi Matplotlib Community!

I am a first-year Data Science student, and I wanted to share a project where Matplotlib played a crucial role in evaluating my AI models. I built a Sales Forecasting System using LSTM (Long Short-Term Memory) neural networks to predict future trends based on historical data.

The Visualization Challenge

In time-series forecasting, simply seeing the numbers isn’t enough. I used Matplotlib to create a clear comparison between Actual Sales and Predicted Trends.

Key Matplotlib features I used:

  • plt.plot() with distinct styling: Used solid lines for ground truth and dashed lines for predictions to make the chart instantly readable.

  • Grid and Legend Customization: Focused on a minimalist design to ensure the data remains the hero of the visual.

  • Figure Scaling: Adjusted DPI and figsize to ensure the LSTM’s subtle trend shifts were visible.

GSoC 2026 Context

I am currently preparing my proposal for the GSoC 2026 “Indirect Transforms” project. Working on this forecasting tool helped me realize how important precise coordinate positioning is, which is why I’m excited to dive into the transform machinery and help refactor the positioning API!

I will be get confidence if any mentors can guide me over this to make it more better and useful…

Check out the code here:

I’d love to hear any feedback on how I can make these AI visualizations even more effective!

Best regards, [Rahasya Pandey]

2

Thanks for sharing. However, there’s a slight misconception here. GSoC applications are not about us telling you what to do better.

They are about you presenting your ideas and skills to convince us you’ll be able to meaningfully contribute to the project.

3

I appreciate that clarification! That makes total sense.

4

Also @Rahasya_pandey as a showcase, your repo is missing the demo image (it says insert image here) and your notebook is broken, which means that anyone who wants to check out your project has to run the code themselves.

5

"Hi @story645 , I’ve pushed a final update to the sales-forecasting-lstm repository.

  1. Fixed the ‘Invalid Notebook’ rendering issue on GitHub.

  2. Optimized the notebook by clearing unnecessary logs and ensuring a clean ‘Restart and Run All’ execution.

  3. Added the forecast visualization directly into the README.

Everything is now reproducible and documented. Thanks for the guidance!"

6

Hi Hannah (@story645), following up on our previous discussion, I’ve been diving deeper into the matplotlib.transforms module to prepare for the Indirect Transforms project.

I’m particularly interested in the CompositeTransform logic—specifically how the pipeline handles the interaction between a BboxTransform and an Affine2D transformation when data is being re-scaled. I’ve been looking at the transform_path method to see how it handles non-affine components efficiently.

Would you recommend a specific ‘Good First Issue’ related to the transforms core, or perhaps a small area of the lib/matplotlib/transforms.py file I should focus on to better understand the current bottlenecks in indirect mapping?"

7

I’m gonna be frank w/ you. I gave you feedback on your library to showcase it better b/c you posted to showcase, but your library comes across as AI generated. That’s borderline disqualifying for GSoC b/c we here at Matplotlib very much want to see what the human in the loop (you) can do.

What do you think composite transforms have to with indirect transform?

8

I wanted to apologize for the library I shared earlier. To be honest, I used AI tools to help generate parts of it, and I realize now that it didn’t showcase my own work or thought process clearly. I understand why this raised concerns about authenticity, especially for GSoC.

I’m committed to transparency and would like to revisit this topic. Could we discuss how I can improve and better demonstrate my skills? I’m eager to learn and make it right.

Thanks for your understanding and guidance.