deepdetect VS Pandas

For those seeking a lightweight solution for setting up deep learning REST APIs across platforms without the complexity of Kubernetes, Deepdetect is worth considering.

But you need to have good reasons to do it. Ours is that we have a multi-backend framework, and that we don't want any step in between dev & run. C++ allows for this since the same code can run on training server and edge device as needed. It also allows for building full AI applicatioms with great performances (e g. real time) We dev & use https://github.com/jolibrain/deepdetect for these purposes and it serves us very well, but it's not the faint of heart !

- Real-time is achieved through our full C++ Open Source backend DeepDetect, https://github.com/jolibrain/deepdetect. We use CUDA along with OpenCV and TensorRT to chain multiple models (ring detection and generator mostly), and we make sure the data remain within CUDA memory at all time. This allows us to reach ~60 FPS on 1080Ti and 20% more on average on an RTX3090.

Interesting, thanks. Recently benchmarked inference with Vulkan/MoltenVK/NCNN, M1 GPU is roughly 30% faster than M1 CPU, https://github.com/jolibrain/deepdetect/pull/1105 for single batch inference (NCNN does not really support batch size > 1).

Python is a natural fit for serverless development. It boasts a vast array of libraries, including Powertools for AWS and robust libraries for data engineers. Its versatility and excellent developer experience make it a top choice for serverless projects, offering a seamless and enjoyable development experience.

In data analysis, managing the structure and layout of data before analyzing them is crucial. Python offers versatile tools to manipulate data, including the often-used Pandas reset_index() method.

Dash is a Python framework that enables you to build interactive frontend applications without writing a single line of Javascript. Internally and in projects we like to use it in order to build a quick proof of concept for data driven applications because of the nice integration with Plotly and pandas. For this post, I'm going to assume that you're already familiar with Dash and won't explain that part in detail. Instead, we'll focus on what's necessary to make it run serverless.

there is pull request to integrate in both pydantic extra types and into pandas cose [1]

[1]: https://github.com/pandas-dev/pandas/issues/53999

Last year I worked through the challenges using VisiData, Datasette, and Pandas. I walked through my thought process and solutions in a series of posts.

pandas: A library for data analysis in Python

Data analysis involves scrutinizing datasets for class imbalances or protected features and understanding their correlations and representations. A classical tool like pandas would be my obvious choice for most of the analysis, and I would use OpenCV or Scikit-Image for image-related tasks.

Pandas, a powerful data manipulation library in Python, has become an essential tool for data scientists and analysts. One of its key functions is read_csv(), which allows users to read data from CSV (Comma-Separated Values) files into a Pandas DataFrame. In this tutorial, brought to you by CodesWithPankaj.com, we will explore the intricacies of read_csv() with clear examples to help you harness its full potential.

So, what I'd like to do is write a documentation package in Python to recreate what I've lost. I plan to build upon the fantastic python-docx and docxtpl packages, and I'll probably rely on pandas from much of the tabular stuff. Here are the features I intend to include:

Weirdly most of my time spent with data analysis was in the C layers in pandas.