Have you ever thought about how many open source data science libraries you would need to know to be able to reproduce some of the basic functionalities of the Peltarion platform? If so, you probably work at Peltarion...If not, are you looking for a job? ;)
How many open source libraries does it take to make one Peltarion Platform?
The more natural thing is of course that you haven't pondered this before - there really is no reason to. But we think it's an interesting question nonetheless, so we have created a short list for you.
Here are some of the libraries* you would need to know and string together, to have similar functionalities as the platform:
Data view
- Pandas to manipulate and analyze tabular data
- Pillow or OpenCV for fanciful image manipulations
- Numpy for more surgical data manipulations
- Matplotlib / Plotly for visualizing the data
- DVC for the versioning
Modeling view
- Tensorflow, Keras, Pytorch or similar for the modeling
- MLFlow for the versioning and experiment tracking (incl. code, data, config, and results)
Evaluation view
- Tensorboard, Sacred or Matplotlib / Plotly for visualizing the results
- MLFlow for the versioning and experiment tracking (mainly the results)
Deployment view
- Onnx to export / import your model into your preferred coding framework
- Tensorflow, Keras, Pytorch or similar for serving your model
- MLFlow, Docker for packaging / containerizing the model
- Kubernetes for container orchestration
* I can hear you furiously typing a comment already :) Before you do that, let me say that this list is obviously a bit tongue in cheek. I’m by no means saying that these libraries are the chosen ones that everyone needs to use and that there are no other tools for each of the steps above (or that you can do everything with fewer libraries).
Rather, the point I was trying to make is that the platform offers users a lot of functionalities that they would have to otherwise build themselves, likely, using multiple tools. And that was part of our motivation behind building the platform: helping users get rid of mundane, repetitive and non deep learning related tasks like building visualizations for your data, dealing with version handling, implementing/debugging code, etc.
This is not always obvious to non-users, so this was our way of trying to let you know about this.
But hey, maybe you don’t know our product, so this article probably didn’t make much sense to you so far. If that’s the case, no worries, we have a neat overview of our platform right here:
P.S. did you know that the platform also offers you all the training infrastructure (i.e. GPUs) and cloud deployment infrastructure that you need when you create an account? There is nothing to set up or configure, so you can get started right away. Beats even the best of laptops, don’t you think? ;)