ResellReel Video Learning
Space Science with Python - Cassini #5: ML Sample Weights
Video details
About this video
14:28Length
319Views
15Likes
Repository: https://github.com/ThomasAlbin/Astroniz-YT-Tutorials/blob/main/%5BProject%5D-Cassini-CDA/01-Calibration/04_weighted_machine_learning.ipynb
Modeling ML models and training them is a multi-dimensional problem. Last time, we created a simple neural network to create a first iteration of a calibration function.
However, one can do a lot more to create "the perfect" model (though it never really exists). One thing to consider: the distribution of the input data by either down- or upsampling. Here, we use a method called "sample weight".
Our model does not have the most complex data ... however next time we take a look at k-folding and Keras Tuner, to complete this small task. Afterwards we consider more data for calibration purposes and will dive back into our simple "1D" function.
Credit - Thumbnail: NASA/JPL
---
Content
0:00 Introduction
1:48 Coding - Read in the
2:18 Coding - Calibration data distribution
2:46 Coding - Calibration data weighting
5:54 Coding - Training data splitting
7:45 Coding - Scaling data
7:51 Coding - Machine Learning training
9:22 Coding - Solution space in 3D
10:17 Coding - Outlook
11:12 Outlook
---
How to setup a local dev environment: https://www.youtube.com/watch?v=PWzvRDylbQA
---
There is a lot to do and to learn and I hope you will join the journey. Meanwhile, if you have questions or ideas, reach out to me via:
Mastodon: https://astrodon.social/@ThomasAlbin
Twitter: https://twitter.com/MrAstroThomas
Reddit: https://www.reddit.com/user/MrAstroThomas
GitHub: https://github.com/ThomasAlbin
Or drop a comment!
Talk to you later,
Thomas
#space #science #Python #tutorial #datascience #cassini #nasa #saturn #ml #ai #machinelearning #deeplearning
Modeling ML models and training them is a multi-dimensional problem. Last time, we created a simple neural network to create a first iteration of a calibration function.
However, one can do a lot more to create "the perfect" model (though it never really exists). One thing to consider: the distribution of the input data by either down- or upsampling. Here, we use a method called "sample weight".
Our model does not have the most complex data ... however next time we take a look at k-folding and Keras Tuner, to complete this small task. Afterwards we consider more data for calibration purposes and will dive back into our simple "1D" function.
Credit - Thumbnail: NASA/JPL
---
Content
0:00 Introduction
1:48 Coding - Read in the
2:18 Coding - Calibration data distribution
2:46 Coding - Calibration data weighting
5:54 Coding - Training data splitting
7:45 Coding - Scaling data
7:51 Coding - Machine Learning training
9:22 Coding - Solution space in 3D
10:17 Coding - Outlook
11:12 Outlook
---
How to setup a local dev environment: https://www.youtube.com/watch?v=PWzvRDylbQA
---
There is a lot to do and to learn and I hope you will join the journey. Meanwhile, if you have questions or ideas, reach out to me via:
Mastodon: https://astrodon.social/@ThomasAlbin
Twitter: https://twitter.com/MrAstroThomas
Reddit: https://www.reddit.com/user/MrAstroThomas
GitHub: https://github.com/ThomasAlbin
Or drop a comment!
Talk to you later,
Thomas
#space #science #Python #tutorial #datascience #cassini #nasa #saturn #ml #ai #machinelearning #deeplearning
Explore related learning