antagonistic dimension
- 1: A and B are in conflict
- 2: (A+B) and C are in conflict
- 3: ((A+B)+C) and D are in conflict
Experiment. py
A = np.array([-1, 0, 0])
B = np.array([1, 0, 0])
C = np.array([0, 2, 0])
D = np.array([0, 0, 4])
-
- SD=0.1
- Lack of dimensionality causes A and B, which have the least separation in PCA, to mix and become one.
- The UMAP is clearly divided
-
- SD=0.5
- They are also placed in close proximity in UMAP by A and B data becoming inseparable in the first place.
-
- SD=0.75
- A, B, and C cannot be separated and become connected.
-
- SD=1
-
- SD=1.5
- It’s all in one piece.
- string artifact is occurring.
- Because of a lack of data?
- You can’t increase it by a factor of 10.
It looks like Variational Bayesian is needed to distinguish between them
- Implemented EM algorithm for mixed Gaussian distribution in Numpy - Yasukazu Nishio’s Hatena Diary
- Estimating clusters of mixed Gaussian distribution model by k-means method, EM algorithm and variational Bayesian respectively - Yasukazu Nishio’s Hatena Diary
- But it’s not a good idea to assume a normal distribution for real data in the first place, so it’s not a good idea.
This page is auto-translated from /nishio/対立次元 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.