Please enable JavaScript.
Coggle requires JavaScript to display documents.
Data reduction (dimensionality reduction (Autoencoder (Undercomplete…
Data reduction
dimensionality reduction
-
=Feature selection : reduce # of patterns in the patterns, easier to understand
Autoencoder
非線性轉換,透過類神經網絡方式做reduction
Unlike PCA,
autoencoder uses activation functions to achieve non-linearity mapping
-
-
-
-
numerosity reduction
-
nonparametric models
histograms
Divide data into buckets and store average (sum) for each bucket. Can be constructed optimally in one dimension using dynamic programming
clusters
Partition data set into clusters, and one can store cluster representation only
-
-
-
-
統初講義 P.14
實作:
- 計算出資料的covariance matrix (Σ)
- 計算covariance matrix的eigenvector 跟eigenvalue,將其從大排到小 maximize
- 拿前k個eigenvector(ak)跟現有資料內積做出新的k 個 feature zk
For 𝑘 1, 2, ... , 𝑝 the k-th PC is given by 𝑧𝑘 = 𝜶'𝑘 𝒙, where 𝜶𝑘 is an eigenvector of Σ corresponding to its k-th largest eigenvalue 𝜆𝑘. If 𝜶𝑘 is chosen to have unit length (i.e., 𝜶'𝑘 𝜶𝑘 = 1), then 𝑉𝑎𝑟(𝑧𝑘) = 𝜆𝑘
-
the data are replaced by alternative, smaller representations
小波轉換原則(照片壓縮):小波轉化是一種壓縮的前置動作,類似於JPEG的壓縮方法,將整張影像越重要的地方放在左上方,越不重要的放在右下方。
左上方(SS)所有的值皆為累加,可以只留下左上方的值
-
-
Matrix V in the SVD decomposition. For a new data element x, xV defines the
transformed data.
-
-
-