curse of Dimensionality
case 1) few feature
- Attendance ( % )
- Internal Marks
- Assignment Marks
case 2; Too many feature
- AH
- IM
- Assignment
- Lab marks
- project score
overcome the curse of dimensionality
- feature selection
- feature Extraction
- collect more training Data
- Regularization -> unwanted
- use Domain knowledge.
underfiting ; few data
overfiting ; To many data
Datasets
| Training | Validation | Testing |
|---|
Bias and Variance
plotting s data
- scatter plots
- Bar plot
- Histreogram
- line plot.
Plotting Libraries
- Matplotlib
- pandas.
Vectorization
Gradient Descent
Types
- Batch G.D
- Stochastic G.D (S.G.D)
- Mini-Batch G.D
Resampling
Types;
- cross-validation
- a. Holdout
- b. leave one-off
- c. k. fold
- Bootstrap Resampling.
Resampling methods
1 cross validation Dataset
↙ ↘ -> 70% -> 30% validation
a. Hold out -> random (level by level)
b. Leave one-off out (cross-validation) (LOOCV)
c. K-fold.
2. Bootstrap Resampling
without repeating data
Linear Discriminant Analysis (LDA)
| parametric Model | Non-Parametric Model |
|---|---|
| 1. fixed no. of Parameter | Flexible |
| 2. ASSUME a specific Relationship | does not assume a fixed relationship |
| 3. Simple and faster | More flexible and slower. |
| 4. Need less training data | Needs more training data |
Unit-2
Principal Component Analysis (PCA)
How PCA works (steps)
- collect the dataset
- Find relationship (correlation) between features.
- create New feature (principle components)
- keep the components with Most importance.
- Remove less importance components.
Definition;
- PCA is A dimensionality reduction techquie.
- used to reduce No. of input features while keeping as Much important information as possible
- PCA converts Many Related features into small of features without losing Much information.
- New features is called principal component.
0 Comments