Principal Component Analysis (PCA) in ML Python - Model Metrics & Evaluation

Metrics & Evaluation - Principal Component Analysis (PCA)

Which metric matters for Principal Component Analysis (PCA) and WHY

PCA is a method to reduce data size by keeping the most important parts. The key metric is explained variance ratio. It tells us how much of the original data's information is kept in the new smaller set. We want this number to be high, so we don't lose important details.

Confusion matrix or equivalent visualization

PCA does not use a confusion matrix because it is not a classification method. Instead, we look at the explained variance ratio for each principal component. For example:

Component 1: 60% variance explained
Component 2: 25% variance explained
Component 3: 10% variance explained
Component 4: 5% variance explained

Total variance explained by first 2 components: 85%

This means the first two components keep 85% of the original data's information.

Tradeoff: Number of components vs Information kept

Choosing more components keeps more information but makes data bigger and harder to use. Choosing fewer components makes data smaller but loses some details.

For example, if you keep only 2 components out of 10 original features, you might keep 85% of information but lose 15%. This is a tradeoff between simplicity and accuracy.

What "good" vs "bad" metric values look like for PCA

Good: Explained variance ratio above 80% with few components. This means you keep most information with less data.

Bad: Explained variance ratio below 50% with many components. This means you lose too much information or need many components, defeating the purpose of PCA.

Common pitfalls when evaluating PCA

• Ignoring explained variance ratio and choosing components arbitrarily.
• Using PCA on data without scaling features first, which can distort results.
• Assuming PCA improves model accuracy directly; it only reduces data size.
• Overfitting by keeping too many components, losing the benefit of simplification.

Self-check question

Your PCA model keeps 3 components out of 10 and explains 40% of the variance. Is this good? Why or why not?

Answer: This is not good because 40% variance explained is low. It means most information is lost. You should try to keep more components or check data preprocessing.

Key Result

Explained variance ratio shows how much original data information PCA keeps; higher is better for effective dimensionality reduction.

Practice

1. What is the main purpose of Principal Component Analysis (PCA) in machine learning?

easy

A. To reduce the number of features while keeping important information

B. To increase the number of features for better accuracy

C. To split data into training and testing sets

D. To normalize data by scaling features between 0 and 1

Solution

1. Step 1: Understand PCA's goal

   PCA aims to reduce many features into fewer main features without losing important information.

2. Step 2: Compare options with PCA's goal

   To reduce the number of features while keeping important information correctly states PCA reduces features while keeping key info. Other options describe different tasks.

3. Final Answer:

   To reduce the number of features while keeping important information -> Option A

4. Quick Check:

   PCA purpose = reduce features + keep info [OK]

Hint: PCA simplifies data by reducing features, not increasing them [OK]

Common Mistakes:

• Thinking PCA increases features
• Confusing PCA with data splitting
• Mixing PCA with data normalization

2. Which of the following is the correct way to import PCA from scikit-learn in Python?

easy

A. import PCA from sklearn.decomposition

B. import decomposition.PCA from sklearn

C. from sklearn import PCA

D. from sklearn.decomposition import PCA

Solution

1. Step 1: Recall correct import syntax for PCA

   In scikit-learn, PCA is in the decomposition module, imported as: from sklearn.decomposition import PCA.

2. Step 2: Check each option's syntax

   from sklearn.decomposition import PCA matches the correct Python import syntax. Others are invalid or incorrect.

3. Final Answer:

   from sklearn.decomposition import PCA -> Option D

4. Quick Check:

   Correct import = from sklearn.decomposition import PCA [OK]

Hint: Remember: PCA is in sklearn.decomposition module [OK]

Common Mistakes:

• Using wrong import order
• Trying to import PCA directly from sklearn
• Incorrect Python import syntax

3. Given the following Python code using PCA, what will be the shape of X_pca?

from sklearn.decomposition import PCA
import numpy as np
X = np.random.rand(100, 5)
pca = PCA(n_components=2)
X_pca = pca.fit_transform(X)

medium

A. (100, 5)

B. (2, 100)

C. (100, 2)

D. (5, 2)

Solution

1. Step 1: Understand input and PCA parameters

   Input X has shape (100 samples, 5 features). PCA is set to reduce to 2 components.

2. Step 2: Determine output shape after PCA transform

   fit_transform returns transformed data with shape (samples, n_components), so (100, 2).

3. Final Answer:

   (100, 2) -> Option C

4. Quick Check:

   Output shape = (samples, components) = (100, 2) [OK]

Hint: Output shape = (samples, n_components) after PCA transform [OK]

Common Mistakes:

• Confusing samples and features in output shape
• Mixing up n_components with original features
• Thinking PCA changes sample count

4. Identify the error in this PCA code snippet:

from sklearn.decomposition import PCA
import numpy as np
X = np.random.rand(50, 4)
pca = PCA(n_components=5)
X_pca = pca.fit_transform(X)

medium

A. n_components cannot be greater than number of features

B. Random data cannot be used with PCA

C. fit_transform should be called before PCA initialization

D. Import statement for PCA is incorrect

Solution

1. Step 1: Check PCA parameter n_components vs data features

   Data has 4 features, but n_components is set to 5, which is invalid.

2. Step 2: Validate other code parts

   Random data is fine, fit_transform is correctly called, and import is correct.

3. Final Answer:

   n_components cannot be greater than number of features -> Option A

4. Quick Check:

   n_components ≤ features = True [OK]

Hint: n_components must be ≤ number of original features [OK]

Common Mistakes:

• Setting n_components larger than features
• Misunderstanding fit_transform order
• Thinking random data is invalid

5. You have a dataset with 10 features and want to keep 95% of the variance using PCA. Which approach is best to choose the number of components?

hard

A. Set n_components=5 because it's half of 10 features

B. Set n_components=0.95 to keep 95% variance automatically

C. Set n_components=10 to keep all variance

D. Set n_components=1 to simplify data as much as possible

Solution

1. Step 1: Understand PCA's explained variance ratio parameter

   Setting n_components as a float between 0 and 1 tells PCA to select components to keep that fraction of variance.

2. Step 2: Evaluate options for keeping 95% variance

   Set n_components=0.95 to keep 95% variance automatically uses n_components=0.95, which automatically selects the right number of components. Others do not guarantee 95% variance.

3. Final Answer:

   Set n_components=0.95 to keep 95% variance automatically -> Option B

4. Quick Check:

   n_components=0.95 keeps 95% variance [OK]

Hint: Use n_components as float for variance ratio in PCA [OK]

Common Mistakes:

• Choosing fixed number without variance check
• Setting n_components too low or too high blindly
• Confusing number of features with variance percentage