Status Updates From Python Machine Learning

Back to work, back to reading about python and machine learning 😆

LDA for topic modeling

NLP

NLP

Introducing Learning Curves with an example

Feature Selection in ML

Feature Selection in ML

Feature Selection in ML

L1 vs L2 regularization

Feature Scaling:
Standardization
MinMax
RobustScaling

124

One hot encoding for class labels

Explaining kNN classifier

Explaining kNN classifier

The theory and logic behind Random Forest

The theory and logic behind Random Forest

Decision trees split the data on each feature in order to increase "order" or information gain. This is in a sense how much pure this split will be compared to the sample before splitting

Decision trees split the data on each feature in order to increase "order" or information gain. This is in a sense how much pure this split will be compared to the sample before splitting

Decision trees have a great advantage of interpretability

for non-linear problem, SVM has a trick: The kernel trick. and the most famous kernel is the RBF kernel

Logistic Regression often gives similar yield to linear SVM. However,
1. LR is more prone to outliers
2. LR is easier to interpret
3. LR is easier to update, which might make it more attractive to streaming data

Logistic Regression often gives similar yield to linear SVM. However,
1. LR is more prone to outliers
2. LR is easier to interpret
3. LR is easier to update, which might make it more attractive to streaming data

SVM is a generalization of perceptron, where instead of minimizing classification errors, we know maximize the hyperbolic margin

L2 regularization

L2 regularization.

Logistic Regression is actually a classification algorithm

Logistic Regression is actually a classification algorithm

Logistic Regression is actually a classification algorithm

60

3 Types of ways to find local minima:
Batch: Updating weights over the whole sample
Stochastic: Updating one example at a time
Mini Batch: A compromise between both previous examples