Machine Learning Introduction Series by Women Who Code- Notes

I am attending ML Series initiated by Women Who Code which is six week long program where each week some topics in ML are covered. I will be posting my notes and assignments for each week in this blog.

ML Intro Series 1:

Series 1 was divided into two parts ML basics and Hypothesis testing, also Introductory lab on how to use colab was provided.

ML Basics/Intro

What is Machine Learning?

• a subset of AI 
• class of computer algorithms that learns from data 
• algorithms that improve with experience
• data and outputs are provided that results in a function that maps input to output, which can be used in multiple scenarios

Why ML now?

• large computing power
• big Data available
• technologies that deal with data available
• high storage capacity
• higher RAM available
• reduction in the gap between academia and industry

Terms

• data
• features
• target variable

Types of ML

• supervised
• unsupervised
• semi Supervised
• reinforcement Learning 

Supervised Learning 

• known input and output, training examples
• unknown, function that maps input to output
• goal to find the function

Types of Supervised Learning

• regression when target is continuous
• classification when target is categorical

We basically need Supervised Learning when there is no human expert for the task, humans can't describe task, function is changing frequently or we need personalised function for each use case.

Hypothesis Testing

• Hypothesis test calculates some quantity under a given assumption. 
• The result of the quantity tells us whether assumption holds true or is violated.

Normal Distribution is a type of population distribution that is most commonly found in natural phenomena.

Conducting a hypothesis test

Test starts with an assumption that a null hypothesis, also called default hypothesis hold true and a violation this assumption called first hypothesis is also called alternate hypothesis.

P-Value

• It is a quantity that can be used to interpret the result of hypothesis test. 
• Many test statistics can be used to calculate p-value.

Alpha

• It is the significance level used to accept or reject a hypothesis.
• It is generally 5% or 0.05. Lower the alpha higher the confidence.
• Confidence is 1 minus alpha.

Errors in statistical tests

• type 1 error which is false positive
• type 2 error which is false negative

Homework

• Linking colab with github
• Test Statistics Understanding on Wikipedia
• Z- Score, two tailed test for numerical problem

ML Intro Series 2:

In series 2 of ML series, Conditional Probability, Naive Bayes, Bayesian Learning were discussed and a lab on implementation of Naive based classifier using Scikit learn was also there. 

Classification

In classification responses are categorical in nature.

Events can be

• dependent
• independent

Conditional Probability

• defines probability between dependent events
• occurrence of one event changes the probability of other event

Bayes Theorem Links

https://machinelearningmastery.com/bayes-theorem-for-machine-learning/

https://stattrek.com/probability/bayes-theorem.aspx

https://towardsdatascience.com/intro-to-bayesian-statistics-5056b43d248d

Bayes Theorem Formula

P(A|B)=P(B|A)*P(A)/P(B)

Prior Probability

Probability of an event that has occurred.

Posterior Probability

Probability of an event that is going to occur.

Naive Bayes Classifier is based on the principal of Bayes Theorem

• It assumes all features are independent of each other.
• All features contributes equally.

These assumptions can be wrong and due to these assumptions, this classifier is called naive.