Bringing machine learning to a practical level in one month # 1 (Starting edition)
Introduction
I received a one-month free research period from Company, so I started studying machine learning, which I had been interested in for some time. Since it's a big deal, I'll keep a record of my learning. Today, the first day, it was a disorganized reference to the literature and learning methods for learning machine learning.
List of past articles
Goal setting
We have set the goals for one month as follows.
- You will be able to propose solutions to new problems using a machine learning approach.
- You will be able to make proposals to replace the work that people have done manually with computer work.
- Put into practice what you have proposed, or create a project that can be put into practice.
- Solve simple problems that are actually present and use them as the result of independent research.
Advance preparation
First of all, in order to spend this short month meaningfully, I prepared some books instead of the run-up before the free study period.
The book I bought (not read)
The book I bought and read
- Introduction to Strategic Data Science
- What is data science?
- What can and cannot be done
- Usefulness of data science
- If you read it first, you will feel more excited before studying!
- (If you don't read this and get excited, it's not suitable.)
- It is good to read it in the sense of judging it.
- Collective Intelligence Programming
- I read it 8 years ago, so I would like to review it.
- At that time I even started Python to read this.
- This time, I reconfirmed only the table of contents.
- Introduction to Machine Learning Theory for IT Engineers
- Leave the title. I learned the theory of machine learning broadly and shallowly.
- The reading memo has become a linked list for future learning.
- Many tools are introduced, so it is easy to practice.
- I will summarize the reading notes at the end as an appendix to this article.
Finally free research start
And today, which is the first day, I went through various sites and slides at random to decide what kind of course to study.
Seen slide
The site I saw
Stanford University Machine Learning Online Course
Machine Learning - Stanford University | Coursera
Finished WEEK 1
- The machine learning field is in demand anyway!
- Machine learning can be broadly divided into "supervised learning and unsupervised learning."
- (I'm interested in supervised learning for the time being)
- Supervised learning means that the correct answer is given.
- The list of features can be infinitely long, but there are algorithms that can handle it.
- Regression problem => Prediction of continuous value output
- Classification problem => Prediction of discrete value output
- Choose Octave as the language to use when learning machine learning!
- Fastest learning of machine learning with Octave in past achievements
- Than Python.
- Python I was trying to prepare various things ...
- Impressions after WEEK 1
- About 3 hours per WEEK.
- It will take about March if you do it according to the schedule
- If you do your best, one WEEK in the morning and another WEEK in the afternoon
- The content is similar to "Introduction to Machine Learning Theory for IT Engineers"
- However, it is easy to understand because it explains more slowly and in detail.
From tomorrow, we will be holding this Stanford machine learning course. It seems that there is up to WEEK 11, and even if you do one in the morning and one in the afternoon, it will end in a week, so I would like to finish this course before Christmas and get a feel for it.
Then, when I try to move my hand, I don't want to be context-switched because I'm stumbling or taking time in a non-essential place, so I'll review the machine learning environment in Python and create the strongest environment tomorrow. I thought, but I have to create an Octave environment as well.
What to do tomorrow
appendix
Introduction to Machine Learning Theory Reading Memo
Overview
Introduction to Machine Learning Theory for IT Engineers
Technical Review Company
Etsuji Nakai
Classification of machine learning algorithms
- Classification: Algorithm that produces class judgment
- Regression analysis: Algorithm for predicting numerical values
- Clustering: An algorithm for unsupervised grouping
- Other (not handled)
- Similar matching
- Co-occurrence analysis
- Link prediction
Review term
- Least squares
- Training set, feature variables, objective variables
- Polynomial approximation, error function
- Overfitting problem
- Maximum likelihood estimation method
- Setting the probability of data generation
- Parameter evaluation (parameter that maximizes probability)
- Estimator evaluation (match positive and universality)
- Perceptron
- The equation of the straight line that divides the plane
- Evaluation of classification results by error function
- Stochastic gradient descent-> Parameter modification by gradient vector
- Geometric interpretation
- Bias term arbitrariness and algorithm convergence speed
- Geometric interpretation / Geometric meaning of bias term
- Logistic regression
- Definition of data occurrence probability
- Determining parameters by maximum likelihood estimation method
- ROC curve
- Application of logistic regression to real problems
- Performance evaluation by ROC curve
- IRLS method
- k-means (means?)
- Basics of unsupervised learning model
- EM algorithm
- Unsupervised learning model by maximum likelihood estimation method
- Bernoulli distribution
- Bernui mixture distribution
- Clustering by EM algorithm
- Bayesian inference
- Bayes' theorem
- Application of Bayesian inference to regression analysis
Review analysis tool
Further reference books