Photo by the author Ideogram
You do not need a strict degree of mathematics or computer science to get to the data of data. But you must understand the mathematical concepts behind the algorithms and analyzes that you will apply every day. But why is it tough?
Well, most people approach the mathematics of data on data back. They add a bit of abstract theory, are overwhelmed and giving up. True? Almost all mathematics needed to learn data is based on the concepts you already know. You just have to connect dots and see how these ideas solve real problems.
This road map focuses on mathematical foundations, which are actually essential in practice. Lack of theoretical holes of rabbits, no unnecessary complexity. I hope you find it helpful.
Part 1: Statistics and probability
Statistics are not optional in learning data. In this way, you separate the signal from the noise and claim that you can defend yourself. Without statistical thinking, you simply guess educated with the facilitate of fancy tools.
Why does it matter: each set of data tells the story, but statistics facilitate to find out which parts of this story are true. After understanding the distribution, you can immediately detect data quality problems. When you know hypothesis testing, you know if your test results A/B really mean something.
What you learn: Start with descriptive statistics. As you probably know, it includes agents, median, standard deviations and quarter. These are not just summary numbers. Learn to visualize distribution and understand what different shapes say about data behavior.
Then there is a probability. Learn the basis for probability and conditional probability. Bayes may seem a bit tough, but it is only a systematic way to update your beliefs with recent evidence. This pattern of thinking appears everywhere, from detecting spam to medical diagnosis.
Testing hypotheses gives a framework for sending essential and possible claims. Learn T tests, Chi-Kwadrat tests and confidence intervals. More importantly, understand what the PI values actually mean when they are useful compared to misleading.
Key resources:
Coding component: Utilize Python Scyka.stats and Panda for practical practice. Calculate the summary statistics and run appropriate statistical tests on real data sets. You can start with pure data from sources such as built -in Searorn data sets, and then complete the Messier Real World data.
Part 2: Linear algebra
Each machine learning algorithm you apply is linent algebra. Understanding this transforms these algorithms from mysterious black boxes into tools that you can apply.
Why is it necessary: your data is in matrices. Thus, every operation – filtering, transformation, modeling – uses linear algebra under the hood.
Basic concepts: First focus on vectors and matrices. The vector represents a data point in a multidimensional space. The matrix is a set of vectors or a transformation that transfers data from one space to another. Multiplication of the matrix is not only arithmetic; In this way, algorithms transform and combine information.
Own values and own vectors reveal basic patterns in your data. They are delayed with the analysis of the main ingredients (PCA) and many other dimensional reduction techniques. Not only remember the formulas; Understand that own values show the most essential data directions.
Practical application: Implementation of matrix surgery in Numba before using higher level libraries. Build a uncomplicated linear regression using only matrix surgery. This exercise will strengthen your understanding of how mathematics becomes a working code.
Educational resources:
Try this exercise: Take a super uncomplicated IRIS data set and manually perform PCA using EIGEGENDECOMPOSITION (code using Numpy from scratch). Try to see how mathematics reduces four dimensions to two, while maintaining the most essential information.
Part 3: Account
During training, machine learning model learns optimal parameter values by optimizing. You need an account in action to get optimization. You do not have to solve complicated integrals, but understanding derivatives and gradients is necessary to understand how algorithms improve their performance.
Photo by the author Ideogram
Optimization connection: Every time the model trains, he uses a differential bill to find the best parameters. The descent of the gradient literally follows the derivative to find optimal solutions. Understanding this process helps to diagnose training problems and effectively tune hyperparameters.
Key areas: Focus on partial derivatives and gradients. When you understand that the gradient indicates in the direction of the highest growth, you understand why the descent of the gradient works. You will have to move towards the most steep decline to minimize the loss function.
Don’t try to wrap your head of complicated integration if you have difficulties. In scientific data projects you will work with derivatives and optimization for the most part. The bill you need is more about understanding the speed of changes and finding optimal points.
Resources:
Exercise: Try to cod from scratch descent to get a uncomplicated linear regression model. Utilize a number to calculate gradients and update parameters. See how the algorithm coincides with the optimal solution. Such practical practice builds intuition that no theory can provide.
Part 4: Some advanced topics regarding statistics and optimization
When you feel comfortable with the basics, these areas will facilitate improve your knowledge and introduce you to more sophisticated techniques.
Information theory: entropy and mutual information facilitate to understand the choice of functions and model assessment. These concepts are particularly essential in the case of function models and function engineering.
Optimization theory: In addition to the basic descent of the gradient, understanding the optimization of the convexity helps to choose the right algorithms and understand convergence guarantees. This becomes very useful when working with real problems.
Bayesian statistics: Bayesowski’s exit beyond the repeated statistics opens powerful modeling techniques, especially to support uncertainty and take into account previous knowledge.
Learn these project topics after the project, not in insulation. When you work on the recommendation system, immerse yourself deeper into factoring of the matrix. When building a classifier, study various optimization techniques. This contextual learning is better than an abstract examination.
Part 5: What should your learning strategy be like?
Start with statistics; It is immediately useful and builds confidence. Spend 2-3 weeks convenient with descriptive statistics, probability and basic hypothesis tests using real data sets.
Then go to the linear algebra. The visual nature of linear algebra makes it addictive and you will see immediate applications in the field of reduction of dimensions and basic machine learning models.
Add the bill gradually when you encounter optimization problems in your projects. You don’t have to master the differential account before starting machine learning – learn it as you need.
The most essential advice: code next to each mathematical concept you learn. Mathematics without application is just a theory. Mathematics with direct practical apply becomes intuition. Build petite projects that present each concept: uncomplicated but useful statistical analysis, PCA implementation, visualization of the gradient.
Do not strive for perfection. Goal for knowledge and confidence. You should be able to choose between techniques based on their mathematical assumptions, look at the implementation of the algorithm and understand mathematics behind him and the like.
Wrapping
Learning mathematics can definitely facilitate in development as scientists from data. This transformation does not happen by remembering or academic rigor. This happens through consistent practice, strategic learning and the desire to combine mathematical concepts with real problems.
If you receive one thing from this road map, it is: mathematics needed to learn about data is possible to study, practical and immediate.
Start with statistics this week. The code next to each concept you learn. Build petite projects that show your growing understanding. In six months you are wondering why you thought that the mathematics behind the data is intimidating!
Bala Priya C He is a programmer and technical writer from India. He likes to work at the intersection of mathematics, programming, data science and content creation. Its interest areas and specialist knowledge include Devops, Data Science and Natural Language Processing. He likes to read, write, cod and coffee! He is currently working on learning and sharing his knowledge of programmers, creating tutorials, guides, opinions and many others. Bal also creates a coding resource and tutorial review.