Deep Learning Prerequisites: Logistic Regression in Python (Udemy.com)

I'm giving this a 5/5 because by the first three sections, I've already achieved my goal of at least typing a run through from import to finding the weights, without looking at his code, before running. I'm still far away from fully absorbing the concepts that the course offers, but for anyone looking for a Udemy course to explain a deeper view into Logistic Regression in a few days' sitting (That is balanced enough that we don't study things like creating a for loop in Python like some introductory courses do), I'd say that this course delivers what you're looking for.

Instructor has been very responsive to any question I posted in Q&A. The course covers the actual maths behind all the code examples which is great and is what I was looking exactly looking for. I was able to understand what L1/L2 is in real vs fancy regularisation diagram showing some kind of smoothness happening. The contents are covers a lot of facts and requires you to think and analyse, which is good. On some forums have seen some negative reviews about lazy programmer which does not make any sense me to. After taken multiple courses from the instructor , I can say every fact covered in the lectures amongst the courses I have taken, have been very valuable.

The author interprets the classification a categorical random variable with two outputs defined on a sample set X contained in a vector space as finding a separating hyperplane. He also relates this problem to the biological neuron. To achieve this classification, a probability P is defined on X, and the predictions are the values of P rounded to either 0 or 1. The formula of P uses the sigmoid function. This is motivated by the fact that a similar expression appears when some components of X are normally distributed. The author defines the corresponding error function and computes it's derivative. This enables to implement the theory and it's L1 and L2 regularizations versions with gradient descent using Python. Remark that even when the variables cannot be separated by an hyperplane (nonlinear problems) , a wise embedding may lead to the same solution as shows the author for the donut problem. He also applies this method to an Ecommerce project, and a facial expression recognition project. In this course, all the theory underlying the concepts is presented in details and mathematical formulas are demonstrated. A basic knowledge of Python in particular control flow, arrays, randomization functions of it's numpy library is required to understand codes included. I recommend this course to anyone interested to understand the foundations (mathematical) of logistic regression and to apply it using Python.

Following the linear regression course, this course dives into slightly more complicated material. Similar to the previous course, the instructor provides clear and concise explanations to the theory (math) and demonstrates how to implement these in code. I've now taken 6 courses from LazyProgrammer and noticed that in the earlier courses, students ask for more projects to practice implementation. My recommendation is not to get hung up on feeling like you need to understand implementation 100%. If you continue to progress through the series, you'll find many of the questions answered at the appropriate time. That is one of the helpful aspects of the course layouts. The material is covered in a way that builds upon the prior courses. The next two courses "Supervised learning in Python" and "Deep Learning in Python" will use many of these topics again in more complex situations.

The course goes into the ground of Logistic Regression. I felt sometimes as not everything is exhaustively explained. This on the other hand makes you think more. The lecturer really knows his stuff and answers very fast in a manner that makes you think - I really like that. Probably the width of the course is smaller than others on the machine learning topic, but because of the depth here I think was good. More practicals would be great though. Overall recommendation and thanks

It is great that there is a series of more in-depth classes about ML/DL. However, this class is way too fast and expects a GREAT amount of knowledge in calculus, linear algebra and statistics. Even though the prerequisites are background knowledge in these fields, there is a great difference between having this knowledge or doing these things on a daily basis. Thus, with some more explanations about how you get to the next step and what you are doing exactly people would understand much better.... Remember that learning has a lot to do with repetition and not with saying it once (like in another course) and expecting everybody to remember every detail...

As with the Linear Regression course, I found the course to be very detailed, in-depth, and enlightening. The instructor has a really great way of explaining and motivating the concepts that makes everything very clear. This course is also more in-depth than other courses out there on udemy (or other sites) on the same subject. Really great! If I could make a request, I'd like to see a section that deals with model performance metrics, like accuracy, precision, recall, and AUROC. I'd love to see Lazy Programmer's treatment of these topics.

[Thanks & Suggestions from Statistics Instructor (Alex University)] LP is very good teacher, he depend on giving 90% of the knowlegde and 10% to be investigated by you (Thats a good appouch for learning), He is explaining the Logistic regression as a Classifier to be used in Prediction [ Machine Learning main objective]. However, in epidimic research and in statistical analyisis , one may be looking for Test the significance of coeficient, understand the meaning behind reg coeficient, getting Risk Ratio, Checking the Logistic Assumtption, Calculate the Goodness of fit of the model, ..... all this aspect is not covered in the Course, and will be very nice to be added, so the course fit more categories of studients.

Now that I have completed the course I can give you a fair evaluation. The only confusion with me are the two words Training and Learning. They both are technically forms of learning(Cognitive Functions). It is not until you ask me to do something that I was trained to do is when you realize that I have learned it. The end of the course does emphasize why doing the problems yourself first and then seeing where you got stuck is an excellent way of learning, rather than getting frustrated not being able to solve the problem and giving up. All in all, the Logistic Regression course is well worth the effort to go through to understand how the classification model works.

Brilliant as always. LazyProgrammer never ceases to amaze me. Not only does this course give you the math behind logistic regression, it gives you the math from two perspectives... The geometrical as well as the bayesian perspective. I find these courses the best refreshers on the week of a machine learning job interview. Also note that the course has both, the theory and the code, for logistic regression. So there's never a feeling of being shortchanged on either aspect. More importantly, it does not have ready-to-use scikit-learn library usage that hides all ML detailed from a student .