Experience in Data Science specialization on Coursera in 6 months

Greetings

In this post, I would like to record my fading memories of the acquisition of “Data Science, a 9-course specialization by Johns Hopkins University on Coursera. Specialization Certificate ”, focusing on the organizational side of the issue. Those. I will not talk about how prestigious this certificate is, what teacher is more boring and whether enough knowledge is given. In my opinion - these are subjective questions. Instead of why, I will talk about how: in what order and how to take courses, what to look for when taking tests and coursework, and what happens as a result.

It's not free

This specialization (everywhere in different ways) consists of 9 online courses + capstone project (graduation project). For each course you receive a certificate, and it must be verified. You can listen to the course for free and get a regular certificate, but this one will not qualify for your signature track. Most likely, the signature track (the list of matching specialization courses) will be able to pick up the course already completed for a fee, which is part of your chosen specialization. Do not listen to him again, right? But I did not have such.

Total, the cost of Data Science specialization will consist of the price of 9 verified courses + capstone project. And the prices in your currency. You can pay for each course as needed (the price will be slightly higher), but I caught the moment when the ruble exchange rate had already fallen, and the prices had not yet been adjusted - so I paid immediately for all 10800r. Yes, and I am more motivated to pay in advance. Now a full set of courses costs 18600r (or 28200r, if you pay one at a time). I advise you to take a beam.
')
I didn’t offer to buy any books or services (paid Azure, etc.) in the process of obtaining this specialization.

From pleasant: after payment you can proceed to the course within 2 years. If you fail - the subsequent attempts to retake it are free. This means that if you paid for the course, you will not be able to receive a certificate for it if you stop trying to do it. So getting specialization becomes a matter of time and desire.

What is the verification

1. You will be asked to type a long phrase and take a picture on the webcam. Honestly, I don’t remember exactly, but it may be necessary to send a photo of an identity document with your photograph once. This is also done via web cam. For shooting using flash plugin with frames. No pattern recognition for framing occurs.
2. By sending test results or links to reports, you will enter this phrase each time (apparently printing style is checked) and be photographed on a web-cam. This is part of the submission process. Photos, apparently, then briefly viewed by specially trained monkeys. And you have to do it at least 50 times.

Conclusion: you can watch the video anywhere, but you need to send the results on the machine with the hardware keyboard and web-cam. What will they remind you of all half a year. If the photo does not work out, you can make a submission once per course without it (I could be wrong in details).

Linux

Despite the fact that the courses of specialization were tested by the authors on Win and Mac, I passed it on a laptop with Kubuntu 14-15 + Firefox. All software used in the course is for this system. I didn’t have any need to start Windows to perform any operation. But this, of course, does not prove that any course on coursera can be successfully completed with linux machines.

Course order

There are 2 rules in Data Science specialization:

1. proceed to courses from 2 to 9, you can only passing the 1st.
2. capstone project can be started only by passing the previous 9 courses.

Each course lasts a month (or rather, 4 weeks), a capstone project lasts 2 months. After the course is restarted. Usually the course starts from the 1st to the 5th day - on Monday / Tuesday.

Hence the conclusion: if you pay for the specialization on the 10th, you will still wait until the end of the month to start learning. In addition, when finishing current courses, be sure to watch for when the registration for the next one opens, and register for them. Otherwise, you risk getting 4 weeks of vacation.

As for the order of passage, I’ll say that I experienced developer, read a book on R by that time and had the opportunity to study full time. So I took several courses at a time. Nevertheless (about what below), I believe that such a scheme will be drawn by a student who studies after work / at the weekend.

1. As soon as possible, take the course number 1 and pass it for 4 hours. Well, a maximum of 2 days. Yes, he is so simple + there are some tests. It is not worth a month.
2. Immediately, in order not to lose a month, you will have time to take courses №2 and №3. (i.e. 3 courses for 1st month).
3. Then take two courses in parallel and, as a result, type 9 completed courses. (3 more months).
4. Make capstone (+2 months).

How the course is arranged

During the course you may need to pass 1-4 tests and 1-2 coursework. Immediately see Syllabus (curriculum) - it says that you are required and the main thing - deadlines. Usually, the course looks like this:

1 week: video lessons + test with deadline at the end of the week
Week 2: video lessons + test and / or coursework with deadline at the end of the week
Week 3: video lessons + test and / or coursework with deadline at the end of the week
Week 4: video lessons + test

Please note that on the last week there cannot be a coursework (why - below). And on the first he never came across to me.

Video - hypothetically, you can not watch at all. So that the student does not forget everything, the player can sometimes pause and give a question for repetition. These answers, from the point of view of certification, absolutely do not affect anything. If you download a video on a PC, then this pleasure, of course, will not be there.

Tests in this specialization do not have a timer. (Some of the questions will require working in RStudio, so passing the test can take a long time.) They usually consist of 6-10 questions with 3-5 answers. It is given 3, at least 2 (in capstone) attempts to surrender it. In the offset is the maximum score for an attempt to score. Do not be afraid to close the question page - the attempt will not be counted until you pass the submission process. Questions do not change, the list of answers can be diluted with fresh wrong options and shuffled.

Usually, the test is passed from the first time. If difficult - you should immediately get into 7-8 questions out of 10, the rest will be able to twist.

I have not met incorrectly posed questions or questions that are clearly not having the correct answer among those proposed. If you think so - most likely you misunderstood something.

Tests will give 30-40% of course credit score. (To obtain a certificate you need> 70%)

Course

This can be a report or a 5-page presentation on RPubs / Github, the link to which you send. Or a project on Github. Or a web application on shinyapps.io. Rarely, instead of a link, you will be asked to attach an R file.

Usually course one. But once, if I am not mistaken, there were 2 small ones.

The course has 2 deadlines - the second a week after the first. Therefore, the course does not happen on week 4. By the first deadline you hand over the project. To the second - check 4 strangers, putting them points in a small form. Sometimes put a mark and yourself. The grade for your course will be the average of your classmates' grades. I suspect that course supervisors do not look at these results until they are out of the trend. Those. in 99% of cases your fellow students rate you.

From here the terrible headache follows. The task (by the way, carefully read all the requirements for the task) is usually set vaguely, and can be performed in various ways with varying degrees of zeal. And you will always have a dilemma - how many reasonable efforts to attach to its implementation. The problem is that you will be judged "on their own." And if you have done less or not like your colleague, he will probably slow down your assessment. In addition, requirements may be something like a “presentation for your boss / non-technician”. How much water should be in such a presentation, and how much technology is always a subjective opinion.

Therefore, 2 things are important - carefully read the items in the questionnaire by which you / you will be assessed. I personally, did not reduce the assessment in cases where I do not have a reason for this. Those. the program may work crookedly, but if the form shows {start / presence of user input / successful prediction of at least one result out of 5} and specifically there is this in the program, then I have no formal reason to slow down the estimate in any of these items. You can express your general phi in the field for a detailed commentary, as well as later read comments about your work. But they do not affect the assessment. In general, you do not know how - do it so that it is impossible to find fault with the evaluation questionnaire. In general, I think fellow students tend to overestimate grades.

The second thing is that there are places that are interpreted in two ways / three ways, etc. And this can lead to diametrically opposite results. If you have doubts about something, check out the course forum. I'm sure there will already be a holivar. If not, feel free to write your result and ask the supervisor: “is it possible?”. You will most likely be answered that it is possible both this way and that, and in general - how many people, so many opinions. A link to this discussion marked “I was so allowed” is attached to your work. In this way, you will avoid problems if your opinion is shared by a minority, not by a majority, and save the tester from doubts.

The exchange rate will give 60-70% of the final assessment. In general, the percentage is chosen so that it is impossible to pass a course without it.

Separately, I note that after each course a lot of reports, github projects, and so on. All of this is public, and can be naguleno next stream. Therefore, such materials need to wipe for a. Although cases of plagiarism did not come across to me. I advise you to do this slightly after receiving the certificate, and not immediately after the end of the course - you never know.

Penalties (penalty)

For the delay of the test or coursework shall be a penalty in% of the score for them. Do not forget at the end of the deadline for the delivery of the course return and check out 4 other people's work. Otherwise, get paid for it. And I got it, because passed the course 2 weeks before the first deadline and forgot about it completely.
There are exception tests for tests - forget about them, respect the deadline.

Certificate with distinction

If you type more than 90% of the maximum course credit score, the certificate issued for the course will be marked “with distinction”.
But for capstone this is not. Those. you will have 9 with distinction, and someone has it all without a difference - you will receive the same final certificate for the entire specialization. By cr. least visually the same. If you follow the link to such a certificate, then 9 included in its program will be visible.

By the way, certificates appear 2-3 days after the end of the course. Although, once the delay was 1.5 weeks, and the public began to go crazy.

Swirl

For some courses, the authors allow you to perform additional tasks and get points for them. All tasks are to pass a small interactive training directly in RStudio through the Swirl package. I will not explain the mechanics - nothing complicated there. Swirl lessons are not available for every specialization course. Usually there are a dozen of such lessons, but passing through any 5 of them has practical meaning. This is because for each given +1 point, and more than an additional 5 points for the course to collect in this way do not give. If you have not received the maximum score for the test, or are not sure that the reviewers will appreciate your project - make yourself a handicap - go through at least 5 first lessons. It will hardly save you from failure, but it can tip the scales in the direction of a certificate with a distinction. I, for example, thus partially closed the penalty obtained by forgetfulness.

And once again on the order of the course

After I grinded the first course in 4 hours, and even a week later I passed everything by the next 2 (of course, it was necessary to return to them and evaluate other people's work, but it takes no more than half an hour for 1 foreign project, depending on your consciousness) I have 2 free weeks left before the next one. (Do not profukayte opening registration on them). I spent these 2 weeks competing on kaggle and contributing some code to VW. Then the question arose: continue at this pace (2 courses in 1.5-3 weeks + self-education / kaggle) or take 3 courses in parallel. So I advise you first. And for those who study in their free time, probably there will be no such question.

The problem is that all the courses you take at once start in the same first week of the month. And with high probability deadlines for all coursework will fall on the same date (2-3 weeks). Well, with the progress on the signature track, the courses become harder, and coursework - more creative. Plus, each subsequent course may slightly rely on the knowledge gained from the previous course. In general, at some point you will not be able to effectively switch between them, and the delivery of coursework will begin to cling to deadlines (and all deadlines are in one day). In general, despite the fact that when studying on two projects at once, the student has enough time (at least at the beginning of the journey) enough time to take three or more courses at a time is not an option, because The peak load in the 2nd-3rd week will knock you out. Now, if the beginning of the courses could be shifted for a week or two, then yes ...

Capstone project

Minimum video - the information you need to collect yourself. Two coursework - a report with exploratory analysis and the actual web app on shinyaps.io.



Note that participation in capstone requires completion of 9 courses. So you cannot have anything parallel with capstone, and there should be enough time to study a new subject area.

It is not known how often the capstone project assignments change. Most likely, it has never changed - in Google there are reports on the same topic left over from students of previous launches of the capstone project. By the way, do not forget to delete your own.

Result

The result of the training is a record in the coursera certificates database and a URL to admire this record. And the button automatically assigns it to you in linkedin profile.

Article compiled from memories and in small details, especially over time, may differ from your experience.

If you forgot something important - I will add.