I find my masters topic extremely hard what should I do?

Question

I'm writing this because I'm in a real trouble and I need to hear from experienced people.

I'm new here so please excuse me if I get anything wrong. I'm doing my masters in Deep Learning and I'm having toughest time ever.

I have a research paper very similar to what I wish to make, with the codes in Python which is working fine and is giving results. The problem is I can't understand that code it's beyond my knowledge.

I read books looking at codes but couldn't achieve that level of being able to understand the code or the paper.

The paper is about Recurrent Nueral Networks and motion prediction.

Unfortunately I'm not good at Python or Deep Learning and feel like I'm really stuck.

My supervisor didn't know much about Deep Learning but he approved my proposal. Now I don't know what to do. Should I stick to the topic and take courses in Deep Learning like the one in Coursera even if It means to take extensions beyond the deadline?

Or should I change the topic completely?

I asked my supervisor and he was against changing the topic he suggested to simplify the topic and choose something simpler than Deep Learning like interpolation. He said that submitting a proposal will take time and it's not guaranteed that your new topic will be easier than your current topic.

I tried to find some codes about interpolation motion data to begin with but can't find anything related to my topic.

I'm not lazy at all, I love researching and I have hundreds of codes in my machine but still can't understand the code I have about RNNs.

I contacted the authors of that paper but they weren't cooperative.

I feel like I ran out of energy and very confused.

If you were in the same situation what would you do?

Answer 1

It's a bit hard to bring this all together to give good advice but let me make a few suggestions.

With regard to the Python code you are looking at, it is possible that it is just terrible code that no one could understand easily. That may not be the case also, but it happens. On the other hand, if it is a sensible program, then it is probably based on some underlying model that doesn't itself depend on the code. If that is the case, you could find that model and understand it first, then attack the code. The paper you have found probably has some bibliographic references that may help you understand everything.

Furthermore, you could try to find someone locally who is a Python guru and have them look at the code and give you advice about it - perhaps with the model in hand.

But your advisor seems to be giving good advice. He or she is the one who will need to approve your result in the end, so it is good to take the advice offered. If doing a bit less is acceptable to them, then it is acceptable, and probably easier. Doing a simpler project should also give you more experience and better enable you to take on harder work later. As a Master's level student it may be a mistake to go beyond the understanding of your professor. Doctoral students normally need to do that, but they have a lot more experience by then.

Running out of energy (burn out) is a serious academic issue. One way to avoid it is to make a plan, in conjunction with your advisor, that you can execute properly. Once you have a plan it won't feel like you are just thrashing around, as long as you follow the plan.

Answer 2

I should say that mistakes seem to have been made, and not (only) by you: putting a student with (as you say) limited programming skills and no experience in deep learning on a topic as notoriously hard to get into as deep learning when the advisor also has little understanding of the topic seems to me like a recipe for desaster. That said, it happens to all of us occassionally.

Your advisor's proposal to move forward with a simpler algorithm seems smart - this is also what I generally do if the original plan turns out not to be feasible. That said, your general approach to "find codes" that implement your algorithm for your use case would worry me, to be honest. Extrapolation is a fairly simple principle - simple enough that if I told a student to try extrapolation for a given problem, I would expect that they can come up with a solution themselves in fairly finite time (that is not to say that you shouldn't use an existing solution if one is already out there, but in the absence of an existing library just writing it yourself should totally be doable).

If that is outside your reach, actually changing topics may really be the most sensible thing to do for you, but I suggest having this conversation with your advisor.

Answer 3

Your advisor suggestion is wise. simplify the technical basis for your thesis while you try to improve your python programming skills.

you could be able to get some support from other python programmers in your university and online. people would be able to hep more if you have specific programming question related to one problem at a time. so try to break your project into smaller chunks of problem and then solve it one by one.

Also There is many websites that run python courses like data camp.com and, as you mentioned, coursera.

Answer 4

Remember that compilers aren't smart. If they can understand the code, you can too.

What to do now:

• Relax. Code written by someone else is not easily understood. The last time I tried, it took weeks before the code started making sense, and even longer before I was able to use it confidently.
• Get a high-level understanding of the code. You can do this off the Coursera course for instance. Knowing roughly what each part of the code is trying to do helps a lot with understanding it.
• Test the code with different parts changed. For example, if you find a piece of code that doesn't seem to do anything, try commenting it out and rerunning the code.
• If you still can't understand it, phrase your questions in a way someone else can understand and ask at Stack Overflow.