You're stuck on a complex Machine Learning problem. How do you know when to ask for help?

Before seeking assistance, it's crucial to evaluate your situation by pinpointing the source and type of problem you're facing. Determine if it's a conceptual, technical, or data-related issue, assess your understanding of the problem and potential solutions, consider the available time and resources, and review your past efforts and their outcomes. This assessment helps in narrowing down the problem and establishing clear goals for resolving it.

Situation is quite broad. You should be in one of the following scenarios: -you cannot translate the idea/concept into a code. In this case either the idea/concept is too vague to be implemented or some details are missing (either you missed or your boss/colleague missed). You should clarify to yourself what it is missing and say something like 'to implement X we need Y,Z etc.' -code to use is complex and you are lost in it. In this case, assuming the code is huge with many libraries (known to you) you have few options: 1. ALWAYS try 'stupid, trivial' tests first. Never underestimate naive mistakes. 2. understand which libraries/classes are called. Exclude the irrelevant. You may ask someone with experience in the code if you have done 2.

When stuck with complex machine learning problem, always try: 1. Simpler machine learning model or approach 2. Drill down if problem is with model architecture, data, underfitting or overfitting , not performing for specific kind of data, model latency Once tried, then only check with peers

When facing a problem, consider seeking relevant resources that can assist you in finding a solution. Numerous online and offline materials are available for learning new concepts and skills in machine learning, including books, courses, blogs, podcasts, videos, tutorials, forums, and documentation. These resources can help you gain knowledge, discover fresh ideas, and draw inspiration from others' solutions. However, it's important to avoid information overload and concentrate on the most pertinent and dependable sources for your specific problem.

The 1 day rule can help with balancing self research and amount of time spent - if you can't figure out a solution by yourself in one day, ask for help!

Stack overflow is a nice resource for finding solutions for your problem. And talking to your peers with similar interests is going to help a lot and there is a high chance of unlocking many other possibilities and perspectives when you indulge in such discussions. When finding the right person even if your peers are not that skilled, there is no harm in discussing your problems with them. Explain and make them understand the problem and you will be amazed to see the ideas that come out of these kind of discussions.

Stackoverflow is good for pesky errors in your code. However, for more framework (also modality like image and text) specific help I prefer pytorch and huggingface forums. However, depending on the complexity and confidentiality of the problem, it might be better to ask help from your colleagues or mentors.

When seeking help, it's important to prepare your question effectively. This involves creating a clear and concise question that includes elements such as a descriptive title, a summary of the problem, a statement of your goal, details of your previous efforts and what didn't work, relevant code or data samples (if applicable), and a specific request for assistance. Proper formatting, spelling, and grammar should be used, while avoiding vague, broad, or subjective queries. Preparing your question in this way enhances the likelihood of receiving a helpful and prompt response

Showing gratitude and following up with your results with the people or community who helped you is not only going to foster the positivity but also is going to help you immensely in the long run. Keeping them in the loop will make it easier for you to discuss your future problems or deadlocks with them because they already know what you are doing and you will get the solutions much faster.

Getting stuck with a problem and not able to solve by it yourself doesn't make you dumb. Everyone goes through this phase and even the most experienced people get stuck at times and it's completely normal and ok. So don't hesitate to reach out for help. Nothing to be ashamed of.

Before reaching out, articulate your problem clearly - what you've tried, what you don't understand, and what you're hoping to achieve. This makes it easier for others to assist you effectively. Remember, asking for help is a sign of strength and growth. Don't be afraid to tap into the collective knowledge of the ML community - it's there to empower you, not judge you. So, raise your hand, share your struggle, and watch your problem magically transform into an opportunity for learning and collaboration.

It is impossible to know what you do not know. Knowledge is gained via experience and there will always be someone who has walked the path before you. Reaching out and helping each other is how a thriving community will be established.

Even though one cannot be prepared for absolutely everything, having a solid background in statistics, system dynamics and the machine learning techniques you are using (ex: SGD, kernel tricks, etc.) can help you a lot in formulating solid, specific and dense questions that will help whoever is assisting you in answering them also with more solidity, specificity and depth. Knowing your trade and never stopping to learn is a very good way for both knowing when you are really stuck and efficiently asking for help.