Trying to get into Data science
I have 3 years of experience in other domain and i am trying to get into Data science. Any suggestions amid these layoffs ?
SleepyHeadF
Stealth
a year ago
We are resume oriented country especially in data science. Not degree. I have recruited and I don't care.
So best suggestion is pick some real world projects(if possible pick one from your current company itself, your best bet) and implement it, then put it on your resume and apply. Just 1 project won't do. You need to show you have done this again n again so that we don't have to handhold you every single time.
steppenwolf
Stealth
a year ago
Thanks. Do you work in Data science ?
SleepyHeadF
Stealth
a year ago
2 is good.
And ofcourse that's just to pass the resume screening. You don't have to learn every algorithm there is, but basics are important, and most importantly, be thorough about the parts used in your projects.
In interview I focus on these parts.
1. Data Processing and engineering (atleast 60% weightage). Most important.
-Be good with pandas. Make sure you learn time series data handling, be as good as possible with pandas.
-if the profile is for NLP, then be clear on how to do text cleaning, data tagging, (like create BIOS tags for NER), fuzzy text matching, a little bit of regex etc... Much more I can't recollect all.
- If you are using PySpark then pyspark or atleast be very very good with SQL. Focus on window functions, partitioning & ranking, merge on nearest date, and several other ops on date etc...
No need to know all 3 especially when you are beginning
2. Problem solving:(25% weightage)
I will ask you to tell me how you would go about building a machine learning solution. It can be something like build query correction for bhojpuri or kannada language where ready made datasets are not available.
You would have to focus on how you would get your data (annotate is not the answer). Think of some ways you can create synthetic data.
How do you frame your machine learning problem. What's your input what's your Target and how you will train
What loss metrics and business metrics you will focus on. You have to sell your work to stakeholders, make them understand the metric clearly and defend why reaching x% is good enough for business
3. Depth in algorithms and math.(15% weight)
- whatever you implement, learn in and outs of the algorithm. If it's xgboost, then first learn how trees work and then GBM. I might ask you to give pseudo code too.
If it's NLP & transformers, I show a custom bert based architecture pytorch code to candidates and ask them to explain what the architecture is doing and what could be the objective
- Loss functions. (Contd...)
See more comments
We are a degree obsessed country... Get a part time Mtech and push all your projects to your GitHub profile
steppenwolf
Stealth
a year ago
Thanks.
Discover More
Curated from across
