Photo by Author | Canva
A strong portfolio often differences in making and breaking it. But what makes a portfolio stronger? Many complex projects? Clever design? Imagination of impressive data? Yes and no Although these nesses are essential elements of the greatness of the portfolio, those elements are so clear that everyone knows that you cannot work without them.
However, many data scientists make mistakes while trying to go beyond. As a result, he is interviewing with Portfolios that has everything that is nominated but is not really so big.
. Framework
This is the framework that will help you avoid normal mistakes when building a great portfolio.
. Errors
Now let’s talk about portfolio mistakes and how to avoid using this framework.
!! Error #1: Of building plans you don’t care
Many departments have given the impression that projects are only to mark a box: Titanic Survival, Irris Dataset, MNIST digits. You know – ordinary things. Not only is it that you will be submerged in thousands of similar departments, it also reflects the lack of origin and interest in your work. Auto pilot projects.
Correct: Start with domains that you are interested in, such as, GameFor, for, for,. FinancialFor, for, for,. Music. When the subject is interested in you, you will be deep without trying. If you are a sports fan, you can analyze the shot performance in the NBA or choose from these cold project ideas for practice. A music fan can create a model of playlist recommendations.
!! Error #2: To use whatever data comes in your lap
Candidates often catch the first clean CSV that they can find. The problem is that real data science does not work like this.
Fine: You should show you know how to find the original data, access it, and to re -establish it with more modeling stages. In your plans, use APIS (such as, Twitter/X API), Open Official Datases (such as, Data.gov), And web scrapped sources (such as, Great public datases on Gut Hub) Use more data sources, evaluate the data, integrate them in a datastate, and develop it for modeling.
!! Error #3: Treatment of projects such as Kagal competition
Cogl The competition focuses on the only matriculation correction. It’s great for exercise but does not cut it into the real world. Accuracy in itself is not a purpose. You have to trade between the technical aspects of your model and the actual business or social impact.
Fix: Even if you use ordinary datases from Kegal, always offer a different angle and frame the problem so that it has business or social value. For example, do not just rate the real news. Which words, phrases, or titles run misinformation. Another example: Don’t just predict the mantra.
Show how a 10 % reduction in rotating can save $ 2 million in annual income.
!! Error
Many projects read as a continuation of the Paper Notebook: Importing libraries, then data pre -processing, then fitting model – here is accuracy. It is incomplete and boring. What is lost is how you handle the different stages of a project and why you make some decisions.
Fix: Make them projects from the end to the end. Show everything from data collection and everything between it. Tell us why you made key choices, such as why you chose one model more than the other, or why you engineered a special feature. Use tolls like StreamlitFor, for, for,. FlaskOr Power bi Dashboards to use others. All this will look like your plans implemented the problem (such as, Portfolio of Arch Desai), Code walkthrough not (such as, This one,
!! Error #5: Expired with a model, not process
Data scientists are often eliminated at the technical level, such as the accuracy score. OK, but what do you do with it? You should remember that what is important is the practical use of the model. The technical aspect of the model is just one part of it, the other business or the social impact.
Fix: Eliminate the project with the recommendation of what to do. For example, “This model recommends inspection in restaurants that serve high -risk food during the winter.”
. Project Example: City City Predicts City Energy Demand Predicts
In this section, I will create a fun project walkthrough to tell you how the framework can be used practically.
Domain: The domain I chose is energy consumption and stability. Living in a large city, I was aware of how cities are struggling with high demand for electricity during high hours. Demand for forecast can be more accurately helping to balance the grid, reduce costs and reduce emissions.
Data: can be a central source US Energy Information Administration (EIA). In addition, I could use Noaa weather api (Like, temperature and humidity)), and the calendars of the holiday/event (increased demand)).
Setting up the problem: Instead of preparing this problem as a “prediction of electricity demand over time, I would frame it,” How much money can the city save if it transmits the burden using better demand predictions? ” With this, I turn a technical prediction into a problem of distribution and cost savings.
Building from the end to the end: The project will include these steps.
- Data Cleaning: handle the missing times, align the time stamp, bring the weather variable to normal.
- Feature Engineering:
- Interval Features: Demand in previous hours/days
- Weather features: temperature, humidity
- Calendar Features: Saturday, Holiday Flag, Important Events
- Modeling:
- Deployment: For example, I can create a dashboard that contains a 24 -hour forecast vs. the original demand and adjusting the demand for “if what happens”, such as industrial burden.
Action: We will not keep on “low RMSE in forecast”. Instead, let’s give a recommendation that has a business and social impact, such as, “If the city encourages big businesses to remove 5 percent of consumption from the peak (predicted by the model), it can save grid costs annually M 3.5 million.”
. Bonus: Resources
As a bonus, here are some tips on which platform you can use for practice and where to find the data.
!! Platform to practice
!! Open data sources
!! APIS for real -time data
. Conclusion
You may have seen that the above mistakes are not technical. This is not accidental. The biggest mistake is to forget that a portfolio is demonstrating how you solve the problems.
Focus on these two aspects-demonstrations and solving the problem-and your portfolio will eventually begin to look like the evidence you can work.
Net Razii A data is in a scientist and product strategy. He is also an affiliated professor of Teaching Analytics, and is the founder of Stratskrich, a platform that helps data scientists prepare for his interview with the real questions of high companies. The net carrier writes on the latest trends in the market, gives interview advice, sharing data science projects, and everything covers SQL.