** ATTENTION**: PLEASE, do not act on the challenge before sending us an email with your Resume. We will talk over the task during our fist talk, when we plan on explaining better what we expect from it, and what we are expecting from your work.
Note that the data needed to complete the challenge will only be released AFTER our first talk.
Read carefully through the complete description bellow, before sending us an email.
Here at Mediar, we are obsessed with data. So, we are looking for people that can extract useful insights from it.
We collect location data from brick and mortar retail stores, and, by referencing these and sales data, we generate insights and analytics for our partners and clients.
As a Data Scientist, your role is to generate metrics, insights and reports for manufacturers and retail partners, using behavior and sales data, in a quick and scalable manner.
Right now, we are looking for a Junior Data Scientist to join our team. Our Data Science stack is composed by Python components, such as Pandas, Sklearn and Numpy. Most of our work is accomplished on Jupyter Notebooks, but you are free to use the tool that best suits your needs.
In this position, you will be part of Mediar's core product team, projecting experiments and building models using machine learning and data science techniques, on location and sales data on some of the largest retail stores in the world. You will also be responsible for deploying improvements on our current code base, improving on our algorithms and methodologies.
We are looking for professionals willing to work on Belo Horizonte, Brazil
If you fit most (but not necessarily all) of these, let us know!
- BSc on a highly technical degree, such as (but not limited to) Computer Science, Computer Engineering, Engineering, Math or Statistics
- Experience with Python programming
- Some experience (or knowledge) on PyData stack (Pandas, SkLearn, Numpy, etc)
- Knowledge in Machine Learning algorithms
- Database modelling experience (SQL)
- Experience with Front-end development using HTML, CSS and JavaScript
- Strong knowledge of Data visualization tools or JavaScript library for manipulating documents based on data (D3.js)
- English fluency
It would be great if you have any of these.
- Experience with statistical modeling
- Knowledge in client-side frameworks and libraries such as React and AngularJs
- REST API (Python Flask) - Complete understanding of the REST architectural style
- Experience with Google Cloud platform core infrastructure (App Engine, Storage, BigQuery e Cloud Functions)
If you wish to apply for this position, let us know by sending an email with your resume to carlos[at]mediarsolutions.com, so we can talk. Our process consists of the following steps:
-
Resume Analysis: After sending us your resume, we will try to get back to you ASAP.
-
First Talk: You come to our office (or Skype us) and we talk for 30 minutes to one hour. We explain you more about the position and what we expect from you.
-
We release the data needed for the challenge: Right after the first talk, if everything goes well, we send you the data needed for completing the challenge bellow.
-
Challenge presentation: You come back to our office, where we will talk about the challenge, and you will have the chance to show us what you did.
-
Final Interviews: We will call you back for one or two extra interviews, with the product team and the management.
-
Proposal: If we choose to move forward, we will send you a proposal for the position.
For this specific position, we are planning on completing the entire process in up to two weeks.
Mediar Solutions (formerly IDXP Analytics) is a brazilian company with it's HQ in California. It was awarded as "Global Entrepreneur of the year", by IBM and "Editor's Choice" in the world biggest Shopper Marketing fair. We were cited in some of the largest media outlets, like Forbes, Wall Street Journal, Venture Beat, Revista Exame, Folha de São Paulo and many others. The company developed a Big Data and Analytics platform that monitors in-store shopper beheavior in real time for retailers. The system helps retailers and manufacturers increases its sales and measure the efficacy of promotions and other actions across the store. The company has already raised multiple funding rounds in Brazil and abroad.
Some perks that you will find here:
- Relaxed environment
- Competitive pay
- Stock Options
Here at Mediar, we use the data described below in order to generate metrics and insights to costumers and partners. In this challenge, you will do the same.
We provide you one month of real, in-store data, located in the data/ folder that will be sent to you after our first talk. The provided files are the following, all as described below.
sales/: Folder with sales datalocation/: Folder with location datamapping.csv: CSV with mapping between category and store areastore.png: A high-res blueprint of the store, with some of the areas of interest shown in red (The image is not necesseraly from the same store).
Every analysis that we perform on the data relies on, at least, the following three metrics, calculated for each category and for each area of the store:
-
Exposure : How many people were exposed to that category or area? Meaning, how many people were identified on the area of that item?
-
Engagement : What is the average engagement power of a given category or area? Meaning, how long does someone, on average, spend on a given area? It's defined as the average time spent by shoppers spent on each area.
-
Conversion : On average, how many people need to be exposed to an item so one purchase will be made? This can be defined as #_baskets_sold/exposure (We define one basket as one shopper that purchased that item)
Every other basic metric and analysis that we use are crafted by combining these three base metrics and other variants on that.
What we ask you to do is the following:
-
First, we ask you to Calculate the three metrics above on the sample data given. The first two must be calculated for each each category and each area. The conversion, for each category.
-
If you find any inconsistence with the data, we want you to try to come up with plausible explanations as why that happened.
-
Since the role of a Data Scientist also involves the creation of new and interesting metrics and insights, we also want you to suggest (and calculate, of course) other metrics and insights from this data. We want to see how well you can use your imagination and knowledge in order to find patterns and insights on the data. We are looking forward to see what else can we do with the data provided. Please, surprise us!
Note that the data, both location and sales, is real. Therefore, any inconsistence should be considered and will probably come up on your daily work at Mediar (and any other data company). Therefore, it's up to you how to handle these. Some inconsistences that may occour are (but not limited to): 0-seconds exposure, more baskets on a area than exposure, missing data and others.
We want to see how you deal with these inconsistences, and it would be great if you could come up with your own explanations on why these issues may happen.
Our location data is based on video information collected at the store. Our system can understand when someone enters any designated area, and for how long that person remained on that area, as such:
| AREA | TIMESTAMP | DURATION |
|---|---|---|
| 104 | 1464752382 | 6 |
| 104 | 1464752042 | 6 |
| 104 | 1464753356 | 0 |
| 104 | 1464752953 | 13 |
| 108 | 1464752797 | 6 |
| 108 | 1464757765 | 6 |
| 103 | 1464775340 | 2 |
| 104 | 1464777447 | 13 |
| 100 | 1464787092 | 6 |
The data has the following columns:
AREA: An unique ID of a location in the storeTIMESTAMP: The moment that someone entered thatAREADURATION: How many seconds that person remained on thatAREA
Each row on the data can be interpreted as a single event (Say, one shopper captured on that given area, starting at the given tiemstamp and ending after the number of seconds indicated.
Below, we show the blueprint of one store, with the areas that we are interested shown in red. The green and blue squares are cameras that are used to track users in store. Only the blue ones were used on this data.
from IPython.display import Image
Image(filename='data/store.png') 
The sales data is more complete, and has the following columns:
external_id: An unique ID representing the complete basket. You can understand this information as the ID of this shopper.date: Date and hour that the given item was scanned on the Point of Sale.sku: An unique product identifer, unique to this retailer.upc: An unique product identifier, unique to the product (AKA the barcode of the product)product_name: The description of the product, given by the retailermanufacturer_id: the ID of the manufacturer, given by the retailercategory_[1-5]_id: IDs of the category and every subdivision of the category, given by the retailercategory_[1-5]_name: Names of the category and every subdivision of the category, given by the category, given by the categoryvalue: Price of one unity of the itemquantity: How many items were solddiscount: How much were given as discount to that item (if any)total_value: Total value paid for that item (Calculated as (value*quantity)-discount
Finally, we also have a CSV file, called mapping.csv, formatted as such:
| area | category_name | Comentario |
|---|---|---|
| X | VINHOS SUAVES | |
| 424 | VODKA | |
| 424 | WHISKY | |
| 425 | CEREAIS MATINAIS | |
| 425 | LEITES | |
| 424 | CERVEJAS | |
| 424 | CERVEJAS ESPECIAIS | |
| 421 | CHOC.CKOUT |
In that sample, the area 424 Contains products with the categories VODKA, WHISKY, CERVEJAS and CERVEJAS ESPECIAIS, for instance. An X means that the given category has no mapped area.
You are NOT required to complete this using any specific tool or language. Use whathever you are more confortable with. You are required to write a report, explaining what, how and why you did what you did, specifically, why do you think the extra metrics you generated are usefull. As stated before, you do not need to use any specific tool, but, for the report, we would love to see it using the Jupyter Notebook. If you are not familiar with it, you can check http://jupyter.org/ in order to install and give your first steps with it. Again, it's not required to use it, but it would be nice to.
Finally, your code should be avaiable on a private github repository. If you can't make it private, you can use bitbucket.
You can send your reports and link to repository to the email [email protected].
We believe that one week is enough time to complete this challenge with interesting results. But, if you think you will need more time, let us know as soon as possible. If you need any further clarification, also let us know, so we can help you through the proccess.