casper-house / machinelearningchallenge

Machine Learning Challenge

Paytm has several merchants who sell merchandise across different categories. As a merchant performance manager I want to understand how well a merchant is doing but since there are so many conflicting dimensions it is hard to say how is merchant performing overall. As an example some merchants may have great service like product they advertise is exactly what they deliver, others may deliver a variation or overpromise to customer. Similarly, a merchant may sell a lot of units but they get ship late consistently and hence cause bad customer experience.

We want you to come up with a universal score for every merchant to summarize their performance. Think of this as a consumer credit score where many variables are considered and you get positive points for some behaviors and negative for others.

##There are three datasets provided to you –

1. [Transactions:] – These are all the transactions from users in a fixed time period. This table has merchant id which will let you calculate summary statistics for every merchant. Here is the schema and data is in CSV format.

2. merchant_id: Unique ID representing a merchant

3. T1:Root Category

4. T2: Sub Category

5. T4: Primary Category

6. order_id: Order id, think of an order as shopping cart

7. order_item_id: Unique ID for every item in an order

8. product_id: Product ID

9. item_created_at: Time of order creation

10. item_ship_by_date: Latest date by which item must be shipped to the customer ( SLA driven)

11. qty_ordered: Number of units ordered for an item

12. item_mp: Original price of an item

13. item_price: Price at which merchant is offering this product

14. item_selling_price: Price at which this item was sold.

15. item_discount: Discount offered by merchant

16. fulfillment_shipped_at: Time when item was shipped from merchant to customer

17. fulfillment_created_at: Time when shipment request was created for the item and merchant starts preparing the item for shipment.

18. [Profit Metrics]– This dataset has aggregated view of 3 metrics for all merchants- a. Commission%- Average commission paid by merchant to us, aggregated by primary category (see concept of categories section below). b. Discount %- Average discount offered by merchant to customers, aggregated by primary category. c. Cash back %- Average cash back offered by us to customers on items sold by merchant, aggregated by primary category. Paytm uses cashback as a preferred method for promotions. Here is the schema and data is in CSV format.

19. merchant_id: Unique Merchant_id

20. T1: Root category

21. T2: Sub Category

22. T4: Primary Category

23. commission_percent: Average % commission charged from merchant. This is aggregated at primary category level.

24. cashback_percent: Average % cashback (discount from Paytm) given across all items sold by merchant. This is aggregated at primary category level.

25. discount_percent: Average % discount given by merchant across all the orders. This is aggregated at primary category level.

26. [Returned Cancelled Metrics] – This dataset has all merchant related cancelled and retuned orders. Merchants sometime cancel the order in case they are not able to acquire inventory or maybe they accepted an order but does not want to ship to a particular pin code. Returns can happen if customer receives order late since merchant did not ship in time or item received is not as promised on the platform. Cancelled and returned order here are only because of merchants’ fault and general customer cancellations are not a part of it.Here is the schema and data is in CSV format.

27. merchant_id: Unique Merchant_id

28. T1: Root category

29. T2: Sub Category

30. T4: Primary Category

31. cancel_num: Total number of orders cancelled by merchant, aggregated at primary category level

32. return_num: Total number of returned orders fulfilled by merchant, aggregated at primary category level

##Concept of categories One thing you will notice in the datasets is category fields T1, T2, T4. All the catalogs are structured in some hierarchy. We have shared 3 levels of hierarchy where T1 -> T2 -> T4 -> Product ID. As an example Mobile and Accessories is a Root category or T1, under this we have 2 Sub categories 1) Mobiles and 2) Accessories and finally each T2 category has many T4 categories (primary category) as an example T2 category Mobiles can have 2 primary categories 1) Smart phones 2) Feature phones.

##How to complete this challenge:

1. Get the data from the links mentioned in data section below.

2. Fork this repo in github https://github.com/PaytmLabs/MachineLearningChallenge

3. Complete the processing and analytics as defined first to the best of your ability with the time provided.

4. Place notes in your code to help with clarity where appropriate. Make it readable enough to present to the Paytm Labs interview team.

5. Complete your work in your own github repo and send the results and code to us and/or present them during your interview.

##What is expected:

1. Provide a list of merchant id and score and explanation of why you think some merchants should have higher score than others.

2. We expect you to be good in at least 1 programming language (Scala, java, Python, R). Please don’t complete this using one of the packaged solutions (example SAS/SPSS)

3. A good story is as important as an algorithm. We expect you to be able to communicate to us your idea, methodology and implementation. Please create a read-me describing your approach. (Visualizations can be submitted in appendix and are always appreciated but they should convey some message. Interactive visualizations get extra points)

4. Please make assumptions where necessary, we are interested in approach and if you can defend your assumption we will accept it.

5. If you find some parts of this assignment challenging or not able to finish it, don't hestitate to let us know. We can always help answer the questions and/or reduce the scope. Ultimately,journey is more important than destination!

##Data

1. Transaction Data - https://s3.amazonaws.com/databricks-dump/datalakedr/mlTest/final/transactions/part-00000
2. Profit Metrics Data - https://s3.amazonaws.com/databricks-dump/datalakedr/mlTest/final/profitMetrics/part-00000)
3. Returned Cancelled Data - https://s3.amazonaws.com/databricks-dump/datalakedr/mlTest/final/returnedCancelledMetrics/part-00000