ML skills for product managers

New Xm17
Oct 13, 2018 6 Comments

Any product managers working with data scientists -ML powered products? as a product guy what ML expertise you need to have ?

comments

Want to comment? LOG IN or SIGN UP
TOP 6 Comments
  • I'm a research scientist but I can tell what skills classical pms would lack. Understanding the risk associated with ML solutions as a non-deterministic software (read on precision recall, false positive, false neg. 2. Feature pipelines and versioning. 3. Model training and precision decay 4. Defining KPIs which correlate with algorithm performance
    Oct 13, 2018 2
    • New Xm17
      OP
      @ebay thanks .. risk associated I understand needs to be shared with data scientists.. KPI / precision decay on algo performance also makes sense for feedback loop ..

      What kinds of extra inpute are required for features pipeline and versioning and model training
      Oct 13, 2018
    • Less of your input for data scietist but communication with the data teams and designing processes to make the data always available and valid for model training
      Oct 13, 2018
  • Oracle xxPh88
    All of the above plus you really have to understand that ML is a data problem, so classic ETL, data normalization, just knowing what to chase is valuable. You need to have some skills in data analysis as the data science team may likely know hyper parameter tuning but need help bridging to the problem. Even when you get to feature engineering, you need to be the voice of the use case or the problem. The last thing you need to know is which classifier.
    Oct 13, 2018 1
    • New Xm17
      OP
      I have a sense of data processing, training n serving needs.. have written some basic models.. but if PMs are able to define the success criteria comprehensively.. shouldn't that be enough? I can learn more data prep mechanisms..n model tuning but shouldn't that be responsibility of Lead/ Managei data scientists
      Oct 13, 2018
  • Amazon Manamana!
    As a PM for this sort of thing you need to find and attract talent on both the data and the science side. You need to know how you will make a difference for customers, where you will get training data, and how to run controlled experiments. You need to ship experiments that deliver results and that gets you the proof points to go back for more resources (people) that you need to extend the initial use cases.
    Oct 13, 2018 0