More Detailed EDA

[Anteemony]

• more detailed EDA
  • [ ] #5
  • [ ] #6
  • [ ] #7
  • [ ] #8
  • action: formulate some basic questions and answer them
    • write down questions at the start, what are you investigating?
    • e.g., promotions impact sales?
    • ability to forecast sales?
    • up to you, but ensure you have these at the start
    • think about ways to address these
  • ensure to have "basic" EDA elements
    • report size of data, type of data
    • report summary statistics
    • e.g., how many shops are there
    • how do shops distribute by sales
    • univariate plots of summaries by shop
      • interesting summary - average sale, type of good
    • correlation analyses and plots for things like
      • shops vs sales
      • good type vs sales
      • sales per time unit by shop or good type
      • promotions
  • put the above in markdown cells (observations) and python cells

[fkiraly]

Adding my thoughts here, based on the current notebook:

• basic parameters of the data set should be summarized at the start
• basic summaries should also be at the start:
  • how many stores? how many product groups?
  • coverage in terms of start/end of record per store
  • summary of sales by store, do statistical summaries on the summaries. E.g., average sales in a year, variation, etc.
• there are marked "jumps" in the mean aggregate over stores. What explains these? Could be artefact start/end
• similar analysis for product groups
• interaction analysis product groups and stores. Which stores sell which products? How does the product compositiòn distribute?