construct a label from other benchmark fields #3

Activity

1. cwpearson

   assigned

   abduld

   ,

   cwpearson

   and

   cli99

   on Aug 13, 2018
2. abduld commented on Aug 13, 2018

   

   abduld

   on Aug 13, 2018

   More actions

   

   e.g. from https://ml-arc-minsky.netlify.com/machine/AWS_g3_xlarge
3. cwpearson commented on Aug 13, 2018

   

   cwpearson

   on Aug 13, 2018

   MemberAuthor

   More actions

   Can you add a snippet of the benchmark.json file as well?
4. abduld commented on Aug 13, 2018

   

   abduld

   on Aug 13, 2018

   More actions

   e.g.

    {
         "name": "SGEMM/384/169/1",
         "iterations": 13074,
         "real_time": 5.3525743996501726e+04,
         "cpu_time": 5.3539960532357516e+04,
         "time_unit": "ns",
         "K": 1.0000000000000000e+00,
         "M": 3.8400000000000000e+02,
         "N": 1.6900000000000000e+02
       }, 
   

   https://github.com/rai-project/scope/blob/master/results/raw_data/minsky/sgemm_smt_0.json
5. cwpearson commented on Aug 16, 2018

   

   cwpearson

   on Aug 16, 2018

   Last edited by cwpearson

   MemberAuthor

   More actions

   What about something like this. This would construct a string with the specified seperator, from the fields in each benchmark object

   In the spec file

   label:
     seperator: "x" # optional, there would be some default
     fields:
     -  K
     -  M
     -  N

   or for command line utilities

   ... --label-seperator x --label-fields K M N ...
   

   The result for the offered snippet would be

   1x384x169
   
6. cwpearson commented on Aug 21, 2018

   

   cwpearson

   on Aug 21, 2018

   MemberAuthor

   More actions

   Some more thoughts.
   In the spec file, at the series level, series.label is the name of the series in the plot legend right now, so perhaps add another two optional fields series.tick_fields and series.tick_seperator

   series:
     - ...
       tick_seperator: "x" # optional, there would be some default
       tick_fields: [K, M, N]
     - ...

   or for command line utilities

   ... --tick-seperator x --tick-fields K M N ...
   

   The result for the offered snippet would be

   1x384x169
   

   There will need to be some changes in how dataframes are generated in scope_plot/benchmark.py and consumed in the generate_bar() portions of scope_plot/backends/....
   I think when the dataframe is generated, the index of the dataframe should be set to the desired tick label.
   Then, when that dataframe is consumed the logic would need to be adjusted a bit to plot it correctly.
7. cwpearson commented on Aug 21, 2018

   

   cwpearson

   on Aug 21, 2018

   MemberAuthor

   More actions

   Added support for series.tick_fields and series.tick_seperator in schema.py, but no logic to actually use them.

   bc739ac