Model_Trees_Full_Dataset: NHL Draft Data

For the NBA draft dataset, please see the readme.

If you use our datasets, please cite our paper

Provenance

• on-line sources: hockey data crawled from NHL, Eliteprospects and Draft Analyst. Our dataset contains all the players who got an NHL contract drafted from 1998 to 2008( excluded 2003 due to too many missing CSS_rank data that year). Thank you to David Wilson for letting us use his draft data set!
• local sources: Databases are stored in cs-oschulte-01.cs.sfu.ca
  • NHL season stats. Used to compute 7-years career stats.
    • github name: NHL_season_stats_1997_2017_original.csv
    • Database Table Name: chao_draft.NHL_season_stats_1998_2016_original
  • Draft Stats.
    • github name: elite_prospects_skaters_stats_1998_2008_original.csv
    • database name: chao_draft.elite_prospects_skaters_stats_1998_2008_original
  • Complete Table. Combines NHL season stats and Draft Stats. This is the file on which we base our learning method.(po_PlusMinus were dropped due to missing values in Eliteprospects
    • github name: raw_datasets.csv
    • Database Table Name: chao_draft.join_skater_and_season_stats_10_years_without_po_plusminus
  • Processed Stats. Applying preprocessing steps to the complete raw data table.
    • Processing Steps:
      1. sum_7yr_GP > 0 is entered in column gp_7_yr > 0 as target class label.
      2. null values for counts are converted to 0 (e.g. po_GP = null becomes po_GP = 0). So if a player didn't play in the playoffs, they are assigned 0 for all playoff stats.
      3. countries -> country_group = USA, CAN, EURO. So European countries are lumped together (no disrespect intended).
      4. null values for CSS_rank are converted to the (maximum_rank + 1) of the corresponding year.
    • github name: preprocessed_datasets.csv
    • database name: chao_draft.join_skaters_preprocessed_without_po_PlusMinus
  • Normalized Stats. This is the input for model tree learning.
    • Processing Steps:
      1. standardize data to the same range, i.e. x := x - min(x)/(max(x) - min(x)), see here. The max is chosen with respect to each cohort (cohort 1 has draft year 1998-2002, cohort 2 has draft year 2004-2008).
    • github name: normalized_datasets.csv
    • database name: chao_draft.norm_data_without_po_PlusMinus

Semantics

Field Name	Explanation
id	nhl.com id for NHL players, otherwise Eliteprospects.com id
Draft Age	Age in Draft Year
Country_group	Nationality. Canada -> 'CAN', USA -> 'USA', countries in Europe -> 'EURO'
Position	Position in Draft Year. Left Wing -> 'L', Right Wing -> 'R', Center -> 'C', Defencemen -> 'D'
Overall	Overall pick in NHL Entry Draft
CSS_rank	Central scouting service ranking in the draft year
rs_GP	Games played in regular seasons in the draft year
rs_G	Goals in regular seasons in the draft year
rs_A	Assists in regular seasons in the draft year
rs_P	Points in regular seasons in the draft year
rs_PIM	Penality_in_Minutes in regular seasons in the draft year
rs_PlusMinus	Goal Differential in regular seasons in the draft year
po_GP	Games played in playoffs in the draft year
po_G	Goals in playoffs in the draft year
po_A	Assists in playoffs in the draft year
po_P	Points in playoffs in the draft year
po_PIM	Penality_in_Minutes in playoffs in the draft year
po_PlusMinus	Goal Differential in playoffs in the draft year
sum_7yr_GP	Total games played in player's first 7 years of NHL career
sum_7yr_TOI	Total Time_on_Ice in player's first 7 years of NHL career
GP_7yr_greater_than_0	Played a game or not in player's first 7 years of NHL career