KNN exercise with NBA player data

Introduction

NBA player statistics from 2014-2015 (partial season): data, data dictionary
Goal: Predict player position using assists, steals, blocks, turnovers, and personal fouls

Step 1: Read the data into Pandas



In [1]:

    
# read the data into a DataFrame
import pandas as pd
url = 'https://raw.githubusercontent.com/kjones8812/DAT4-students/master/kerry/Final/NBA_players_2015.csv'
nba = pd.read_csv(url, index_col=0)
nba.head()









    Out[1]:






  
    
      
      season_end
      player
      pos
      age
      bref_team_id
      g
      gs
      mp
      fg
      fga
      ...
      TOV%
      USG%
      OWS
      DWS
      WS
      WS/48
      OBPM
      DBPM
      BPM
      VORP
    
  
  
    
      0
      2015
      Quincy Acy
      F
      24
      NYK
      52
      21
      19.2
      2.2
      4.6
      ...
      15.1
      14.7
      0.6
      0.5
      1.0
      0.050
      -2.6
      -0.7
      -3.4
      -0.3
    
    
      1
      2015
      Jordan Adams
      G
      20
      MEM
      18
      0
      7.3
      1.0
      2.1
      ...
      15.9
      17.7
      0.0
      0.2
      0.2
      0.076
      -2.3
      1.8
      -0.5
      0.0
    
    
      2
      2015
      Steven Adams
      C
      21
      OKC
      51
      50
      24.2
      3.0
      5.5
      ...
      19.2
      14.8
      1.0
      1.8
      2.8
      0.109
      -2.0
      2.0
      -0.1
      0.6
    
    
      3
      2015
      Jeff Adrien
      F
      28
      MIN
      17
      0
      12.6
      1.1
      2.6
      ...
      12.9
      14.1
      0.2
      0.2
      0.4
      0.093
      -2.6
      0.8
      -1.8
      0.0
    
    
      4
      2015
      Arron Afflalo
      G
      29
      TOT
      60
      54
      32.5
      5.0
      11.8
      ...
      10.9
      19.6
      1.4
      0.7
      2.1
      0.051
      -0.2
      -1.4
      -1.6
      0.2
    
  

5 rows × 49 columns



In [2]:

    
# examine the columns



In [3]:

    
# examine the positions

Step 2: Create X and y

Use the following features: assists, steals, blocks, turnovers, personal fouls



In [4]:

    
# map positions to numbers



In [5]:

    
# create feature matrix (X) (use fields: 'ast', 'stl', 'blk', 'tov', 'pf')



In [6]:

    
# create response vector (y)

Step 3: Train a KNN model (K=5)



In [7]:

    
# import class



In [8]:

    
# instantiate with K=5



In [9]:

    
# fit with data

Step 4: Predict player position and calculate predicted probability of each position

Predict for a player with these statistics: 1 assist, 1 steal, 0 blocks, 1 turnover, 2 personal fouls



In [10]:

    
# create a list to represent a player



In [11]:

    
# make a prediction



In [12]:

    
# calculate predicted probabilities

Step 5: Repeat steps 3 and 4 using K=50



In [13]:

    
# repeat for K=50



In [14]:

    
# calculate predicted probabilities

Bonus: Explore the features to decide which ones are predictive



In [15]:

    
# allow plots to appear in the notebook
%matplotlib inline
import matplotlib.pyplot as plt

# increase default figure and font sizes for easier viewing
plt.rcParams['figure.figsize'] = (6, 4)
plt.rcParams['font.size'] = 14

	season_end	player	pos	age	bref_team_id	g	gs	mp	fg	fga	...	TOV%	USG%	OWS	DWS	WS	WS/48	OBPM	DBPM	BPM	VORP
0	2015	Quincy Acy	F	24	NYK	52	21	19.2	2.2	4.6	...	15.1	14.7	0.6	0.5	1.0	0.050	-2.6	-0.7	-3.4	-0.3
1	2015	Jordan Adams	G	20	MEM	18	0	7.3	1.0	2.1	...	15.9	17.7	0.0	0.2	0.2	0.076	-2.3	1.8	-0.5	0.0
2	2015	Steven Adams	C	21	OKC	51	50	24.2	3.0	5.5	...	19.2	14.8	1.0	1.8	2.8	0.109	-2.0	2.0	-0.1	0.6
3	2015	Jeff Adrien	F	28	MIN	17	0	12.6	1.1	2.6	...	12.9	14.1	0.2	0.2	0.4	0.093	-2.6	0.8	-1.8	0.0
4	2015	Arron Afflalo	G	29	TOT	60	54	32.5	5.0	11.8	...	10.9	19.6	1.4	0.7	2.1	0.051	-0.2	-1.4	-1.6	0.2