In [1]:

import os
import glob
from collections import defaultdict

import numpy as np
import pandas as pd

import matplotlib.pyplot as plt

import watchcbb.utils as utils

In [2]:

dfs = []
for fname in glob.glob("../data/game_data/*.csv"):
    dfs.append(pd.read_csv(fname))
df = pd.concat(dfs)
df.Date = pd.to_datetime(df.Date)
df = df.sort_values("Date").reset_index()
# df = df.query("Season==2019").reset_index()
df["poss"] = 0.5*(df["WFGA"] + 0.44*df["WFTA"] - df["WOR"] + df["WTO"] + df["LFGA"] + 0.44*df["LFTA"] - df["LOR"] + df["LTO"])
print("Shape:",df.shape)
df.head(10)

Shape: (54760, 37)

Out[2]:

	index	Season	Date	Type	WTeamID	WScore	LTeamID	LScore	WLoc	...	LFTM	LFTA	LOR	LDR	LAst	LTO	LStl	LBlk	LPF	poss
0	0	2011	2010-11-08	RG	illinois	79	california-irvine	65	H	...	14	22	11	26	12	14	7	1	21	74.90
1	3	2011	2010-11-08	RG	pittsburgh	83	rhode-island	75	H	...	13	16	9	20	16	16	7	6	25	75.62
2	1	2011	2010-11-08	RG	maryland	105	seattle	76	H	...	17	22	13	16	11	27	18	1	31	98.36
3	2	2011	2010-11-08	RG	texas	83	navy	52	H	...	10	18	5	21	8	16	5	5	24	80.44
4	4	2011	2010-11-10	RG	maryland	75	college-of-charleston	74	H	...	16	19	10	25	12	22	8	8	19	77.64
5	5	2011	2010-11-10	RG	pittsburgh	97	illinois-chicago	54	H	...	7	9	5	15	9	11	0	3	18	67.88
6	6	2011	2010-11-10	RG	illinois	84	toledo	45	H	...	10	19	3	16	9	23	6	0	14	72.00
7	7	2011	2010-11-10	RG	texas	89	louisiana-tech	58	H	...	13	20	13	25	4	22	6	5	21	83.74
8	78	2011	2010-11-12	RG	tulane	91	maryland-eastern-shore	62	A	...	19	26	9	16	11	21	4	2	28	78.04
9	77	2011	2010-11-12	RG	marquette	97	prairie-view	58	H	...	7	11	3	13	12	20	11	3	27	79.12

10 rows × 37 columns

In [3]:

first, second = utils.partition_games(df, frac=0.7)
print(df.iloc[first].shape[0], df.iloc[second].shape[0])

38328 16432

In [4]:

season_stats_dict = utils.compute_season_stats(df.iloc[first])
season_stats_df = utils.stats_dict_to_df(season_stats_dict)
utils.add_advanced_stats(season_stats_df)
season_stats_dict = utils.stats_df_to_dict(season_stats_df)
print(season_stats_df.shape)
season_stats_df.head()

(3497, 48)

Out[4]:

	year	team_id	wins	losses	totOT	TScore	TFGM	TFGA	TFGM3	TFGA3	...	Torbp	Tftr	Oposs	Oeff	Oastr	Otovr	Oefgp	Oorbp	Oftr	rawpace
0	2011	air-force	12	8	2	1328	463	941	133	333	...	0.192678	0.414453	1261.96	101.825731	0.157057	0.169381	0.485780	0.342949	0.283486	61.890370
1	2011	akron	12	10	2	1498	527	1263	175	508	...	0.289833	0.306413	1522.28	96.434296	0.137450	0.190418	0.480897	0.286472	0.383721	68.349663
2	2011	alabama	15	7	1	1493	566	1249	96	311	...	0.305990	0.293034	1502.48	83.595123	0.110456	0.199252	0.409393	0.350689	0.346633	67.763164
3	2011	alabama-am	6	10	0	959	348	896	62	237	...	0.299213	0.377232	1143.32	90.875695	0.136156	0.209393	0.472254	0.305206	0.399773	71.501250
4	2011	alabama-birmingham	16	6	5	1537	547	1224	166	480	...	0.291248	0.307190	1472.84	94.375492	0.130914	0.164221	0.457404	0.330330	0.287420	64.801768

5 rows × 48 columns

In [5]:

season_stats_df.query('year==2019').sort_values('rawpace').tail()

Out[5]:

	year	team_id	wins	losses	totOT	TScore	TFGM	TFGA	TFGM3	TFGA3	...	Torbp	Tftr	Oposs	Oeff	Oastr	Otovr	Oefgp	Oorbp	Oftr	rawpace
2903	2019	green-bay	10	12	2	1787	632	1412	184	533	...	0.257468	0.345609	1720.64	108.854845	0.144713	0.158850	0.539202	0.255114	0.360656	77.333933
3000	2019	north-carolina	19	4	1	2031	726	1532	203	535	...	0.276273	0.327023	1776.72	96.019632	0.144185	0.170009	0.489184	0.261494	0.305652	77.711568
3063	2019	savannah-state	7	15	1	1654	581	1508	253	884	...	0.300539	0.241379	1759.56	112.869126	0.183773	0.164379	0.560926	0.273128	0.322668	80.038870
2876	2019	eastern-kentucky	8	14	3	1766	622	1486	189	592	...	0.296044	0.327052	1788.84	105.096040	0.145986	0.188988	0.538293	0.240936	0.409190	80.484469
2887	2019	florida-international	10	10	0	1604	563	1313	169	568	...	0.282468	0.369383	1592.84	106.037016	0.144561	0.192815	0.547049	0.264339	0.367624	80.731000

5 rows × 48 columns

In [6]:

year = 2019
tid1, tid2 = 'north-carolina', 'virginia'
poss1 = df.query(f'Season=={year} & (WTeamID=="{tid1}" | LTeamID=="tid1")').poss.values
poss2 = df.query(f'Season=={year} & (WTeamID=="{tid2}" | LTeamID=="tid2")').poss.values
plt.figure(figsize=(9,7))
plt.hist(poss1, bins=np.linspace(50,100,26), histtype='step', lw=2, label=str(year)+' '+tid1, color='c')
plt.hist(poss2, bins=np.linspace(50,100,26), histtype='step', lw=2, label=str(year)+' '+tid2, color='r')
plt.legend(fontsize='xx-large')
plt.xlabel('Pace', fontsize='x-large')

Out[6]:

Text(0.5, 0, 'Pace')

In [7]:

mean_poss = {year:season_stats_df.query(f'year=={year}').rawpace.mean() for year in season_stats_df.year.unique()}
for i,row in season_stats_df.iterrows():
    season_stats_dict[row.year][row.team_id]["rawpace"] = row.rawpace

def predict_poss(row):
    p1 = season_stats_dict[row.Season][row.WTeamID]["rawpace"]
    p2 = season_stats_dict[row.Season][row.WTeamID]["rawpace"]
    return p1*p2/mean_poss[row.Season]

pred_poss = df.iloc[second].apply(predict_poss, axis=1).values

In [8]:

from sklearn.linear_model import LinearRegression

plt.figure(figsize=(9,7))
xs = pred_poss.reshape(-1,1)
ys = df.iloc[second].poss.values.reshape(-1,1)
plt.scatter(xs, ys, s=20, alpha=0.05)

linreg = LinearRegression()
linreg.fit(xs, ys)
linreg.score(xs,ys)
xl = np.array([50,100]).reshape(-1,1)
yl = linreg.predict(xl)
plt.plot(xl, yl, 'r-')

Out[8]:

[<matplotlib.lines.Line2D at 0x7fcb7d425c90>]

In [9]:

plt.figure(figsize=(9,8))
xs = season_stats_df.query('year>=2016').Teff
ys = season_stats_df.query('year>=2016').Oeff
plt.scatter(xs, ys, s=20, alpha=0.3)
plt.xlabel("Offensive efficiency", fontsize='large')
plt.ylabel("Defensive efficiency", fontsize='large')

Out[9]:

Text(0, 0.5, 'Defensive efficiency')

In [10]:

season_stats_df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 3497 entries, 0 to 3496
Data columns (total 48 columns):
 #   Column   Non-Null Count  Dtype  
---  ------   --------------  -----  
 0   year     3497 non-null   int64  
 1   team_id  3497 non-null   object 
 2   wins     3497 non-null   int64  
 3   losses   3497 non-null   int64  
 4   totOT    3497 non-null   int64  
 5   TScore   3497 non-null   int64  
 6   TFGM     3497 non-null   int64  
 7   TFGA     3497 non-null   int64  
 8   TFGM3    3497 non-null   int64  
 9   TFGA3    3497 non-null   int64  
 10  TFTM     3497 non-null   int64  
 11  TFTA     3497 non-null   int64  
 12  TOR      3497 non-null   int64  
 13  TDR      3497 non-null   int64  
 14  TAst     3497 non-null   int64  
 15  TTO      3497 non-null   int64  
 16  TStl     3497 non-null   int64  
 17  TBlk     3497 non-null   int64  
 18  TPF      3497 non-null   int64  
 19  OScore   3497 non-null   int64  
 20  OFGM     3497 non-null   int64  
 21  OFGA     3497 non-null   int64  
 22  OFGM3    3497 non-null   int64  
 23  OFGA3    3497 non-null   int64  
 24  OFTM     3497 non-null   int64  
 25  OFTA     3497 non-null   int64  
 26  OOR      3497 non-null   int64  
 27  ODR      3497 non-null   int64  
 28  OAst     3497 non-null   int64  
 29  OTO      3497 non-null   int64  
 30  OStl     3497 non-null   int64  
 31  OBlk     3497 non-null   int64  
 32  OPF      3497 non-null   int64  
 33  Tposs    3497 non-null   float64
 34  Teff     3497 non-null   float64
 35  Tastr    3497 non-null   float64
 36  Ttovr    3497 non-null   float64
 37  Tefgp    3497 non-null   float64
 38  Torbp    3497 non-null   float64
 39  Tftr     3497 non-null   float64
 40  Oposs    3497 non-null   float64
 41  Oeff     3497 non-null   float64
 42  Oastr    3497 non-null   float64
 43  Otovr    3497 non-null   float64
 44  Oefgp    3497 non-null   float64
 45  Oorbp    3497 non-null   float64
 46  Oftr     3497 non-null   float64
 47  rawpace  3497 non-null   float64
dtypes: float64(15), int64(32), object(1)
memory usage: 1.3+ MB