def graph_player_distributions(stats_df,
player_name,
team,
season,
teams_df,
cols,
pct_cols,
filter_by_position=False,
position_to_use=None):
team_info = get_team_info(stats_df=stats_df, teams_df=teams_df, team=team)
df_to_graph, player_stats_df, sub_title = format_dataframe(
stats_df=stats_df,
player_name=player_name,
team=team,
season=season,
cols=cols,
filter_by_position=filter_by_position,
position_to_use=position_to_use)
fig = plt.figure(figsize=(8, len(cols) * 4), constrained_layout=True)
gs = fig.add_gridspec(len(cols), 1)
chart_title = player_name
fig.suptitle(chart_title,
fontsize=30,
fontweight='bold',
y=1.02,
x=-.05,
horizontalalignment='left',
verticalalignment='bottom')
fig.text(x=1.05,
y=1.02,
s=sub_title,
horizontalalignment='right',
verticalalignment='top',
fontsize=24,
style='italic')
fig.text(x=-.05,
y=1.02,
s=f'{team_info["Full"]}, 2019-20',
horizontalalignment='left',
verticalalignment='top',
fontsize=24,
style='italic')
for index, col in enumerate(cols):
ax = fig.add_subplot(gs[index, 0])
is_pct_col = col in pct_cols
dist_plot(totals=df_to_graph,
stats_to_graph=player_stats_df,
col=col,
team=team_info,
ax=ax,
is_pct_col=is_pct_col)
plt.savefig(f'output/{player_name} {season} Shooting Distribution.png',
bbox_inches='tight',
pad_inches=2)
plt.close('all')
def predict(self):
try:
# model = self.load_model("randomForest")
model = self.load_model_by_database(self.config["algorithm"],
self.config["model"])
res = {}
if self.config['oneSample']:
if not self.config['X']:
raise ValueError(
"feature must not be empty when one-sample")
X = [[float(x) for x in self.config['X']]]
predict = model.predict(X)[0] if isinstance(
model.predict(X)[0], str) else "{:.0f}".format(
model.predict(X)[0])
res.update({
"data":
[[",".join([str(s) for s in self.config['X']]), predict]],
"title":
"单样本预测结果",
"col": ["样本特征", "模型预测结果"],
})
else:
# 从数据库拿数据
if not self.config['tableName'] or self.config[
'tableName'] == "":
raise ValueError(
"cannot find table data when multi-sample")
data = self.table_data
log.info("输入数据大小:{}".format(len(data)))
data = data.astype(float)
data["predict"] = model.predict(data.values)
if data["predict"].dtypes != "object":
data = format_dataframe(data, {"predict": ".0f"})
res.update(
transform_table_data_to_html({
"data": data.values.tolist(),
"title": "多样本预测结果",
"col": data.columns.tolist(),
"row": data.index.tolist()
}))
response_data = {"res": res, "code": "200", "msg": "ok!"}
return response_data
except Exception as e:
# raise e
log.exception("Exception Logged")
return {"data": "", "code": "500", "msg": "{}".format(e.args)}
def graph_player_with_shot_chart(stats_df,
player_name,
team,
season,
teams_df,
cols,
pct_cols,
shot_df,
filter_by_position=False,
position_to_use=None):
team_info = get_team_info(stats_df=stats_df, teams_df=teams_df, team=team)
df_to_graph, player_stats_df, sub_title = format_dataframe(
stats_df=stats_df,
player_name=player_name,
team=team,
season=season,
cols=cols,
filter_by_position=filter_by_position,
position_to_use=position_to_use)
fig = plt.figure(figsize=(40, len(cols) * 4), constrained_layout=True)
gs = fig.add_gridspec(len(cols), 6)
chart_title = player_name
fig.suptitle(chart_title,
fontsize=80,
fontweight='bold',
y=1.02,
x=0,
horizontalalignment='left',
verticalalignment='bottom')
fig.text(x=1,
y=1.02,
s=sub_title,
horizontalalignment='right',
verticalalignment='top',
fontsize=40,
style='italic')
fig.text(x=0,
y=1.02,
s=f'{team_info["Full"]}, 2019-20 Regular Season',
horizontalalignment='left',
verticalalignment='top',
fontsize=40,
style='italic')
fig.text(x=0,
y=-.06,
s='By Andrew Lawlor, Twitter: @lawlorpalooza',
horizontalalignment='left',
verticalalignment='top',
fontsize=40,
style='italic')
fig.text(
x=0,
y=-.08,
s='Shot Data from stats.nba.com, Statistics from Basketball Reference',
horizontalalignment='left',
verticalalignment='top',
fontsize=40,
style='italic')
ax1 = fig.add_subplot(gs[:, :5])
ax1 = draw_court(ax=ax1, outer_lines=False)
player_shot_df = shot_df[shot_df['PLAYER_NAME'] == player_name]
shot_chart(shot_df=player_shot_df, ax=ax1)
for index, col in enumerate(cols):
ax = fig.add_subplot(gs[index, 5])
is_pct_col = col in pct_cols
dist_plot(totals=df_to_graph,
stats_to_graph=player_stats_df,
col=col,
team=team_info,
ax=ax,
is_pct_col=is_pct_col,
show_colors=False,
fontsize=32)
plt.savefig(
f'output/player-shot-charts/{player_name} {season} Shot Chart.png',
bbox_inches='tight',
pad_inches=2)
plt.close('all')
def show_regression_result(self, x, y, model, options=[]):
"""
回归模型拟合效果展示
:param x: 特征列
:param y: 标签列
:param model: 已经训练好的回归模型
:param options: 可选参数，控制输出结果["coff", "independence", "resid_normal"]
:return: 给前端的结果
"""
# plt.rcParams['font.sans-serif'] = ['Arial Unicode MS']
# plt.rcParams['axes.unicode_minus'] = False
res = []
# 拟合优度
if "r2" in options:
res.append({
"is_test":
False,
"title":
"拟合优度",
"str":
str(model.summary().tables[0]).replace("\n", "<br/>")
})
# 系数解读
if "coff" in options:
res.append({
"is_test":
False,
"title":
"系数解读",
"str":
str(model.summary().tables[1]).replace("\n", "<br/>")
})
# 独立性检验
if "independence" in options:
res.append({
"is_test":
True,
"title":
"独立性检验",
"str":
str(model.summary().tables[2]).replace("\n", "<br/>")
})
# 残差正态性检验
if "resid_normal" in options:
sns.distplot(a=model.resid,
bins=10,
fit=stats.norm,
norm_hist=True,
hist_kws={
'color': 'green',
'edgecolor': 'black'
},
kde_kws={
'color': 'black',
'linestyle': '--',
'label': 'kernel density curve'
},
fit_kws={
'color': 'red',
'linestyle': ':',
'label': 'normal density curve'
})
plt.legend()
plt.title("残差正态性检验")
res.append({
"is_test":
True,
"title":
"残差正态性检验",
"base64":
"{}".format(self.plot_and_output_base64_png(plt))
})
# 残差pp图
if "pp" in options:
pp_qq_plot = sm.ProbPlot(model.resid)
pp_qq_plot.ppplot(line='45')
res.append({
"is_test":
True,
"title":
"残差pp图",
"base64":
"{}".format(self.plot_and_output_base64_png(plt))
})
# 残差qq图
if "qq" in options:
pp_qq_plot = sm.ProbPlot(model.resid)
pp_qq_plot.qqplot(line='q')
res.append({
"is_test":
True,
"title":
"残差qq图",
"base64":
"{}".format(self.plot_and_output_base64_png(plt))
})
# 标准化残差与预测值之间的散点图(验证残差的方差齐性)
if "var" in options:
plt.scatter(model.predict(),
(model.resid - model.resid.mean()) / model.resid.std())
plt.xlabel('predict value')
plt.ylabel('standardized residual ')
# 添加水平参考线
plt.axhline(y=0, color='r', linewidth=2)
res.append({
"is_test":
True,
"title":
"方差齐性检验",
"base64":
"{}".format(self.plot_and_output_base64_png(plt))
})
# 多重共线性检验
if len(x.columns) > 1 and "vif" in options:
X = sm.add_constant(x)
vif = pd.DataFrame()
vif['features'] = X.columns
vif["VIF Factor"] = [
variance_inflation_factor(X.values, i)
for i in range(X.shape[1])
]
vif = format_dataframe(vif, {"VIF Factor": ".4f"})
res.append(
self.transform_table_data_to_html({
"is_test":
True,
"title":
"多重共线性检验",
"row":
vif['features'].values.tolist(),
"col": ["VIF Factor"],
"data": [vif["VIF Factor"].values.tolist()],
"remarks":
"VIF>10，存在多重共线性，>100则变量间存在严重的多重共线性"
}))
# 线性相关性检验(留在数据探索里面展示)
# 异常值检测（帽子矩阵、DFFITS准则、学生化残差、Cook距离）
if "outliers" in options:
outliers = model.get_influence()
# 帽子矩阵
leverage = outliers.hat_matrix_diag
# dffits值
dffits = outliers.dffits[0]
# 学生化残差
resid_stu = outliers.resid_studentized_external
# cook距离
cook = outliers.cooks_distance[0]
# 合并各种异常值检验的统计量值
"""
"""
contatl = pd.concat([
pd.Series(leverage, name='leverage'),
pd.Series(dffits, name='dffits'),
pd.Series(resid_stu, name='resid_stu'),
pd.Series(cook, name='cook')
],
axis=1)
x.index = range(x.shape[0])
profit_outliers = pd.concat([x, contatl], axis=1)
profit_outliers = format_dataframe(
profit_outliers, {
"leverage": ".4f",
"dffits": ".4f",
"resid_stu": ".4f",
"cook": ".4f"
})
res.append(
self.transform_table_data_to_html({
"is_test":
True,
"title":
"异常值检测",
"row":
profit_outliers.index.tolist(),
"col":
profit_outliers.columns.tolist(),
"data":
profit_outliers.values.tolist(),
"remarks":
"当高杠杆值点（或帽子矩阵）大于2(p+1)/n时，则认为该样本点可能存在异常（其中p为自变量的个数，n为观测的个数）；当DFFITS统计值大于2sqrt((p+1)/n)时，则认为该样本点可能存在异常；当学生化残差的绝对值大于2，则认为该样本点可能存在异常；对于cook距离来说，则没有明确的判断标准，一般来说，值越大则为异常点的可能性就越高；对于covratio值来说，如果一个样本的covratio值离数值1越远，则认为该样本越可能是异常值。"
}))
# 预测值与真实值的散点图
if "pred_y_contrast" in options:
plt.scatter(model.predict(), y)
plt.plot([model.predict().min(),
model.predict().max()], [y.min(), y.max()],
'r-',
linewidth=3)
plt.xlabel("predict value")
plt.ylabel('true value')
res.append({
"is_test":
False,
"title":
"预测值与真实值对比散点图",
"base64":
"{}".format(self.plot_and_output_base64_png(plt))
})
return res
def plot_player_profile(player_name, team, season):
stats_df = pd.read_csv('./data/nba_per_poss.csv')
teams_df = pd.read_csv('./data/nba_team_colors.csv')
advanced_stats_df = pd.read_csv('./data/nba_advanced.csv')
radar_df = advanced_stats_df[[
'player', 'team_id', 'season', 'mp', 'per', 'obpm', 'dbpm', 'ast_pct',
'fg3a_per_fga_pct', 'ts_pct', 'fta_per_fga_pct', 'trb_pct', 'blk_pct',
'stl_pct', 'tov_pct', 'usg_pct'
]]
dists_df = stats_df[[
'player', 'team_id', 'season', 'fga_per_poss', 'fg2a_per_poss',
'fg3a_per_poss', 'fta_per_poss', 'fg2_pct', 'fg3_pct', 'ft_pct',
'pf_per_poss'
]]
shot_df = pd.read_csv('./data/all_shots.csv')
merged_df = radar_df.merge(dists_df, on=['player', 'team_id', 'season'])
merged_df.rename(columns={
'pf_per_poss': 'PF/100',
'fga_per_poss': 'FGA/100',
'fta_per_poss': 'FTA/100',
'fg3a_per_poss': '3PA/100',
'fg2a_per_poss': '2PA/100'
},
inplace=True)
cols = [
'mp',
'per',
'obpm',
'dbpm',
'ast_pct',
'fg3a_per_fga_pct',
'ts_pct',
'fta_per_fga_pct',
'trb_pct',
'blk_pct',
'stl_pct',
'tov_pct',
'usg_pct',
'3PA/100',
'2PA/100',
'FTA/100',
'fg2_pct',
'fg3_pct',
'ft_pct',
'PF/100',
]
pct_cols = [
'fg2_pct',
'fg3_pct',
'ft_pct',
'fg3a_per_fga_pct',
'ts_pct',
'fta_per_fga_pct',
]
team_info = get_team_info(stats_df=stats_df, teams_df=teams_df, team=team)
df_to_graph, player_stats_df, sub_title = format_dataframe(
stats_df=merged_df,
player_name=player_name,
team=team,
season=season,
cols=cols,
)
N = len(cols) + 1
fig = plt.figure(figsize=(50, N * 2), constrained_layout=True)
gs = fig.add_gridspec(N, 6)
chart_title = player_name
fig.suptitle(chart_title,
fontsize=120,
fontweight='bold',
y=1.02,
x=0,
horizontalalignment='left',
verticalalignment='bottom')
fig.text(x=1,
y=1.02,
s=sub_title,
horizontalalignment='right',
verticalalignment='top',
fontsize=60,
style='italic')
fig.text(x=0,
y=1.02,
s=f'{team_info["Full"]}, 2019-20',
horizontalalignment='left',
verticalalignment='top',
fontsize=60,
style='italic')
ax1 = fig.add_subplot(gs[9:, :5])
ax1 = draw_court(ax=ax1, outer_lines=False)
player_shot_df = shot_df[shot_df['PLAYER_NAME'] == player_name]
shot_chart(shot_df=player_shot_df, ax=ax1, ylim=(-47.5, 300))
player_stats = merged_df[(merged_df['player'] == player_name)
& (merged_df['season'] == season) &
(merged_df['team_id'] == team)].squeeze()
radar_plot(
totals_df=merged_df,
stats_to_graph=player_stats,
player_name=player_name,
team_info=team_info,
season=2020,
cols=[
'fta_per_fga_pct',
'ts_pct',
'fg3a_per_fga_pct',
'ast_pct',
'usg_pct',
'tov_pct',
'PF/100',
'stl_pct',
'blk_pct',
'trb_pct',
],
pct_cols=[
'ts_pct',
'fg3a_per_fga_pct',
'fta_per_fga_pct',
],
inverse_cols=[
'tov_pct',
'PF/100',
],
fig=fig,
gs=gs[1:9, :5],
label_font_size=48,
tick_font_size=32,
label_padding=60,
)
for index, col in enumerate(cols):
ax = fig.add_subplot(gs[index + 1, 5])
is_pct_col = col in pct_cols
dist_plot(totals=df_to_graph,
stats_to_graph=player_stats_df,
col=col,
team=team_info,
ax=ax,
is_pct_col=is_pct_col)
plt.savefig(f'output/{player_name} 2019-20 Profile.png',
bbox_inches='tight',
pad_inches=2)