-
Notifications
You must be signed in to change notification settings - Fork 27
/
Copy pathutils.py
199 lines (176 loc) · 6.82 KB
/
utils.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
import numpy as np
import pandas as pd
import scipy.stats as st
from plotnine import *
from mizani.breaks import *
from mizani.formatters import *
import seaborn as sns
from sklearn.linear_model import LinearRegression
#######################################################################
'''
Snippet for plotnine with thai font by @korakot
https://gist.github.com/korakot/01d181229b21411b0a20784e0ca20d3d
'''
import matplotlib
# !wget https://github.com/Phonbopit/sarabun-webfont/raw/master/fonts/thsarabunnew-webfont.ttf -q
# !cp thsarabunnew-webfont.ttf /usr/share/fonts/truetype/
matplotlib.font_manager._rebuild()
matplotlib.rc('font', family='TH Sarabun New')
theme_set(theme_minimal(11, 'TH Sarabun New'))
#######################################################################
#cleaning
def check_missing(df):
per_missing = df.isnull().mean()
missing_df = pd.DataFrame({'col_name': df.columns, 'per_missing': per_missing})
missing_df = missing_df.sort_values('per_missing',ascending=False).reset_index(drop=True)
missing_df['rnk'] = missing_df.index.map(lambda x: str(x).zfill(2)+'_') + missing_df.col_name
return missing_df
def check_mode(df):
mode_df = []
for col in df.columns:
x = df[col].value_counts()
mode_df.append({'col':col, 'value':x.index[0], 'per_mode': list(x)[0]/df.shape[0],
'nb_value':len(x)})
mode_df = pd.DataFrame(mode_df)[['col','value','per_mode','nb_value']]\
.sort_values('per_mode',ascending=False)
return mode_df.reset_index(drop=True)
def remove_outliers(df,col):
q1 = np.percentile(df[col], 25)
q3 = np.percentile(df[col], 75)
iqr = q3-q1
df = df[(df[col] < q3+1.5*iqr)&(df[col] > q1-1.5*iqr)]
return df.reset_index(drop=True)
def value_dist(df,col):
x = pd.DataFrame(df[col].value_counts()).reset_index()
x.columns = ['value','cnt']
x['per'] = x.cnt / x.cnt.sum()
return x
def otherify(df,col, th=0.03, retain=['NA']):
value_df = value_dist(df,col)
other_cols = list(value_df[value_df.per<th].value)
df[col] = df[col].map(lambda x: 'others' if (x in other_cols and x not in retain) else x)
return df
def replace_dict(x, d):
for key,value_list in d.items():
for v in value_list:
if v in str(x).lower():
return key
return 'others'
#visualization
def thai_text(font_size=11):
#thanks to @korakot for Thai font with plotnine
return theme(text=element_text(size=font_size, family='TH Sarabun New'))
def cat_plot(df,col):
g = (ggplot(df,aes(x=col)) +
geom_bar(stat='bin', #histogram
binwidth=0.5, #histogram binwidth
bins=len(df[col].unique())) + #how many bins
coord_flip()
# theme(axis_text_x=element_blank())
)
return g
def numcat_plot(df,num,cat, no_outliers=True, geom=geom_boxplot()):
if no_outliers:
new_df = remove_outliers(df,num)
else:
new_df = df.copy()
g = (ggplot(new_df, aes(x=cat,y=num)) +
geom
)
return g
def numdist_plot(df, num,cat, geom=geom_density(alpha=0.5), no_outliers=True):
if no_outliers:
new_df = remove_outliers(df,num)
else:
new_df = df.copy()
g = (ggplot(new_df,aes(x=num, fill=cat)) +
geom
)
return g
def catcat_plot(df, cat_dep, cat_ind):
plot_df = df.copy()
plot_df['cnt'] = 1
df_total = plot_df[[cat_ind,'cnt']].groupby(cat_ind).count().reset_index()
df_agg = plot_df[[cat_dep,cat_ind,'cnt']].groupby([cat_dep,cat_ind]).count().reset_index()
df_agg = df_agg.merge(df_total,on='color')
df_agg['per'] = df_agg.cnt_x / df_agg.cnt_y
g = (ggplot(df_agg,aes(x=cat_dep, y='per',fill=cat_dep)) +
geom_col() +
geom_text(aes(x=cat_dep,y='per',label='cnt_x')) +
theme(axis_text_x = element_blank()) +
facet_wrap(f'~{cat_ind}')) + theme(panel_spacing_x=0.5)
return g
def value_dist_plot(df,bins=30):
num_m = df.melt()
g = (ggplot(num_m,aes(x='value')) +
geom_bar(stat='bin', bins=bins) +
facet_wrap('~variable', scales='free') + #facetting by variable
theme_minimal() + theme(panel_spacing_x=0.5)
)
return g
def jointplot(df,col_x, col_y, no_outliers=True, kind='reg'): #'scatter','resid','reg','hex','kde','point'
if no_outliers:
new_df = remove_outliers(df,col_x)
new_df = remove_outliers(new_df,col_y)
else:
new_df = df.copy()
return sns.jointplot(new_df[col_x],new_df[col_y],kind=kind)
def qq_plot(df,col):
qq, reg = calc_qq(df,col)
g = (ggplot(qq,aes(x='theoretical_q',y='sample_q')) +
geom_point() + #plot points
geom_abline(slope=1,intercept=0,color='red') + #perfectly normal line
stat_function(fun=lambda x: x*reg.coef_[0][0]) + #linear estimation
ggtitle(f'y= {np.round(reg.coef_[0][0],2)} * x')+ #display equation
labs(x='Theoretical Quantiles (normalized)', y='Sample Qunatiles (normalized)'))
return g
def boxcox_plot(df, col, ls = [i/10 for i in range(-30,31,5)]):
lamb_df = boxcox_lamb_df(df[col],ls)
g = (ggplot(lamb_df, aes(x='lamb',y='coef',group=1)) +
geom_point() + geom_line())
return g
#transformation
def calc_qq(df,col):
sample_qs = [(np.percentile(df[col],i)-np.mean(df[col]))/np.std(df[col]) for i in range(5,100,5)]
theoretical_qs = [st.norm.ppf(i/100) for i in range(5,100,5)]
qq = pd.DataFrame({'sample_q':sample_qs,'theoretical_q':theoretical_qs})
reg = LinearRegression(fit_intercept=False).fit(np.array(qq['theoretical_q'])[:,None],
np.array(qq['sample_q'])[:,None])
return qq, reg
def boxcox(ser,lamb=0):
ser+= 1 - ser.min()
if lamb==0:
return np.log(ser)
else:
return (ser**lamb - 1)/lamb
def boxcox_lamb_df(ser, ls = [i/10 for i in range(-30,31,5)]):
coefs = []
for l in ls:
df = pd.DataFrame.from_dict({'val': boxcox(ser,l)})
qq, reg = calc_qq(df,'val')
coefs.append(reg.coef_.squeeze().item())
return pd.DataFrame({'lamb':ls,'coef':coefs})
def boxcox_lamb(ser, ls = [i/10 for i in range(-30,31,5)]):
df = boxcox_lamb_df(ser,ls)
return df.lamb[df.coef.idxmax()]
#formatters
class k_format(custom_format):
def k_ize(self,x):
#format one tick
return f'{int(x/1000)}k'
def __call__(self,x):
#format a list of ticks
return [self.k_ize(i) for i in x]
class m_format(custom_format):
def m_ize(self,x):
#format one tick
return f'{int(abs(x)/1e6)}M'
def __call__(self,x):
#format a list of ticks
return [self.m_ize(i) for i in x]
class prettynum_format(currency_format):
def __init__(self, prefix='', suffix='', digits=0, big_mark=','):
self.prefix = prefix
self.suffix = suffix
self.digits = digits
self.big_mark = big_mark