|
1
|
|
|
''' |
|
2
|
|
|
Functions for descriptive analytics. |
|
3
|
|
|
|
|
4
|
|
|
:author: Andreas Kanz |
|
5
|
|
|
|
|
6
|
|
|
''' |
|
7
|
|
|
|
|
8
|
|
|
# Imports |
|
9
|
|
|
import matplotlib.pyplot as plt |
|
10
|
|
|
import matplotlib.ticker as ticker |
|
11
|
|
|
import numpy as np |
|
12
|
|
|
import pandas as pd |
|
13
|
|
|
import scipy |
|
14
|
|
|
import seaborn as sns |
|
15
|
|
|
|
|
16
|
|
|
from .clean import drop_missing |
|
17
|
|
|
from .utils import _corr_selector |
|
18
|
|
|
from .utils import _missing_vals |
|
19
|
|
|
from .utils import _validate_input_0_1 |
|
20
|
|
|
from .utils import _validate_input_bool |
|
21
|
|
|
|
|
22
|
|
|
|
|
23
|
|
|
# Functions |
|
24
|
|
|
|
|
25
|
|
|
# Categorical Plot |
|
26
|
|
|
def cat_plot(data, figsize=(10, 14), top=3, bottom=3, bar_color_top='#5ab4ac', bar_color_bottom='#d8b365'): |
|
27
|
|
|
''' |
|
28
|
|
|
Parameters |
|
29
|
|
|
---------- |
|
30
|
|
|
|
|
31
|
|
|
data: 2D dataset that can be coerced into Pandas DataFrame. If a Pandas DataFrame is provided, the index/column \ |
|
32
|
|
|
information is used to label the plots. |
|
33
|
|
|
|
|
34
|
|
|
figsize: tuple, default (10, 14) |
|
35
|
|
|
Use to control the figure size. |
|
36
|
|
|
|
|
37
|
|
|
top: int, default 3 |
|
38
|
|
|
Show the "top" most frequent values in a column. |
|
39
|
|
|
|
|
40
|
|
|
bottom: int, default 3 |
|
41
|
|
|
Show the "bottom" most frequent values in a column. |
|
42
|
|
|
|
|
43
|
|
|
bar_color_top: color, default '#5ab4ac' |
|
44
|
|
|
Use to control the color of the bars indicating the most common values. |
|
45
|
|
|
|
|
46
|
|
|
bar_color_bottom: color, default '#d8b365' |
|
47
|
|
|
Use to control the color of the bars indicating the least common values. |
|
48
|
|
|
|
|
49
|
|
|
Returns |
|
50
|
|
|
------- |
|
51
|
|
|
figure |
|
52
|
|
|
|
|
53
|
|
|
''' |
|
54
|
|
|
|
|
55
|
|
|
data = pd.DataFrame(data).copy() |
|
56
|
|
|
cols = list(data.select_dtypes(exclude=['number']).columns) # categorical cols |
|
57
|
|
|
data = data[cols].applymap(str) |
|
58
|
|
|
|
|
59
|
|
|
if len(cols) == 0: |
|
60
|
|
|
print('No columns with categorical data were detected.') |
|
61
|
|
|
|
|
62
|
|
|
else: |
|
63
|
|
|
fig = plt.figure(figsize=figsize) |
|
64
|
|
|
gs = fig.add_gridspec(nrows=6, ncols=len(cols), wspace=0.2) |
|
65
|
|
|
|
|
66
|
|
|
for count, col in enumerate(cols): |
|
67
|
|
|
|
|
68
|
|
|
n_unique = data[col].nunique(dropna=False) |
|
69
|
|
|
|
|
70
|
|
|
if n_unique <= min(2, top+bottom): |
|
71
|
|
|
vals = int(n_unique//2) |
|
72
|
|
|
value_counts_top = data[col].value_counts(sort=True)[0:vals] |
|
73
|
|
|
value_counts_idx_top = list(map(str, data[col].value_counts()[0:vals].index.tolist())) |
|
74
|
|
|
value_counts_bot = data[col].value_counts(sort=True)[-vals:] |
|
75
|
|
|
value_counts_idx_bot = list(map(str, data[col].value_counts()[-vals:].index.tolist())) |
|
76
|
|
|
|
|
77
|
|
|
data[col][data[col].isin(value_counts_idx_top)] = 2 |
|
78
|
|
|
data[col][data[col].isin(value_counts_idx_bot)] = -2 |
|
79
|
|
|
data[col][~((data[col] == 2) | (data[col] == -2))] = 0 |
|
80
|
|
|
|
|
81
|
|
|
else: |
|
82
|
|
|
value_counts_top = data[col].value_counts(sort=True)[0:top] |
|
83
|
|
|
value_counts_idx_top = list(map(str, data[col].value_counts()[0:top].index.tolist())) |
|
84
|
|
|
if bottom == 0: |
|
85
|
|
|
value_counts_bot = [] |
|
86
|
|
|
value_counts_idx_bot = [] |
|
87
|
|
|
else: |
|
88
|
|
|
value_counts_bot = data[col].value_counts(sort=True)[-bottom:] |
|
89
|
|
|
value_counts_idx_bot = list(map(str, data[col].value_counts()[-bottom:].index.tolist())) |
|
90
|
|
|
|
|
91
|
|
|
data[col][data[col].isin(value_counts_idx_top)] = 2 |
|
92
|
|
|
data[col][data[col].isin(value_counts_idx_bot)] = -2 |
|
93
|
|
|
data[col][~((data[col] == 2) | (data[col] == -2))] = 0 |
|
94
|
|
|
|
|
95
|
|
|
# Barcharts |
|
96
|
|
|
ax_top = fig.add_subplot(gs[:1, count:count+1]) |
|
97
|
|
|
ax_top.bar(value_counts_idx_top, value_counts_top, color=bar_color_top, width=0.85) |
|
98
|
|
|
ax_top.bar(value_counts_idx_bot, value_counts_bot, color=bar_color_bottom, width=0.85) |
|
99
|
|
|
ax_top.set(frame_on=False) |
|
100
|
|
|
ax_top.tick_params(axis='x', labelrotation=90) |
|
101
|
|
|
|
|
102
|
|
|
# Summary stats |
|
103
|
|
|
ax_bottom = fig.add_subplot(gs[1:2, count:count+1]) |
|
104
|
|
|
ax_bottom.get_yaxis().set_visible(False) |
|
105
|
|
|
ax_bottom.get_xaxis().set_visible(False) |
|
106
|
|
|
ax_bottom.set(frame_on=False) |
|
107
|
|
|
ax_bottom.text(0, 0, f'Unique values: {n_unique}\n\n\ |
|
108
|
|
|
Top {top}vals: {sum(value_counts_top)} ({sum(value_counts_top)/data.shape[0]*100:.1f}%)\n\ |
|
109
|
|
|
Bottom {bottom} vals: {sum(value_counts_bot)} ({sum(value_counts_bot)/data.shape[0]*100:.1f}%)', |
|
110
|
|
|
transform=ax_bottom.transAxes, color='#111111', fontsize=11) |
|
111
|
|
|
|
|
112
|
|
|
data = data.astype('int') |
|
113
|
|
|
|
|
114
|
|
|
# Heatmap |
|
115
|
|
|
ax_hm = fig.add_subplot(gs[2:, :]) |
|
116
|
|
|
sns.heatmap(data, cmap='BrBG', cbar=False, vmin=-4.25, vmax=4.25, ax=ax_hm) |
|
117
|
|
|
ax_hm.set_yticks(np.round(ax_hm.get_yticks()[0::5], -1)) |
|
118
|
|
|
ax_hm.set_yticklabels(ax_hm.get_yticks()) |
|
119
|
|
|
ax_hm.set_xticklabels(ax_hm.get_xticklabels(), |
|
120
|
|
|
horizontalalignment='center', |
|
121
|
|
|
fontweight='light', |
|
122
|
|
|
fontsize='medium') |
|
123
|
|
|
ax_hm.tick_params(length=1, colors='#111111') |
|
124
|
|
|
|
|
125
|
|
|
gs.figure.suptitle('Categorical data plot', x=0.47, y=0.925, fontsize=18, color='#111111') |
|
126
|
|
|
|
|
127
|
|
|
return fig |
|
128
|
|
|
|
|
129
|
|
|
|
|
130
|
|
|
# Correlation Matrix |
|
131
|
|
|
def corr_mat(data, split=None, threshold=0, method='pearson'): |
|
132
|
|
|
''' |
|
133
|
|
|
Parameters |
|
134
|
|
|
---------- |
|
135
|
|
|
|
|
136
|
|
|
data: 2D dataset that can be coerced into Pandas DataFrame. If a Pandas DataFrame is provided, the index/column \ |
|
137
|
|
|
information is used to label the plots. |
|
138
|
|
|
|
|
139
|
|
|
split: {None, 'pos', 'neg', 'high', 'low'}, default None |
|
140
|
|
|
Type of split to be performed. |
|
141
|
|
|
|
|
142
|
|
|
threshold: float, default 0 |
|
143
|
|
|
Value between 0 <= threshold <= 1 |
|
144
|
|
|
|
|
145
|
|
|
method: {'pearson', 'spearman', 'kendall'}, default 'pearson' |
|
146
|
|
|
* pearson: measures linear relationships and requires normally distributed and homoscedastic data. |
|
147
|
|
|
* spearman: ranked/ordinal correlation, measures monotonic relationships. |
|
148
|
|
|
* kendall: ranked/ordinal correlation, measures monotonic relationships. Computationally more expensive but |
|
149
|
|
|
more robus in smaller dataets than 'spearman'. |
|
150
|
|
|
|
|
151
|
|
|
Returns |
|
152
|
|
|
------- |
|
153
|
|
|
Pandas Styler object |
|
154
|
|
|
|
|
155
|
|
|
''' |
|
156
|
|
|
|
|
157
|
|
|
# Validate Inputs |
|
158
|
|
|
_validate_input_0_1(threshold, 'threshold') |
|
159
|
|
|
|
|
160
|
|
|
def color_negative_red(val): |
|
161
|
|
|
color = '#FF3344' if val < 0 else None |
|
162
|
|
|
return 'color: %s' % color |
|
163
|
|
|
|
|
164
|
|
|
data = pd.DataFrame(data) |
|
165
|
|
|
corr = data.corr(method=method) |
|
166
|
|
|
|
|
167
|
|
|
corr = _corr_selector(corr, split=split, threshold=threshold) |
|
168
|
|
|
|
|
169
|
|
|
return corr.style.applymap(color_negative_red).format("{:.2f}", na_rep='-') |
|
170
|
|
|
|
|
171
|
|
|
|
|
172
|
|
|
# Correlation matrix / heatmap |
|
173
|
|
|
def corr_plot(data, split=None, threshold=0, target=None, method='pearson', cmap='BrBG', figsize=(12, 10), annot=True, |
|
174
|
|
|
dev=False, **kwargs): |
|
175
|
|
|
''' |
|
176
|
|
|
Two-dimensional visualization of the correlation between feature-columns, excluding NA values. |
|
177
|
|
|
|
|
178
|
|
|
Parameters |
|
179
|
|
|
---------- |
|
180
|
|
|
data: 2D dataset that can be coerced into Pandas DataFrame. If a Pandas DataFrame is provided, the index/column \ |
|
181
|
|
|
information is used to label the plots. |
|
182
|
|
|
|
|
183
|
|
|
split: {None, 'pos', 'neg', 'high', 'low'}, default None |
|
184
|
|
|
Type of split to be performed. |
|
185
|
|
|
|
|
186
|
|
|
* None: visualize all correlations between the feature-columns. |
|
187
|
|
|
* pos: visualize all positive correlations between the feature-columns above the threshold. |
|
188
|
|
|
* neg: visualize all negative correlations between the feature-columns below the threshold. |
|
189
|
|
|
* high: visualize all correlations between the feature-columns for which abs(corr) > threshold is True. |
|
190
|
|
|
* low: visualize all correlations between the feature-columns for which abs(corr) < threshold is True. |
|
191
|
|
|
|
|
192
|
|
|
threshold: float, default 0 |
|
193
|
|
|
Value between 0 <= threshold <= 1 |
|
194
|
|
|
|
|
195
|
|
|
target: string, list, np.array or pd.Series, default None |
|
196
|
|
|
Specify target for correlation. E.g. label column to generate only the correlations between each feature\ |
|
197
|
|
|
and the label. |
|
198
|
|
|
|
|
199
|
|
|
method: {'pearson', 'spearman', 'kendall'}, default 'pearson' |
|
200
|
|
|
* pearson: measures linear relationships and requires normally distributed and homoscedastic data. |
|
201
|
|
|
* spearman: ranked/ordinal correlation, measures monotonic relationships. |
|
202
|
|
|
* kendall: ranked/ordinal correlation, measures monotonic relationships. Computationally more expensive but |
|
203
|
|
|
more robust in smaller dataets than 'spearman'. |
|
204
|
|
|
|
|
205
|
|
|
cmap: matplotlib colormap name or object, or list of colors, default 'BrBG' |
|
206
|
|
|
The mapping from data values to color space. |
|
207
|
|
|
|
|
208
|
|
|
figsize: tuple, default (12, 10) |
|
209
|
|
|
Use to control the figure size. |
|
210
|
|
|
|
|
211
|
|
|
annot: bool, default True |
|
212
|
|
|
Use to show or hide annotations. |
|
213
|
|
|
|
|
214
|
|
|
dev: bool, default False |
|
215
|
|
|
Display figure settings in the plot by setting dev = True. If False, the settings are not displayed.s |
|
216
|
|
|
|
|
217
|
|
|
**kwargs: optional |
|
218
|
|
|
Additional elements to control the visualization of the plot, e.g.: |
|
219
|
|
|
|
|
220
|
|
|
* mask: bool, default True |
|
221
|
|
|
If set to False the entire correlation matrix, including the upper triangle is shown. Set dev = False in this \ |
|
222
|
|
|
case to avoid overlap. |
|
223
|
|
|
* vmax: float, default is calculated from the given correlation coefficients. |
|
224
|
|
|
Value between -1 or vmin <= vmax <= 1, limits the range of the colorbar. |
|
225
|
|
|
* vmin: float, default is calculated from the given correlation coefficients. |
|
226
|
|
|
Value between -1 <= vmin <= 1 or vmax, limits the range of the colorbar. |
|
227
|
|
|
* linewidths: float, default 0.5 |
|
228
|
|
|
Controls the line-width inbetween the squares. |
|
229
|
|
|
* annot_kws: dict, default {'size' : 10} |
|
230
|
|
|
Controls the font size of the annotations. Only available when annot = True. |
|
231
|
|
|
* cbar_kws: dict, default {'shrink': .95, 'aspect': 30} |
|
232
|
|
|
Controls the size of the colorbar. |
|
233
|
|
|
* Many more kwargs are available, i.e. 'alpha' to control blending, or options to adjust labels, ticks ... |
|
234
|
|
|
|
|
235
|
|
|
Kwargs can be supplied through a dictionary of key-value pairs (see above). |
|
236
|
|
|
|
|
237
|
|
|
Returns |
|
238
|
|
|
------- |
|
239
|
|
|
ax: matplotlib Axes |
|
240
|
|
|
Returns the Axes object with the plot for further tweaking. |
|
241
|
|
|
|
|
242
|
|
|
''' |
|
243
|
|
|
|
|
244
|
|
|
# Validate Inputs |
|
245
|
|
|
_validate_input_0_1(threshold, 'threshold') |
|
246
|
|
|
_validate_input_bool(annot, 'annot') |
|
247
|
|
|
_validate_input_bool(dev, 'dev') |
|
248
|
|
|
|
|
249
|
|
|
data = pd.DataFrame(data) |
|
250
|
|
|
|
|
251
|
|
|
# Obtain correlations |
|
252
|
|
|
if isinstance(target, (str, list, pd.Series, np.ndarray)): |
|
253
|
|
|
target_data = [] |
|
254
|
|
|
if isinstance(target, str): |
|
255
|
|
|
target_data = data[target] |
|
256
|
|
|
data = data.drop(target, axis=1) |
|
257
|
|
|
|
|
258
|
|
|
elif isinstance(target, (list, pd.Series, np.ndarray)): |
|
259
|
|
|
target_data = pd.Series(target) |
|
260
|
|
|
|
|
261
|
|
|
corr = pd.DataFrame(data.corrwith(target_data)) |
|
262
|
|
|
corr.rename_axis(target, axis=1, inplace=True) |
|
263
|
|
|
corr = _corr_selector(corr, split=split, threshold=threshold) |
|
264
|
|
|
corr = corr.sort_values(corr.columns[0], ascending=False) |
|
265
|
|
|
vmax = np.round(np.nanmax(corr)-0.05, 2) |
|
266
|
|
|
vmin = np.round(np.nanmin(corr)+0.05, 2) |
|
267
|
|
|
mask = False |
|
268
|
|
|
square = False |
|
269
|
|
|
|
|
270
|
|
|
else: |
|
271
|
|
|
corr = corr_mat(data, split=split, threshold=threshold, method=method).data |
|
272
|
|
|
|
|
273
|
|
|
mask = np.triu(np.ones_like(corr, dtype=np.bool)) # Generate mask for the upper triangle |
|
274
|
|
|
square = True |
|
275
|
|
|
|
|
276
|
|
|
vmax = np.round(np.nanmax(corr.where(~mask))-0.05, 2) |
|
277
|
|
|
vmin = np.round(np.nanmin(corr.where(~mask))+0.05, 2) |
|
278
|
|
|
|
|
279
|
|
|
fig, ax = plt.subplots(figsize=figsize) |
|
280
|
|
|
|
|
281
|
|
|
# Specify kwargs for the heatmap |
|
282
|
|
|
kwargs = {'mask': mask, |
|
283
|
|
|
'cmap': cmap, |
|
284
|
|
|
'annot': annot, |
|
285
|
|
|
'vmax': vmax, |
|
286
|
|
|
'vmin': vmin, |
|
287
|
|
|
'linewidths': .5, |
|
288
|
|
|
'annot_kws': {'size': 10}, |
|
289
|
|
|
'cbar_kws': {'shrink': .95, 'aspect': 30}, |
|
290
|
|
|
**kwargs} |
|
291
|
|
|
|
|
292
|
|
|
# Draw heatmap with mask and some default settings |
|
293
|
|
|
sns.heatmap(corr, |
|
294
|
|
|
center=0, |
|
295
|
|
|
square=square, |
|
296
|
|
|
fmt='.2f', |
|
297
|
|
|
**kwargs |
|
298
|
|
|
) |
|
299
|
|
|
|
|
300
|
|
|
ax.set_title(f'Feature-correlation ({method})', fontdict={'fontsize': 18}) |
|
301
|
|
|
|
|
302
|
|
|
# Display settings |
|
303
|
|
|
if dev: |
|
304
|
|
|
fig.suptitle(f"\ |
|
305
|
|
|
Settings (dev-mode): \n\ |
|
306
|
|
|
- split-mode: {split} \n\ |
|
307
|
|
|
- threshold: {threshold} \n\ |
|
308
|
|
|
- method: {method} \n\ |
|
309
|
|
|
- annotations: {annot} \n\ |
|
310
|
|
|
- cbar: \n\ |
|
311
|
|
|
- vmax: {vmax} \n\ |
|
312
|
|
|
- vmin: {vmin} \n\ |
|
313
|
|
|
- linewidths: {kwargs['linewidths']} \n\ |
|
314
|
|
|
- annot_kws: {kwargs['annot_kws']} \n\ |
|
315
|
|
|
- cbar_kws: {kwargs['cbar_kws']}", |
|
316
|
|
|
fontsize=12, |
|
317
|
|
|
color='gray', |
|
318
|
|
|
x=0.35, |
|
319
|
|
|
y=0.85, |
|
320
|
|
|
ha='left') |
|
321
|
|
|
|
|
322
|
|
|
return ax |
|
323
|
|
|
|
|
324
|
|
|
|
|
325
|
|
|
# Distribution plot |
|
326
|
|
|
def dist_plot(data, mean_color='orange', figsize=(14, 2), fill_range=(0.025, 0.975), hist=False, bins=None, |
|
327
|
|
|
showall=False, kde_kws=None, rug_kws=None, fill_kws=None, font_kws=None): |
|
328
|
|
|
''' |
|
329
|
|
|
Two-dimensional visualization of the distribution of numerical features. |
|
330
|
|
|
|
|
331
|
|
|
Parameters |
|
332
|
|
|
---------- |
|
333
|
|
|
data: 2D dataset that can be coerced into Pandas DataFrame. If a Pandas DataFrame is provided, the index/column \ |
|
334
|
|
|
information is used to label the plots. |
|
335
|
|
|
|
|
336
|
|
|
mean_color: color, default 'orange' |
|
337
|
|
|
Color of the vertical line indicating the mean of the data. |
|
338
|
|
|
|
|
339
|
|
|
figsize: tuple, default (14, 2) |
|
340
|
|
|
Use to control the figure size. |
|
341
|
|
|
|
|
342
|
|
|
fill_range: tuple, default (0.025, 0.975) |
|
343
|
|
|
Use to control set the quantiles for shading. Default spans 95% of the data, which is about two std. deviations\ |
|
344
|
|
|
above and below the mean. |
|
345
|
|
|
|
|
346
|
|
|
hist: bool, default False |
|
347
|
|
|
Set to True to display histogram bars in the plot. |
|
348
|
|
|
|
|
349
|
|
|
bins: integer, default None |
|
350
|
|
|
Specification of the number of hist bins. Requires hist = True |
|
351
|
|
|
|
|
352
|
|
|
showall: bool, default False |
|
353
|
|
|
Set to True to remove the output limit of 20 plots. |
|
354
|
|
|
|
|
355
|
|
|
kdw_kws: dict, default None |
|
356
|
|
|
Keyword arguments for kdeplot(). |
|
357
|
|
|
|
|
358
|
|
|
rug_kws: dict, default None |
|
359
|
|
|
Keyword arguments for rugplot(). |
|
360
|
|
|
|
|
361
|
|
|
fill_kws: dict, default None |
|
362
|
|
|
Keyword arguments to control the fill. |
|
363
|
|
|
|
|
364
|
|
|
font_kws: dict, default None |
|
365
|
|
|
Keyword arguments to control the font. |
|
366
|
|
|
|
|
367
|
|
|
Returns |
|
368
|
|
|
------- |
|
369
|
|
|
ax: matplotlib Axes |
|
370
|
|
|
Returns the Axes object with the plot for further tweaking. |
|
371
|
|
|
|
|
372
|
|
|
''' |
|
373
|
|
|
|
|
374
|
|
|
# Validate Inputs |
|
375
|
|
|
_validate_input_bool(hist, 'hist') |
|
376
|
|
|
_validate_input_bool(showall, 'showall') |
|
377
|
|
|
|
|
378
|
|
|
# Handle dictionary defaults |
|
379
|
|
|
kde_kws = {} if kde_kws is None else kde_kws.copy() |
|
380
|
|
|
rug_kws = {} if rug_kws is None else rug_kws.copy() |
|
381
|
|
|
fill_kws = {} if fill_kws is None else fill_kws.copy() |
|
382
|
|
|
font_kws = {} if font_kws is None else font_kws.copy() |
|
383
|
|
|
|
|
384
|
|
|
data = drop_missing(pd.DataFrame(data).copy()) # drop empty columns and rows |
|
385
|
|
|
cols = list(data.select_dtypes(include=['number']).columns) # numeric cols |
|
386
|
|
|
data = data[cols] |
|
387
|
|
|
|
|
388
|
|
|
# Default settings |
|
389
|
|
|
kde_kws = {'color': 'k', 'alpha': 0.7, 'linewidth': 1, **kde_kws} |
|
390
|
|
|
rug_kws = {'color': 'brown', 'alpha': 0.5, 'linewidth': 2, 'height': 0.04, **rug_kws} |
|
391
|
|
|
fill_kws = {'color': 'brown', 'alpha': 0.1, **fill_kws} |
|
392
|
|
|
font_kws = {'color': '#111111', 'weight': 'normal', 'size': 11, **font_kws} |
|
393
|
|
|
|
|
394
|
|
|
if len(cols) == 0: |
|
395
|
|
|
print('No columns with numeric data were detected.') |
|
396
|
|
|
ax = None |
|
397
|
|
|
|
|
398
|
|
|
else: |
|
399
|
|
|
if len(cols) >= 20 and showall is False: |
|
400
|
|
|
print(f'Note: The number of numerical features is very large ({len(cols)}), please consider splitting the data.\ |
|
401
|
|
|
Showing plots for the first 20 numerical features. Override this by setting showall=True.') |
|
402
|
|
|
cols = cols[:20] |
|
403
|
|
|
|
|
404
|
|
|
ax = [] |
|
405
|
|
|
for col in cols: |
|
406
|
|
|
# Drop missing values |
|
407
|
|
|
dropped_values = data[col].isna().sum() |
|
408
|
|
|
if dropped_values > 0: |
|
409
|
|
|
print(f'Dropped {dropped_values} missing values from column {col}.') |
|
410
|
|
|
col_data = data[col].dropna(axis=0) |
|
411
|
|
|
else: |
|
412
|
|
|
col_data = data[col] |
|
413
|
|
|
|
|
414
|
|
|
_, ax = plt.subplots(figsize=figsize) |
|
415
|
|
|
ax = sns.distplot(col_data, bins=bins, hist=hist, rug=True, kde_kws=kde_kws, |
|
416
|
|
|
rug_kws=rug_kws, hist_kws={'alpha': 0.5, 'histtype': 'step'}) |
|
417
|
|
|
|
|
418
|
|
|
# Vertical lines and fill |
|
419
|
|
|
line = ax.lines[0] |
|
420
|
|
|
x = line.get_xydata()[:, 0] |
|
421
|
|
|
y = line.get_xydata()[:, 1] |
|
422
|
|
|
ax.fill_between(x, y, |
|
423
|
|
|
where=( |
|
424
|
|
|
(x >= np.quantile(col_data, fill_range[0])) & |
|
425
|
|
|
(x <= np.quantile(col_data, fill_range[1]))), |
|
426
|
|
|
label=f'{fill_range[0]*100:.0f}% - {fill_range[1]*100:.0f}%', |
|
427
|
|
|
**fill_kws) |
|
428
|
|
|
|
|
429
|
|
|
ax.vlines(x=np.mean(col_data), |
|
430
|
|
|
ymin=0, |
|
431
|
|
|
ymax=np.interp(np.mean(col_data), x, y), |
|
432
|
|
|
ls='dotted', color=mean_color, lw=2, label='mean') |
|
433
|
|
|
ax.vlines(x=np.median(col_data), |
|
434
|
|
|
ymin=0, |
|
435
|
|
|
ymax=np.interp(np.median(col_data), x, y), |
|
436
|
|
|
ls=':', color='.3', label='median') |
|
437
|
|
|
ax.vlines(x=np.quantile(col_data, 0.25), |
|
438
|
|
|
ymin=0, |
|
439
|
|
|
ymax=np.interp(np.quantile(col_data, 0.25), x, y), ls=':', color='.5', label='25%') |
|
440
|
|
|
ax.vlines(x=np.quantile(col_data, 0.75), |
|
441
|
|
|
ymin=0, |
|
442
|
|
|
ymax=np.interp(np.quantile(col_data, 0.75), x, y), ls=':', color='.5', label='75%') |
|
443
|
|
|
|
|
444
|
|
|
ax.set_ylim(0,) |
|
445
|
|
|
ax.set_xlim(ax.get_xlim()[0]*1.1, ax.get_xlim()[1]*1.1) |
|
446
|
|
|
|
|
447
|
|
|
# Annotations and legend |
|
448
|
|
|
ax.text(0.01, 0.85, f'Mean: {np.round(np.mean(col_data),2)}', |
|
449
|
|
|
fontdict=font_kws, transform=ax.transAxes) |
|
450
|
|
|
ax.text(0.01, 0.7, f'Std. dev: {np.round(scipy.stats.tstd(col_data),2)}', |
|
451
|
|
|
fontdict=font_kws, transform=ax.transAxes) |
|
452
|
|
|
ax.text(0.01, 0.55, f'Skew: {np.round(scipy.stats.skew(col_data),2)}', |
|
453
|
|
|
fontdict=font_kws, transform=ax.transAxes) |
|
454
|
|
|
ax.text(0.01, 0.4, f'Kurtosis: {np.round(scipy.stats.kurtosis(col_data),2)}', # Excess Kurtosis |
|
455
|
|
|
fontdict=font_kws, transform=ax.transAxes) |
|
456
|
|
|
ax.text(0.01, 0.25, f'Count: {np.round(len(col_data))}', |
|
457
|
|
|
fontdict=font_kws, transform=ax.transAxes) |
|
458
|
|
|
ax.legend(loc='upper right') |
|
459
|
|
|
|
|
460
|
|
|
return ax |
|
461
|
|
|
|
|
462
|
|
|
|
|
463
|
|
|
# Missing value plot |
|
464
|
|
|
def missingval_plot(data, cmap='PuBuGn', figsize=(12, 12), sort=False, spine_color='#EEEEEE'): |
|
465
|
|
|
''' |
|
466
|
|
|
Two-dimensional visualization of the missing values in a dataset. |
|
467
|
|
|
|
|
468
|
|
|
Parameters |
|
469
|
|
|
---------- |
|
470
|
|
|
data: 2D dataset that can be coerced into Pandas DataFrame. If a Pandas DataFrame is provided, the index/column \ |
|
471
|
|
|
information is used to label the plots. |
|
472
|
|
|
|
|
473
|
|
|
cmap: colormap, default 'PuBuGn' |
|
474
|
|
|
Any valid colormap can be used. E.g. 'Greys', 'RdPu'. More information can be found in the matplotlib \ |
|
475
|
|
|
documentation. |
|
476
|
|
|
|
|
477
|
|
|
figsize: tuple, default (20, 12) |
|
478
|
|
|
Use to control the figure size. |
|
479
|
|
|
|
|
480
|
|
|
sort: bool, default False |
|
481
|
|
|
Sort columns based on missing values in descending order and drop columns without any missing values |
|
482
|
|
|
|
|
483
|
|
|
spine_color: color, default '#EEEEEE' |
|
484
|
|
|
Set to 'None' to hide the spines on all plots or use any valid matplotlib color argument. |
|
485
|
|
|
|
|
486
|
|
|
Returns |
|
487
|
|
|
------- |
|
488
|
|
|
figure |
|
489
|
|
|
|
|
490
|
|
|
''' |
|
491
|
|
|
|
|
492
|
|
|
data = pd.DataFrame(data) |
|
493
|
|
|
|
|
494
|
|
|
if sort: |
|
495
|
|
|
mv_cols_sorted = data.isna().sum(axis=0).sort_values(ascending=False) |
|
496
|
|
|
final_cols = mv_cols_sorted.drop(mv_cols_sorted[mv_cols_sorted.values == 0].keys().tolist()).keys().tolist() |
|
497
|
|
|
data = data[final_cols] |
|
498
|
|
|
print('Displaying only columns with missing values.') |
|
499
|
|
|
|
|
500
|
|
|
# Identify missing values |
|
501
|
|
|
mv_cols = _missing_vals(data)['mv_cols'] # data.isna().sum(axis=0) |
|
502
|
|
|
mv_rows = _missing_vals(data)['mv_rows'] # data.isna().sum(axis=1) |
|
503
|
|
|
mv_total = _missing_vals(data)['mv_total'] |
|
504
|
|
|
mv_cols_ratio = _missing_vals(data)['mv_cols_ratio'] # mv_cols / data.shape[0] |
|
505
|
|
|
total_datapoints = data.shape[0]*data.shape[1] |
|
506
|
|
|
|
|
507
|
|
|
if mv_total == 0: |
|
508
|
|
|
print('No missing values found in the dataset.') |
|
509
|
|
|
else: |
|
510
|
|
|
# Create figure and axes |
|
511
|
|
|
fig = plt.figure(figsize=figsize) |
|
512
|
|
|
gs = fig.add_gridspec(nrows=6, ncols=6, left=0.05, wspace=0.05) |
|
513
|
|
|
ax1 = fig.add_subplot(gs[:1, :5]) |
|
514
|
|
|
ax2 = fig.add_subplot(gs[1:, :5]) |
|
515
|
|
|
ax3 = fig.add_subplot(gs[:1, 5:]) |
|
516
|
|
|
ax4 = fig.add_subplot(gs[1:, 5:]) |
|
517
|
|
|
|
|
518
|
|
|
# ax1 - Barplot |
|
519
|
|
|
colors = plt.get_cmap(cmap)(mv_cols / np.max(mv_cols)) # color bars by height |
|
520
|
|
|
ax1.bar(range(len(mv_cols)), np.round((mv_cols_ratio)*100, 2), color=colors) |
|
521
|
|
|
ax1.get_xaxis().set_visible(False) |
|
522
|
|
|
ax1.set(frame_on=False, xlim=(-.5, len(mv_cols)-0.5)) |
|
523
|
|
|
ax1.set_ylim(0, np.max(mv_cols_ratio)*100) |
|
524
|
|
|
ax1.grid(linestyle=':', linewidth=1) |
|
525
|
|
|
ax1.yaxis.set_major_formatter(ticker.PercentFormatter(decimals=0)) |
|
526
|
|
|
ax1.tick_params(axis='y', colors='#111111', length=1) |
|
527
|
|
|
|
|
528
|
|
|
# annotate values on top of the bars |
|
529
|
|
|
for rect, label in zip(ax1.patches, mv_cols): |
|
530
|
|
|
height = rect.get_height() |
|
531
|
|
|
ax1.text(.1 + rect.get_x() + rect.get_width() / 2, height+0.5, label, |
|
532
|
|
|
ha='center', |
|
533
|
|
|
va='bottom', |
|
534
|
|
|
rotation='90', |
|
535
|
|
|
alpha=0.5, |
|
536
|
|
|
fontsize='small') |
|
537
|
|
|
|
|
538
|
|
|
ax1.set_frame_on(True) |
|
539
|
|
|
for _, spine in ax1.spines.items(): |
|
540
|
|
|
spine.set_visible(True) |
|
541
|
|
|
spine.set_color(spine_color) |
|
542
|
|
|
ax1.spines['top'].set_color(None) |
|
543
|
|
|
|
|
544
|
|
|
# ax2 - Heatmap |
|
545
|
|
|
sns.heatmap(data.isna(), cbar=False, cmap='binary', ax=ax2) |
|
546
|
|
|
ax2.set_yticks(np.round(ax2.get_yticks()[0::5], -1)) |
|
547
|
|
|
ax2.set_yticklabels(ax2.get_yticks()) |
|
548
|
|
|
ax2.set_xticklabels( |
|
549
|
|
|
ax2.get_xticklabels(), |
|
550
|
|
|
horizontalalignment='center', |
|
551
|
|
|
fontweight='light', |
|
552
|
|
|
fontsize='medium') |
|
553
|
|
|
ax2.tick_params(length=1, colors='#111111') |
|
554
|
|
|
for _, spine in ax2.spines.items(): |
|
555
|
|
|
spine.set_visible(True) |
|
556
|
|
|
spine.set_color(spine_color) |
|
557
|
|
|
|
|
558
|
|
|
# ax3 - Summary |
|
559
|
|
|
fontax3 = {'color': '#111111', |
|
560
|
|
|
'weight': 'normal', |
|
561
|
|
|
'size': 12, |
|
562
|
|
|
} |
|
563
|
|
|
ax3.get_xaxis().set_visible(False) |
|
564
|
|
|
ax3.get_yaxis().set_visible(False) |
|
565
|
|
|
ax3.set(frame_on=False) |
|
566
|
|
|
|
|
567
|
|
|
ax3.text(0.1, 0.9, f"Total: {np.round(total_datapoints/1000,1)}K", |
|
568
|
|
|
transform=ax3.transAxes, |
|
569
|
|
|
fontdict=fontax3) |
|
570
|
|
|
ax3.text(0.1, 0.7, f"Missing: {np.round(mv_total/1000,1)}K", |
|
571
|
|
|
transform=ax3.transAxes, |
|
572
|
|
|
fontdict=fontax3) |
|
573
|
|
|
ax3.text(0.1, 0.5, f"Relative: {np.round(mv_total/total_datapoints*100,1)}%", |
|
574
|
|
|
transform=ax3.transAxes, |
|
575
|
|
|
fontdict=fontax3) |
|
576
|
|
|
ax3.text(0.1, 0.3, f"Max-col: {np.round(mv_cols.max()/data.shape[0]*100)}%", |
|
577
|
|
|
transform=ax3.transAxes, |
|
578
|
|
|
fontdict=fontax3) |
|
579
|
|
|
ax3.text(0.1, 0.1, f"Max-row: {np.round(mv_rows.max()/data.shape[1]*100)}%", |
|
580
|
|
|
transform=ax3.transAxes, |
|
581
|
|
|
fontdict=fontax3) |
|
582
|
|
|
|
|
583
|
|
|
# ax4 - Scatter plot |
|
584
|
|
|
ax4.get_yaxis().set_visible(False) |
|
585
|
|
|
for _, spine in ax4.spines.items(): |
|
586
|
|
|
spine.set_color(spine_color) |
|
587
|
|
|
ax4.tick_params(axis='x', colors='#111111', length=1) |
|
588
|
|
|
|
|
589
|
|
|
ax4.scatter(mv_rows, range(len(mv_rows)), s=mv_rows, c=mv_rows, cmap=cmap, marker=".", vmin=1) |
|
590
|
|
|
ax4.set_ylim((0, len(mv_rows))[::-1]) # limit and invert y-axis |
|
591
|
|
|
ax4.set_xlim(0, max(mv_rows)+0.5) |
|
592
|
|
|
ax4.grid(linestyle=':', linewidth=1) |
|
593
|
|
|
|
|
594
|
|
|
gs.figure.suptitle('Missing value plot', x=0.45, y=0.94, fontsize=18, color='#111111') |
|
595
|
|
|
|
|
596
|
|
|
return fig |
|
597
|
|
|
|
akanz1 /
klib
