|
1
|
|
|
''' |
|
2
|
|
|
Utilities 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 seaborn as sns |
|
14
|
|
|
|
|
15
|
|
|
|
|
16
|
|
|
# _functions |
|
17
|
|
|
|
|
18
|
|
|
def _memory_usage(data): |
|
19
|
|
|
''' |
|
20
|
|
|
Gives the total memory usage in kilobytes. |
|
21
|
|
|
|
|
22
|
|
|
Parameters |
|
23
|
|
|
---------- |
|
24
|
|
|
data: 2D dataset that can be coerced into Pandas DataFrame. If a Pandas DataFrame is provided, the index/column \ |
|
25
|
|
|
information is used to label the plots. |
|
26
|
|
|
|
|
27
|
|
|
Returns |
|
28
|
|
|
------- |
|
29
|
|
|
memory_usage: float |
|
30
|
|
|
|
|
31
|
|
|
''' |
|
32
|
|
|
|
|
33
|
|
|
data = pd.DataFrame(data) |
|
34
|
|
|
memory_usage = round(data.memory_usage(index=True, deep=True).sum()/1024, 2) |
|
35
|
|
|
|
|
36
|
|
|
return memory_usage |
|
37
|
|
|
|
|
38
|
|
|
|
|
39
|
|
|
def _missing_vals(data): |
|
40
|
|
|
''' |
|
41
|
|
|
Gives metrics of missing values in the dataset. |
|
42
|
|
|
|
|
43
|
|
|
Parameters |
|
44
|
|
|
---------- |
|
45
|
|
|
data: 2D dataset that can be coerced into Pandas DataFrame. If a Pandas DataFrame is provided, the index/column \ |
|
46
|
|
|
information is used to label the plots. |
|
47
|
|
|
|
|
48
|
|
|
Returns |
|
49
|
|
|
------- |
|
50
|
|
|
mv_total: float, number of missing values in the entire dataset |
|
51
|
|
|
mv_rows: float, number of missing values in each row |
|
52
|
|
|
mv_cols: float, number of missing values in each column |
|
53
|
|
|
mv_rows_ratio: float, ratio of missing values for each row |
|
54
|
|
|
mv_cols_ratio: float, ratio of missing values for each column |
|
55
|
|
|
''' |
|
56
|
|
|
|
|
57
|
|
|
data = pd.DataFrame(data) |
|
58
|
|
|
mv_rows = data.isna().sum(axis=1) |
|
59
|
|
|
mv_cols = data.isna().sum(axis=0) |
|
60
|
|
|
mv_total = data.isna().sum().sum() |
|
61
|
|
|
mv_rows_ratio = mv_rows/data.shape[1] |
|
62
|
|
|
mv_cols_ratio = mv_cols/data.shape[0] |
|
63
|
|
|
|
|
64
|
|
|
return {'mv_total': mv_total, |
|
65
|
|
|
'mv_rows': mv_rows, |
|
66
|
|
|
'mv_cols': mv_cols, |
|
67
|
|
|
'mv_rows_ratio': mv_rows_ratio, |
|
68
|
|
|
'mv_cols_ratio': mv_cols_ratio} |
|
69
|
|
|
|
|
70
|
|
|
|
|
71
|
|
|
# Functions |
|
72
|
|
|
|
|
73
|
|
|
# Correlation Matrix |
|
74
|
|
|
def corr_mat(data, split=None, threshold=0): |
|
75
|
|
|
def color_negative_red(val): |
|
76
|
|
|
color = '#FF3344' if val < 0 else None |
|
77
|
|
|
return 'color: %s' % color |
|
78
|
|
|
|
|
79
|
|
|
data = pd.DataFrame(data) |
|
80
|
|
|
|
|
81
|
|
|
if split == 'pos': |
|
82
|
|
|
corr = data.corr().where((data.corr() >= threshold) & (data.corr() > 0)) |
|
83
|
|
|
print('Displaying positive correlations. Use "threshold" to further limit the results.') |
|
84
|
|
|
elif split == 'neg': |
|
85
|
|
|
corr = data.corr().where((data.corr() <= threshold) & (data.corr() < 0)) |
|
86
|
|
|
print('Displaying negative correlations. Use "threshold" to further limit the results.') |
|
87
|
|
|
elif split == 'high': |
|
88
|
|
|
corr = data.corr().where(np.abs(data.corr()) >= threshold) |
|
89
|
|
|
print('Displaying absolute correlations above a chosen threshold.') |
|
90
|
|
|
elif split == 'low': |
|
91
|
|
|
corr = data.corr().where(np.abs(data.corr()) <= threshold) |
|
92
|
|
|
print('Displaying absolute correlations below a chosen threshold.') |
|
93
|
|
|
else: |
|
94
|
|
|
corr = data.corr() |
|
95
|
|
|
split = 'None' |
|
96
|
|
|
threshold = 'None' |
|
97
|
|
|
|
|
98
|
|
|
return corr.style.applymap(color_negative_red).format("{:.2f}", na_rep='-') |
|
99
|
|
|
|
|
100
|
|
|
|
|
101
|
|
|
# Missing value plot |
|
102
|
|
|
def missingval_plot(data, cmap='PuBuGn', figsize=(20, 12), sort=False, spine_color='#EEEEEE'): |
|
103
|
|
|
''' |
|
104
|
|
|
Two-dimensional visualization of the missing values in a dataset. |
|
105
|
|
|
|
|
106
|
|
|
Parameters |
|
107
|
|
|
---------- |
|
108
|
|
|
data: 2D dataset that can be coerced into Pandas DataFrame. If a Pandas DataFrame is provided, the index/column \ |
|
109
|
|
|
information is used to label the plots. |
|
110
|
|
|
|
|
111
|
|
|
cmap: colormap, default 'PuBuGn' |
|
112
|
|
|
Any valid colormap can be used. E.g. 'Greys', 'RdPu'. More information can be found in the matplotlib \ |
|
113
|
|
|
documentation. |
|
114
|
|
|
|
|
115
|
|
|
figsize: tuple, default (20,12) |
|
116
|
|
|
Use to control the figure size. |
|
117
|
|
|
|
|
118
|
|
|
sort: bool, default False |
|
119
|
|
|
Sort columns based on missing values in descending order and drop columns without any missing values |
|
120
|
|
|
|
|
121
|
|
|
spine_color: color-code, default '#EEEEEE' |
|
122
|
|
|
Set to 'None' to hide the spines on all plots or use any valid matplotlib color argument. |
|
123
|
|
|
|
|
124
|
|
|
Returns |
|
125
|
|
|
------- |
|
126
|
|
|
figure |
|
127
|
|
|
''' |
|
128
|
|
|
|
|
129
|
|
|
data = pd.DataFrame(data) |
|
130
|
|
|
|
|
131
|
|
|
if sort: |
|
132
|
|
|
mv_cols_sorted = data.isna().sum(axis=0).sort_values(ascending=False) |
|
133
|
|
|
final_cols = mv_cols_sorted.drop(mv_cols_sorted[mv_cols_sorted.values == 0].keys().tolist()).keys().tolist() |
|
134
|
|
|
data = data[final_cols] |
|
135
|
|
|
print('Displaying only columns with missing values.') |
|
136
|
|
|
|
|
137
|
|
|
# Identify missing values |
|
138
|
|
|
mv_cols = _missing_vals(data)['mv_cols'] # data.isna().sum(axis=0) |
|
139
|
|
|
mv_rows = _missing_vals(data)['mv_rows'] # data.isna().sum(axis=1) |
|
140
|
|
|
mv_total = _missing_vals(data)['mv_total'] |
|
141
|
|
|
mv_cols_ratio = _missing_vals(data)['mv_cols_ratio'] # mv_cols / data.shape[0] |
|
142
|
|
|
total_datapoints = data.shape[0]*data.shape[1] |
|
143
|
|
|
|
|
144
|
|
|
if mv_total == 0: |
|
145
|
|
|
print('No missing values found in the dataset.') |
|
146
|
|
|
else: |
|
147
|
|
|
# Create figure and axes |
|
148
|
|
|
fig = plt.figure(figsize=figsize) |
|
149
|
|
|
grid = fig.add_gridspec(nrows=6, ncols=6, left=0.05, right=0.48, wspace=0.05) |
|
150
|
|
|
ax1 = fig.add_subplot(grid[:1, :5]) |
|
151
|
|
|
ax2 = fig.add_subplot(grid[1:, :5]) |
|
152
|
|
|
ax3 = fig.add_subplot(grid[:1, 5:]) |
|
153
|
|
|
ax4 = fig.add_subplot(grid[1:, 5:]) |
|
154
|
|
|
|
|
155
|
|
|
# ax1 - Barplot |
|
156
|
|
|
colors = plt.get_cmap(cmap)(mv_cols / np.max(mv_cols)) # color bars by height |
|
157
|
|
|
ax1.bar(range(len(mv_cols)), np.round((mv_cols_ratio)*100, 2), color=colors) |
|
158
|
|
|
ax1.get_xaxis().set_visible(False) |
|
159
|
|
|
ax1.set(frame_on=False, xlim=(-.5, len(mv_cols)-0.5)) |
|
160
|
|
|
ax1.set_ylim(0, np.max(mv_cols_ratio)*100) |
|
161
|
|
|
ax1.grid(linestyle=':', linewidth=1) |
|
162
|
|
|
ax1.yaxis.set_major_formatter(ticker.PercentFormatter(decimals=0)) |
|
163
|
|
|
ax1.tick_params(axis='y', colors='#111111', length=1) |
|
164
|
|
|
|
|
165
|
|
|
# annotate values on top of the bars |
|
166
|
|
|
for rect, label in zip(ax1.patches, mv_cols): |
|
167
|
|
|
height = rect.get_height() |
|
168
|
|
|
ax1.text(.1 + rect.get_x() + rect.get_width() / 2, height+0.5, label, |
|
169
|
|
|
ha='center', |
|
170
|
|
|
va='bottom', |
|
171
|
|
|
rotation='90', |
|
172
|
|
|
alpha=0.5, |
|
173
|
|
|
fontsize='small') |
|
174
|
|
|
|
|
175
|
|
|
ax1.set_frame_on(True) |
|
176
|
|
|
for _, spine in ax1.spines.items(): |
|
177
|
|
|
spine.set_visible(True) |
|
178
|
|
|
spine.set_color(spine_color) |
|
179
|
|
|
ax1.spines['top'].set_color(None) |
|
180
|
|
|
|
|
181
|
|
|
# ax2 - Heatmap |
|
182
|
|
|
sns.heatmap(data.isna(), cbar=False, cmap='binary', ax=ax2) |
|
183
|
|
|
ax2.set_yticks(np.round(ax2.get_yticks()[0::5], -1)) |
|
184
|
|
|
ax2.set_yticklabels(ax2.get_yticks()) |
|
185
|
|
|
ax2.set_xticklabels( |
|
186
|
|
|
ax2.get_xticklabels(), |
|
187
|
|
|
horizontalalignment='center', |
|
188
|
|
|
fontweight='light', |
|
189
|
|
|
fontsize='medium') |
|
190
|
|
|
ax2.tick_params(length=1, colors='#111111') |
|
191
|
|
|
for _, spine in ax2.spines.items(): |
|
192
|
|
|
spine.set_visible(True) |
|
193
|
|
|
spine.set_color(spine_color) |
|
194
|
|
|
|
|
195
|
|
|
# ax3 - Summary |
|
196
|
|
|
fontax3 = {'color': '#111111', |
|
197
|
|
|
'weight': 'normal', |
|
198
|
|
|
'size': 12, |
|
199
|
|
|
} |
|
200
|
|
|
ax3.get_xaxis().set_visible(False) |
|
201
|
|
|
ax3.get_yaxis().set_visible(False) |
|
202
|
|
|
ax3.set(frame_on=False) |
|
203
|
|
|
|
|
204
|
|
|
ax3.text(0.1, 0.9, f"Total: {np.round(total_datapoints/1000,1)}K", |
|
205
|
|
|
transform=ax3.transAxes, |
|
206
|
|
|
fontdict=fontax3) |
|
207
|
|
|
ax3.text(0.1, 0.7, f"Missing: {np.round(mv_total/1000,1)}K", |
|
208
|
|
|
transform=ax3.transAxes, |
|
209
|
|
|
fontdict=fontax3) |
|
210
|
|
|
ax3.text(0.1, 0.5, f"Relative: {np.round(mv_total/total_datapoints*100,1)}%", |
|
211
|
|
|
transform=ax3.transAxes, |
|
212
|
|
|
fontdict=fontax3) |
|
213
|
|
|
ax3.text(0.1, 0.3, f"Max-col: {np.round(mv_cols.max()/data.shape[0]*100)}%", |
|
214
|
|
|
transform=ax3.transAxes, |
|
215
|
|
|
fontdict=fontax3) |
|
216
|
|
|
ax3.text(0.1, 0.1, f"Max-row: {np.round(mv_rows.max()/data.shape[1]*100)}%", |
|
217
|
|
|
transform=ax3.transAxes, |
|
218
|
|
|
fontdict=fontax3) |
|
219
|
|
|
|
|
220
|
|
|
# ax4 - Scatter plot |
|
221
|
|
|
ax4.get_yaxis().set_visible(False) |
|
222
|
|
|
for _, spine in ax4.spines.items(): |
|
223
|
|
|
spine.set_color(spine_color) |
|
224
|
|
|
ax4.tick_params(axis='x', colors='#111111', length=1) |
|
225
|
|
|
|
|
226
|
|
|
ax4.scatter(mv_rows, range(len(mv_rows)), s=mv_rows, c=mv_rows, cmap=cmap, marker=".", vmin=1) |
|
227
|
|
|
ax4.set_ylim((0, len(mv_rows))[::-1]) # limit and invert y-axis |
|
228
|
|
|
ax4.set_xlim(0, max(mv_rows)+0.5) |
|
229
|
|
|
ax4.grid(linestyle=':', linewidth=1) |
|
230
|
|
|
|
|
231
|
|
|
ax1.set_title('Missing value plot', pad=40, fontdict={'fontsize': 18}) |
|
232
|
|
|
return grid |
|
233
|
|
|
|
|
234
|
|
|
|
|
235
|
|
|
# Correlation matrix / heatmap |
|
236
|
|
|
def corr_plot(data, split=None, threshold=0, cmap='BrBG', figsize=(12, 10), annot=True, dev=False, **kwargs): |
|
237
|
|
|
''' |
|
238
|
|
|
Two-dimensional visualization of the correlation between feature-columns, excluding NA values. |
|
239
|
|
|
|
|
240
|
|
|
Parameters |
|
241
|
|
|
---------- |
|
242
|
|
|
data: 2D dataset that can be coerced into Pandas DataFrame. If a Pandas DataFrame is provided, the index/column \ |
|
243
|
|
|
information is used to label the plots. |
|
244
|
|
|
|
|
245
|
|
|
split: {None, 'pos', 'neg', 'high', 'low'}, default None |
|
246
|
|
|
Type of split to be performed. |
|
247
|
|
|
|
|
248
|
|
|
* None: visualize all correlations between the feature-columns. |
|
249
|
|
|
* pos: visualize all positive correlations between the feature-columns above the threshold. |
|
250
|
|
|
* neg: visualize all negative correlations between the feature-columns below the threshold. |
|
251
|
|
|
* high: visualize all correlations between the feature-columns for which abs(corr) > threshold is True. |
|
252
|
|
|
* low: visualize all correlations between the feature-columns for which abs(corr) < threshold is True. |
|
253
|
|
|
|
|
254
|
|
|
threshold: float, default 0 |
|
255
|
|
|
Value between 0 <= threshold <= 1 |
|
256
|
|
|
|
|
257
|
|
|
cmap: matplotlib colormap name or object, or list of colors, default 'BrBG' |
|
258
|
|
|
The mapping from data values to color space. |
|
259
|
|
|
|
|
260
|
|
|
figsize: tuple, default (12, 10) |
|
261
|
|
|
Use to control the figure size. |
|
262
|
|
|
|
|
263
|
|
|
annot: bool, default True |
|
264
|
|
|
Use to show or hide annotations. |
|
265
|
|
|
|
|
266
|
|
|
dev: bool, default False |
|
267
|
|
|
Display figure settings in the plot by setting dev = True. If False, the settings are not displayed. Use for \ |
|
268
|
|
|
presentations. |
|
269
|
|
|
|
|
270
|
|
|
**kwargs: optional |
|
271
|
|
|
Additional elements to control the visualization of the plot, e.g.: |
|
272
|
|
|
|
|
273
|
|
|
* mask: bool, default True |
|
274
|
|
|
If set to False the entire correlation matrix, including the upper triangle is shown. Set dev = False in this \ |
|
275
|
|
|
case to avoid overlap. |
|
276
|
|
|
* vmax: float, default is calculated from the given correlation coefficients. |
|
277
|
|
|
Value between -1 or vmin <= vmax <= 1, limits the range of the colorbar. |
|
278
|
|
|
* vmin: float, default is calculated from the given correlation coefficients. |
|
279
|
|
|
Value between -1 <= vmin <= 1 or vmax, limits the range of the colorbar. |
|
280
|
|
|
* linewidths: float, default 0.5 |
|
281
|
|
|
Controls the line-width inbetween the squares. |
|
282
|
|
|
* annot_kws: dict, default {'size' : 10} |
|
283
|
|
|
Controls the font size of the annotations. Only available when annot = True. |
|
284
|
|
|
* cbar_kws: dict, default {'shrink': .95, 'aspect': 30} |
|
285
|
|
|
Controls the size of the colorbar. |
|
286
|
|
|
* Many more kwargs are available, i.e. 'alpha' to control blending, or options to adjust labels, ticks ... |
|
287
|
|
|
|
|
288
|
|
|
Kwargs can be supplied through a dictionary of key-value pairs (see above). |
|
289
|
|
|
|
|
290
|
|
|
Returns |
|
291
|
|
|
------- |
|
292
|
|
|
figure |
|
293
|
|
|
''' |
|
294
|
|
|
|
|
295
|
|
|
data = pd.DataFrame(data) |
|
296
|
|
|
|
|
297
|
|
|
# Obtain correlation matrix |
|
298
|
|
|
corr = corr_mat(data, split=split, threshold=threshold).data |
|
299
|
|
|
|
|
300
|
|
|
# Generate mask for the upper triangle |
|
301
|
|
|
mask = np.triu(np.ones_like(corr, dtype=np.bool)) |
|
302
|
|
|
|
|
303
|
|
|
# Compute dimensions and correlation range to adjust settings |
|
304
|
|
|
vmax = np.round(np.nanmax(corr.where(~mask))-0.05, 2) |
|
305
|
|
|
vmin = np.round(np.nanmin(corr.where(~mask))+0.05, 2) |
|
306
|
|
|
|
|
307
|
|
|
# Set up the matplotlib figure and generate colormap |
|
308
|
|
|
fig, ax = plt.subplots(figsize=figsize) |
|
309
|
|
|
|
|
310
|
|
|
# Specify kwargs for the heatmap |
|
311
|
|
|
kwargs = {'mask': mask, |
|
312
|
|
|
'cmap': cmap, |
|
313
|
|
|
'annot': annot, |
|
314
|
|
|
'vmax': vmax, |
|
315
|
|
|
'vmin': vmin, |
|
316
|
|
|
'linewidths': .5, |
|
317
|
|
|
'annot_kws': {'size': 10}, |
|
318
|
|
|
'cbar_kws': {'shrink': .95, 'aspect': 30}, |
|
319
|
|
|
**kwargs} |
|
320
|
|
|
|
|
321
|
|
|
# Draw heatmap with mask and some default settings |
|
322
|
|
|
sns.heatmap(corr, |
|
323
|
|
|
center=0, |
|
324
|
|
|
square=True, |
|
325
|
|
|
fmt='.2f', |
|
326
|
|
|
**kwargs |
|
327
|
|
|
) |
|
328
|
|
|
|
|
329
|
|
|
ax.set_title('Feature-correlation Matrix', fontdict={'fontsize': 18}) |
|
330
|
|
|
|
|
331
|
|
|
# Display settings |
|
332
|
|
|
if dev: |
|
333
|
|
|
fig.suptitle(f"\ |
|
334
|
|
|
Settings (dev-mode): \n\ |
|
335
|
|
|
- split-mode: {split} \n\ |
|
336
|
|
|
- threshold: {threshold} \n\ |
|
337
|
|
|
- annotations: {annot} \n\ |
|
338
|
|
|
- cbar: \n\ |
|
339
|
|
|
- vmax: {vmax} \n\ |
|
340
|
|
|
- vmin: {vmin} \n\ |
|
341
|
|
|
- linewidths: {kwargs['linewidths']} \n\ |
|
342
|
|
|
- annot_kws: {kwargs['annot_kws']} \n\ |
|
343
|
|
|
- cbar_kws: {kwargs['cbar_kws']}", |
|
344
|
|
|
fontsize=12, |
|
345
|
|
|
color='gray', |
|
346
|
|
|
x=0.35, |
|
347
|
|
|
y=0.85, |
|
348
|
|
|
ha='left') |
|
349
|
|
|
|
|
350
|
|
|
return ax |
|
351
|
|
|
|
akanz1 /
klib
