forked from luke14free/aso-sherlock
/
sherlock.py
281 lines (231 loc) · 11.5 KB
/
sherlock.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
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
from optparse import OptionParser
from typing import Dict, List, Any, Union, Tuple
import jinja2
import pandas as pd
import sys
import logging
from matplotlib.pylab import plt
from pandas.errors import ParserError
from pmprophet import PMProphet, Sampler
import numpy as np
import pymc3 as pm
from lib.helpers import figure_to_base64, safe_mean, ERRORS, REQUIRED_COLUMNS, WARNINGS, OPTIONAL_COLUMNS
from lib.plotting import plot_nowcast, plot_seasonality
def read_input_file(file_path: str) -> pd.DataFrame:
try:
df = pd.read_csv(file_path)
except FileNotFoundError:
logging.error(ERRORS['not_found'])
sys.exit()
except ParserError:
logging.error(ERRORS['not_readable'])
sys.exit()
missing_columns = ", ".join(set(REQUIRED_COLUMNS) - set(df.columns))
if missing_columns:
logging.error(ERRORS['missing_columns'].format(missing_columns))
sys.exit()
try:
df['date'] = pd.to_datetime(df['date'], format='%d/%m/%Y')
except ValueError:
logging.error(ERRORS['date_column_not_readable'])
sys.exit()
return df.sort_values('date')
def handle_outliers(data: pd.DataFrame) -> pd.DataFrame:
if options.sigma:
y = []
raw_y = data['y'].values
for idx, val in enumerate(raw_y):
if val < 0 or not (raw_y.mean() - options.sigma * raw_y.std()) < val < (
raw_y.mean() + options.sigma * raw_y.std()):
ts_slice = raw_y[idx - 20:idx + 20]
y.append(np.median(ts_slice[ts_slice >= 0]))
else:
y.append(val)
data['y'] = y
else:
if data['y'].min() < 0:
raise Exception(ERRORS['conversion_less_than_0'])
return data
def fit_beta_regression(model: PMProphet, data: pd.DataFrame) -> PMProphet:
model._prepare_fit()
with model.model:
mean = pm.Deterministic('y_%s' % model.name, model.y) # no scaling needed
hp_alpha = pm.HalfCauchy('y_alpha_%s' % model.name, 2.5)
hp_beta = pm.Deterministic('y_beta_%s' % model.name, hp_alpha * ((1 - mean) / mean))
pm.Beta("observed_%s" % model.name, hp_alpha, hp_beta, observed=data['y'])
pm.Deterministic('y_hat_%s' % model.name, mean)
model.fit(10000 if options.sampler == 'metropolis' else 2000,
method=Sampler.METROPOLIS if options.sampler == 'metropolis' else Sampler.NUTS,
finalize=False,
step_kwargs={'compute_convergence_checks': False} if options.sampler == 'metropolis' else {})
return model
def summary_from_model_regressors(model: PMProphet, regressors: Union[List, Tuple]) -> List[
Dict[str, Union[str, float]]]:
alpha = options.alpha
summary = []
for idx, regressor in enumerate(regressors):
error = (pd.np.percentile(
model.trace['regressors_{}'.format(model.name)][:, idx],
100 - (alpha * 100 / 2)
) - pd.np.percentile(
model.trace['regressors_{}'.format(model.name)][:, idx],
(alpha * 100 / 2)
)) / 2
summary.append({
'name': regressor,
'median': pd.np.round(pd.np.median(model.trace['regressors_{}'.format(model.name)][:, idx] * 100), 2),
'error': pd.np.round(error * 100, 2),
})
return summary
def create_model(model_name: str, data: pd.DataFrame, growth: bool, regressors: Union[List, Tuple] = (),
changepoints: Union[List, Tuple] = ()) -> PMProphet:
model = PMProphet(
data,
growth=growth,
seasonality_prior_scale=options.seasonality_scale,
changepoints=[] if not changepoints else changepoints,
name=model_name,
)
#model.skip_first = 200 if options.sampler == 'nuts' else 10000
for regressor in regressors:
model.add_regressor(regressor)
if not options.weekly:
model.add_seasonality(7, 3)
if (data['ds'].max() - data['ds'].min()).days > 365:
model.add_seasonality(365, 5)
return model
def visual_update_analysis(df: pd.DataFrame) -> Tuple[Dict[str, str], List[
Dict[str, Union[str, float]]]]:
summary = []
template_vars = {}
df = df.rename(columns={'date': 'ds'})
if 'asa_impressions' in df.columns:
df['impressions'] = df['search_impressions'] - df['asa_impressions']
df['conversions'] = df['search_downloads'] - df['asa']
else:
df['impressions'] = df['search_impressions']
df['conversions'] = df['search_downloads']
df['y'] = df['conversions'] / df['impressions']
df.index = df['ds']
if options.weekly:
df = df.resample('W').apply(safe_mean)
df = handle_outliers(df.copy())
time_regressors = []
for _, row in df.iterrows():
if row['update'] == 'visual':
additional_regressor = '{} (visual)'.format(str(row['ds']).split(" ")[0])
df[additional_regressor] = [1 if other_row['ds'] >= row['ds'] else 0 for
_, other_row in df.iterrows()]
time_regressors.append(additional_regressor)
model = create_model('sherlock_visual', df, False, time_regressors)
conversion_model = fit_beta_regression(model, df)
fig = plot_nowcast(conversion_model, [row['ds'] for _, row in df.iterrows() if row['update'] == 'visual'])
plt.title('Conversion & Visual Updates')
template_vars['visual_model'] = figure_to_base64(fig)
summary.extend(summary_from_model_regressors(conversion_model, time_regressors))
seasonality = {}
for period, fig in plot_seasonality(conversion_model, alpha=options.alpha, plot_kwargs={}).items():
seasonality[int(period)] = figure_to_base64(fig)
template_vars['conversion_seasonality'] = seasonality
return template_vars, summary
def textual_update_analysis(df: pd.DataFrame, extra_columns: List) -> Tuple[Dict[str, str], List[
Dict[str, Union[str, float]]]]:
template_vars: Dict[str, Any] = {}
summary = []
df = df.rename(columns={'date': 'ds', 'search_downloads': 'y'})
if 'asa' in df.columns:
df['y'] = df['y'] - df['asa']
df.index = df['ds']
df = handle_outliers(df)
if options.weekly:
df = df.resample('W').apply(safe_mean)
time_regressors = []
for _, row in df.iterrows():
if row['update'] == 'textual':
additional_regressor = '{} (text)'.format(str(row['ds']).split(" ")[0])
df[additional_regressor] = [other_row['y'] if other_row['ds'] >= row['ds'] else 0 for
_, other_row in df.iterrows()]
time_regressors.append(additional_regressor)
model = create_model('sherlock_textual', df, True, time_regressors + extra_columns)
model.fit(10000 if options.sampler == 'metropolis' else 2000,
method=Sampler.METROPOLIS if options.sampler == 'metropolis' else Sampler.NUTS,
step_kwargs={'compute_convergence_checks': False} if options.sampler == 'metropolis' else {})
fig = plot_nowcast(model, [row['ds'] for _, row in df.iterrows() if row['update'] == 'textual'])
plt.title('Downloads & Textual Updates')
template_vars['textual_model'] = figure_to_base64(fig)
summary.extend(summary_from_model_regressors(model, time_regressors + extra_columns))
extra_regressors_plots: List[Dict[str, str]] = []
for i in range(len(time_regressors), len(time_regressors) + len(extra_columns)):
fig = plt.figure()
plt.grid()
plt.hist(model.trace['regressors_{}'.format(model.name)][:, i] * 100, bins=30, alpha=0.8, histtype='stepfilled')
plt.axvline(np.median(model.trace['regressors_{}'.format(model.name)][:, i]) * 100, color="C3", lw=1,
ls="dotted")
plt.title("{} (in %)".format(extra_columns[i - len(time_regressors)]))
extra_regressors_plots.append({
'name': extra_columns[i - len(time_regressors)],
'img_data': figure_to_base64(fig)
})
template_vars['extra_regressors_plots'] = extra_regressors_plots
seasonality = {}
for period, fig in plot_seasonality(model, alpha=options.alpha, plot_kwargs={}).items():
seasonality[int(period)] = figure_to_base64(fig)
template_vars['textual_seasonality'] = seasonality
return template_vars, summary
def run_sherlock() -> None:
template_vars = {'textual_seasonality': {}, 'conversion_seasonality': {}}
df = read_input_file(options.input_file)
for unknown_update in (set(df['update'].unique()) - {'textual', 'visual', pd.np.nan}):
logging.warning(WARNINGS['update_not_understood'].format(unknown_update))
time_span = (df['date'].max() - df['date'].min()).days
if time_span < 7:
logging.error(ERRORS['timespan_too_short'])
sys.exit()
if time_span < 30:
logging.warning(WARNINGS['timespan_too_short'])
extra_columns = list(set(df.columns) - set(REQUIRED_COLUMNS + OPTIONAL_COLUMNS + ['date', 'search_downloads']))
summary = []
if 'visual' in df['update'].unique():
tv, s = visual_update_analysis(df.copy())
template_vars.update(tv)
summary.extend(s)
if 'textual' in df['update'].unique():
tv, s = textual_update_analysis(df.copy(), extra_columns)
template_vars.update(tv)
summary.extend(s)
if options.app_name:
template_vars['app_name'] = options.app_name
template_vars['summary'] = summary
template_env = jinja2.Environment(loader=jinja2.FileSystemLoader(searchpath="./lib"))
template = template_env.get_template("template.html")
with open(options.output_file, 'w') as output_file:
output_file.write(template.render(**template_vars))
if __name__ == '__main__':
parser = OptionParser()
parser.add_option("-a", "--app-name", dest='app_name', default=None,
help="Specify the app name if you want a more personalized report")
parser.add_option("-i", "--input-file", dest="input_file",
help="Input CSV file", metavar="FILE")
parser.add_option("-o", "--output-file", dest="output_file",
help="Output report file (in html format)", metavar="FILE", default='report.html')
parser.add_option("-s", "--sampler", dest='sampler', choices=['metropolis', 'nuts'], default='nuts',
help='Sampler to use ("nuts" is slower but more precise and suggested, otherwise "metropolis")')
parser.add_option("-n", "--no-asa", dest='no_asa', action="store_true", default=False,
help="Do not use ASA as an additional regressor (better seasonality fits)")
parser.add_option("-w", "--weekly", dest='weekly', action="store_true", default=False,
help="Run the analysis on a weekly resampling")
parser.add_option("-r", "--remove-outliers-sigma", dest='sigma', default=False, type='float',
help='''Remove outliers at more than X sigma from the mean (suggested values range between 1.5-3.5).
Default value is: 0 that means that Sherlock will not remove outliers''')
parser.add_option("-l", "--significance-level", dest='alpha', default=0.05, type='float',
help="The significance level for the analysis (default is 0.05)")
parser.add_option("-k", "--seasonality-scale", dest='seasonality_scale', default=5.0, type='float',
help="""The scale of the seasonality, if it fits poorly because you have
great variance due to seasonality increase this""")
(options, args) = parser.parse_args()
if not options.input_file:
logging.error(ERRORS['no_input'])
sys.exit()
if options.no_asa:
OPTIONAL_COLUMNS.append('asa')
run_sherlock()