-
Notifications
You must be signed in to change notification settings - Fork 42
/
test_convoys.py
415 lines (338 loc) · 14.7 KB
/
test_convoys.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
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
import autograd
import datetime
import flaky
import matplotlib
import numpy
import pandas
import pytest
import random
import scipy.stats
matplotlib.use('Agg') # Needed for matplotlib to run in Travis
import convoys
import convoys.plotting
import convoys.regression
import convoys.single
import convoys.utils
def sample_weibull(k, lambd):
# scipy.stats is garbage for this
# exp(-(x * lambda)^k) = y
return (-numpy.log(random.random())) ** (1.0/k) / lambd
def generate_censored_data(N, E, C):
B = numpy.array([c and e < n for n, e, c in zip(N, E, C)])
T = numpy.array([e if b else n for e, b, n in zip(E, B, N)])
return B, T
def test_kaplan_meier_model():
data = [
(2, 0),
(3, 0),
(6, 1),
(6, 1),
(7, 1),
(10, 0)
]
now = pandas.Timestamp('2019-01-22') # fix now end date for easier testing
created_array = [now - pandas.DateOffset(t) for t, e in data]
converted_array = [ts + pandas.DateOffset(t) if e == 1 else numpy.nan for ts, (t, e) in zip(created_array, data)]
df = pandas.DataFrame({
'created_at': created_array,
'converted_at': converted_array,
'group': 1
})
df['now'] = now
unit, groups, (G, B, T) = convoys.utils.get_arrays(
df,
converted='converted_at',
created='created_at',
unit='days'
)
m = convoys.multi.KaplanMeier()
m.fit(G, B, T)
assert m.predict(0, 9) == 0.75
def test_output_shapes(c=0.3, lambd=0.1, n=1000, k=5):
X = numpy.random.randn(n, k)
C = scipy.stats.bernoulli.rvs(c, size=(n,))
N = scipy.stats.uniform.rvs(scale=5./lambd, size=(n,))
E = scipy.stats.expon.rvs(scale=1./lambd, size=(n,))
B, T = generate_censored_data(N, E, C)
# Fit model with ci
model = convoys.regression.Exponential(mcmc=True)
model.fit(X, B, T)
# Generate output without ci
assert model.predict(X[0], 0).shape == ()
assert model.predict([X[0], X[1]], 0).shape == (2,)
assert model.predict([X[0]], [0, 1, 2, 3]).shape == (4,)
assert model.predict([X[0], X[1], X[2]], [0, 1, 2]).shape == (3,)
assert model.predict([[X[0], X[1]]], [[0], [1], [2]]).shape == (3, 2)
assert model.predict([[X[0]], [X[1]]], [[0, 1, 2]]).shape == (2, 3)
# Generate output with ci (same as above plus (3,))
assert model.predict_ci(X[0], 0, ci=0.8).shape == (3,)
assert model.predict_ci([X[0], X[1]], 0, ci=0.8).shape == (2, 3)
assert model.predict_ci([X[0]], [0, 1, 2, 3], ci=0.8).shape == (4, 3)
assert model.predict_ci([X[0], X[1], X[2]], [0, 1, 2], ci=0.8) \
.shape == (3, 3)
assert model.predict_ci([[X[0], X[1]]], [[0], [1], [2]], ci=0.8) \
.shape == (3, 2, 3)
assert model.predict_ci([[X[0]], [X[1]]], [[0, 1, 2]], ci=0.8) \
.shape == (2, 3, 3)
# Assert old interface still works
assert model.cdf(X[0], 0).shape == ()
assert model.cdf(X[0], 0, ci=0.8).shape == (3,)
# Fit model without ci (should be the same)
model = convoys.regression.Exponential(mcmc=False)
model.fit(X, B, T)
assert model.predict(X[0], 0).shape == ()
assert model.predict([X[0], X[1]], [0, 1]).shape == (2,)
@flaky.flaky
def test_exponential_regression_model(c=0.3, lambd=0.1, n=10000):
X = numpy.ones((n, 1))
C = scipy.stats.bernoulli.rvs(c, size=(n,)) # did it convert
N = scipy.stats.uniform.rvs(scale=5./lambd, size=(n,)) # time now
E = scipy.stats.expon.rvs(scale=1./lambd, size=(n,)) # time of event
B, T = generate_censored_data(N, E, C)
model = convoys.regression.Exponential(mcmc=True)
model.fit(X, B, T)
assert 0.80*c < model.predict([1], float('inf')) < 1.30*c
for t in [1, 3, 10]:
d = 1 - numpy.exp(-lambd*t)
assert 0.80*c*d < model.predict([1], t) < 1.30*c*d
# Check the confidence intervals
assert model.predict_ci([1], float('inf'), ci=0.95).shape == (3,)
assert model.predict_ci([1], [0, 1, 2, 3], ci=0.95).shape == (4, 3)
y, y_lo, y_hi = model.predict_ci([1], float('inf'), ci=0.95)
assert 0.80*c < y < 1.30*c
# Check the random variates
will_convert, convert_at = model.rvs([1], n_curves=10000, n_samples=1)
assert 0.80*c < numpy.mean(will_convert) < 1.30*c
convert_times = convert_at[will_convert]
for t in [1, 3, 10]:
d = 1 - numpy.exp(-lambd*t)
assert 0.70*d < (convert_times < t).mean() < 1.30*d
# Fit a linear model
model = convoys.regression.Exponential(mcmc=False, flavor='linear')
model.fit(X, B, T)
model_c = model.params['map']['b'] + model.params['map']['beta'][0]
assert 0.9*c < model_c < 1.1*c
for t in [1, 3, 10]:
d = 1 - numpy.exp(-lambd*t)
assert 0.80*c*d < model.predict([1], t) < 1.30*c*d
@flaky.flaky
def test_weibull_regression_model(cs=[0.3, 0.5, 0.7],
lambd=0.1, k=0.5, n=10000):
X = numpy.array([[r % len(cs) == j for j in range(len(cs))]
for r in range(n)])
C = numpy.array([bool(random.random() < cs[r % len(cs)])
for r in range(n)])
N = scipy.stats.uniform.rvs(scale=5./lambd, size=(n,))
E = numpy.array([sample_weibull(k, lambd)
for r in range(n)])
B, T = generate_censored_data(N, E, C)
model = convoys.regression.Weibull()
model.fit(X, B, T)
# Validate shape of results
x = numpy.ones((len(cs),))
assert model.predict(x, float('inf')).shape == ()
assert model.predict(x, 1).shape == ()
assert model.predict(x, [1, 2, 3, 4]).shape == (4,)
# Check results
for r, c in enumerate(cs):
x = [int(r == j) for j in range(len(cs))]
assert 0.80 * c < model.predict(x, float('inf')) < 1.30 * c
# Fit a linear model
model = convoys.regression.Weibull(mcmc=False, flavor='linear')
model.fit(X, B, T)
model_cs = model.params['map']['b'] + model.params['map']['beta']
for model_c, c in zip(model_cs, cs):
assert 0.8 * c < model_c < 1.2 * c
@flaky.flaky
def test_gamma_regression_model(c=0.3, lambd=0.1, k=3.0, n=10000):
# TODO: this one seems very sensitive to large values for N (i.e. less censoring)
X = numpy.ones((n, 1))
C = scipy.stats.bernoulli.rvs(c, size=(n,))
N = scipy.stats.uniform.rvs(scale=20./lambd, size=(n,))
E = scipy.stats.gamma.rvs(a=k, scale=1.0/lambd, size=(n,))
B, T = generate_censored_data(N, E, C)
model = convoys.regression.Gamma()
model.fit(X, B, T)
assert 0.80*c < model.predict([1], float('inf')) < 1.30*c
assert 0.80*k < numpy.mean(model.params['map']['k']) < 1.30*k
# Fit a linear model
model = convoys.regression.Gamma(mcmc=False, flavor='linear')
model.fit(X, B, T)
model_c = model.params['map']['b'] + model.params['map']['beta'][0]
assert 0.9*c < model_c < 1.1*c
@flaky.flaky
def test_linear_model(n=10000, m=5, k=3.0, lambd=0.1):
# Generate data with little censoring
# The coefficients should be quite close to their actual value
cs = numpy.random.dirichlet(numpy.ones(m))
X = numpy.random.binomial(n=1, p=0.5, size=(n, m))
C = numpy.random.rand(n) < numpy.dot(X, cs.T)
N = scipy.stats.uniform.rvs(scale=20./lambd, size=(n,))
E = numpy.array([sample_weibull(k, lambd) for r in range(n)])
B, T = generate_censored_data(N, E, C)
model = convoys.regression.Weibull(mcmc=False, flavor='linear')
model.fit(X, B, T)
# Check the fitted parameters
model_cs = model.params['map']['b'] + model.params['map']['beta']
for model_c, c in zip(model_cs, cs):
assert c - 0.03 < model_c < c + 0.03
model_lambds = numpy.exp(model.params['map']['a'] +
model.params['map']['alpha'])
for model_lambd in model_lambds:
assert 0.95*lambd < model_lambd < 1.05*lambd
# Check predictions
for i, c in enumerate(cs):
x = numpy.array([float(j == i) for j in range(m)])
p = model.predict(x, float('inf'))
assert c - 0.03 < p < c + 0.03
t = 10.0
p = model.predict(x, t)
f = 1 - numpy.exp(-(t*lambd)**k)
assert c*f - 0.03 < p < c*f + 0.03
@flaky.flaky
def test_exponential_pooling(c=0.5, lambd=0.01, n=10000, ks=[1, 2, 3]):
# Generate one series of n observations with c conversion rate
# Then k1...kn series with zero conversion
# The predicted conversion rates should go towards c for the small cohorts
G = numpy.zeros(n + sum(ks))
C = numpy.zeros(n + sum(ks))
N = numpy.zeros(n + sum(ks))
E = numpy.zeros(n + sum(ks))
offset = 0
for i, k in enumerate([n] + ks):
G[offset:offset+k] = i
offset += k
C[:n] = scipy.stats.bernoulli.rvs(c, size=(n,))
N[:] = 1000.
E[:n] = scipy.stats.expon.rvs(scale=1./lambd, size=(n,))
B, T = generate_censored_data(N, E, C)
# Fit model
model = convoys.multi.Exponential()
model.fit(G, B, T)
# Generate predictions for each cohort
c = numpy.array([model.predict(i, float('inf')) for i in range(1+len(ks))])
assert numpy.all(c[1:] > 0.25) # rough check
assert numpy.all(c[1:] < 0.50) # same
assert numpy.all(numpy.diff(c) < 0) # c should be monotonically decreasing
def _generate_dataframe(cs=[0.3, 0.5, 0.7], k=0.5, lambd=0.1, n=1000):
groups = [r % len(cs) for r in range(n)]
C = numpy.array([bool(random.random() < cs[g]) for g in groups])
N = scipy.stats.expon.rvs(scale=10./lambd, size=(n,))
E = numpy.array([sample_weibull(k, lambd) for r in range(n)])
B, T = generate_censored_data(N, E, C)
x2t = lambda x: datetime.datetime(2000, 1, 1) + datetime.timedelta(days=x)
return pandas.DataFrame(data=dict(
group=['Group %d' % g for g in groups],
created=[x2t(0) for g in groups],
converted=[x2t(t) if b else None for t, b in zip(T, B)],
now=[x2t(n) for n in N]
))
def test_convert_dataframe(n=1000):
df = _generate_dataframe(n=n)
unit, groups, (G, B, T) = convoys.utils.get_arrays(df)
assert G.shape == B.shape == T.shape == (n,)
def test_convert_dataframe_features(n=1000):
df = _generate_dataframe(n=n)
df['features'] = [tuple(numpy.random.randn() for z in range(3))
for g in df['group']]
df = df.drop('group', axis=1)
unit, groups, (X, B, T) = convoys.utils.get_arrays(df)
assert X.shape == (n, 3)
# Generate from multiple columns
df = _generate_dataframe(n=n)
df['feature_1'] = [numpy.random.randn() for g in df['group']]
df['feature_2'] = [numpy.random.randn() for g in df['group']]
df = df.drop('group', axis=1)
unit, groups, (X, B, T) = convoys.utils.get_arrays(
df, features=('feature_1', 'feature_2'))
assert X.shape == (n, 2)
def test_convert_dataframe_infer_now():
df = _generate_dataframe()
df = df.drop('now', axis=1)
unit, groups, (G1, B1, T1) = convoys.utils.get_arrays(df, unit='days')
# Now, let's make the timezone-naive objects timezone aware
utc = datetime.timezone.utc
local = datetime.datetime.now(utc).astimezone().tzinfo
df[['created', 'converted']] = df[['created', 'converted']].applymap(
lambda z: z.replace(tzinfo=local))
unit, groups, (G2, B2, T2) = convoys.utils.get_arrays(df, unit='days')
# Convert everything to UTC and make sure it's still the same
df[['created', 'converted']] = df[['created', 'converted']].applymap(
lambda z: z.tz_convert(utc))
unit, groups, (G3, B3, T3) = convoys.utils.get_arrays(df, unit='days')
# Let's check that all deltas are the same
# There will be some slight clock drift, so let's accept up to 3s
for t1, t2, t3 in zip(T1, T2, T3):
assert 0 <= t2 - t1 < 3.0 / (24*60*60)
assert 0 <= t3 - t1 < 3.0 / (24*60*60)
def test_convert_dataframe_timedeltas():
df = _generate_dataframe()
unit, groups, (G1, B1, T1) = convoys.utils.get_arrays(df, unit='days')
df2 = pandas.DataFrame({'group': df['group'],
'converted': df['converted'] - df['created'],
'now': df['now'] - df['created']})
unit, groups, (G2, B2, T2) = convoys.utils.get_arrays(df2, unit='days')
for t1, t2 in zip(T1, T2):
assert 0 <= t2 - t1 < 3.0 / (24*60*60)
def test_convert_dataframe_more_args():
df = _generate_dataframe()
unit, groups, (G, B, T) = convoys.utils.get_arrays(df, max_groups=2)
assert len(groups) <= 2
unit, groups, (G, B, T) = convoys.utils.get_arrays(df, group_min_size=9999)
assert G.shape == (0,)
def test_convert_dataframe_created_at_nan(n=1000):
df = _generate_dataframe(n=n)
df.loc[df.index[0], 'created'] = None
unit, groups, (G, B, T) = convoys.utils.get_arrays(df)
assert numpy.issubdtype(G.dtype, numpy.integer)
assert numpy.issubdtype(B.dtype, numpy.bool_)
assert numpy.issubdtype(T.dtype, numpy.number)
def _test_plot_cohorts(model='weibull', extra_model=None):
df = _generate_dataframe()
unit, groups, (G, B, T) = convoys.utils.get_arrays(df)
matplotlib.pyplot.clf()
convoys.plotting.plot_cohorts(G, B, T, model=model, ci=0.95, groups=groups)
matplotlib.pyplot.legend()
if extra_model:
convoys.plotting.plot_cohorts(G, B, T, model=extra_model,
plot_kwargs=dict(linestyle='--',
alpha=0.1))
matplotlib.pyplot.savefig('%s-%s.png' % (model, extra_model)
if extra_model is not None else '%s.png' % model)
def test_plot_cohorts_model():
df = _generate_dataframe()
unit, groups, (G, B, T) = convoys.utils.get_arrays(df)
model = convoys.multi.Exponential(mcmc=None)
model.fit(G, B, T)
matplotlib.pyplot.clf()
convoys.plotting.plot_cohorts(G, B, T, model=model, groups=groups)
matplotlib.pyplot.legend()
with pytest.raises(Exception):
convoys.plotting.plot_cohorts(G, B, T, model='bad', groups=groups)
with pytest.raises(Exception):
convoys.plotting.plot_cohorts(G, B, T, model=model, groups=groups,
specific_groups=['Nonsense'])
@flaky.flaky
def test_plot_cohorts_kaplan_meier():
_test_plot_cohorts(model='kaplan-meier')
@flaky.flaky
def test_plot_cohorts_weibull():
_test_plot_cohorts(model='weibull')
@flaky.flaky
def test_plot_cohorts_two_models():
_test_plot_cohorts(model='kaplan-meier', extra_model='weibull')
def test_plot_cohorts_subplots():
df = _generate_dataframe()
unit, groups, (G, B, T) = convoys.utils.get_arrays(df)
matplotlib.pyplot.clf()
fix, axes = matplotlib.pyplot.subplots(nrows=2, ncols=2)
for ax in axes.flatten():
convoys.plotting.plot_cohorts(G, B, T, groups=groups, ax=ax)
ax.legend()
matplotlib.pyplot.savefig('subplots.png')
def test_marriage_example():
from examples.marriage import run
run()
def test_dob_violations_example():
from examples.dob_violations import run
run()