def test_bmt():
# All tests against SAS
# Results taken from here:
# http://support.sas.com/documentation/cdl/en/statug/68162/HTML/default/viewer.htm#statug_lifetest_details03.htm
# Confidence intervals for 25% percentile of the survival
# distribution (for "ALL" subjects), taken from the SAS web site
cb = {"linear": [107, 276],
"cloglog": [86, 230],
"log": [107, 332],
"asinsqrt": [104, 276],
"logit": [104, 230]}
dfa = bmt[bmt.Group == "ALL"]
cur_dir = os.path.dirname(os.path.abspath(__file__))
fp = os.path.join(cur_dir, 'results', 'bmt_results.csv')
rslt = pd.read_csv(fp)
sf = SurvfuncRight(dfa["T"].values, dfa.Status.values)
assert_allclose(sf.surv_times, rslt.t)
assert_allclose(sf.surv_prob, rslt.s, atol=1e-4, rtol=1e-4)
assert_allclose(sf.surv_prob_se, rslt.se, atol=1e-4, rtol=1e-4)
for method in "linear", "cloglog", "log", "logit", "asinsqrt":
lcb, ucb = sf.quantile_ci(0.25, method=method)
assert_allclose(cb[method], np.r_[lcb, ucb])
def test_bmt():
# All tests against SAS
# Results taken from here:
# http://support.sas.com/documentation/cdl/en/statug/68162/HTML/default/viewer.htm#statug_lifetest_details03.htm
# Confidence intervals for 25% percentile of the survival
# distribution (for "ALL" subjects), taken from the SAS web site
cb = {
"linear": [107, 276],
"cloglog": [86, 230],
"log": [107, 332],
"asinsqrt": [104, 276],
"logit": [104, 230]
}
dfa = bmt[bmt.Group == "ALL"]
cur_dir = os.path.dirname(os.path.abspath(__file__))
fp = os.path.join(cur_dir, '#1lab_results', 'bmt_results.csv')
rslt = pd.read_csv(fp)
sf = SurvfuncRight(dfa["T"].values, dfa.Status.values)
assert_allclose(sf.surv_times, rslt.t)
assert_allclose(sf.surv_prob, rslt.s, atol=1e-4, rtol=1e-4)
assert_allclose(sf.surv_prob_se, rslt.se, atol=1e-4, rtol=1e-4)
for method in "linear", "cloglog", "log", "logit", "asinsqrt":
lcb, ucb = sf.quantile_ci(0.25, method=method)
assert_allclose(cb[method], np.r_[lcb, ucb])
def test_kernel_survfunc3():
# cases with tied times
n = 100
np.random.seed(3434)
x = np.random.normal(size=(n, 3))
time = np.random.randint(0, 10, size=n)
status = np.random.randint(0, 2, size=n)
SurvfuncRight(time, status, exog=x, bw_factor=10000)
SurvfuncRight(time, status, exog=x, bw_factor=np.r_[10000, 10000])
def test_survfunc_entry_2():
# entry = 0 is equivalent to no entry time
times = np.r_[1, 3, 3, 5, 5, 7, 7, 8, 8, 9, 10, 10]
status = np.r_[1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1]
entry = np.r_[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
sf = SurvfuncRight(times, status, entry=entry)
sf0 = SurvfuncRight(times, status)
assert_allclose(sf.n_risk, sf0.n_risk)
assert_allclose(sf.surv_times, sf0.surv_times)
assert_allclose(sf.surv_prob, sf0.surv_prob)
assert_allclose(sf.surv_prob_se, sf0.surv_prob_se)
def test_kernel_survfunc2():
# Check that when bandwidth is very large, the kernel procedure
# agrees with standard KM. (Note: the results do not agree
# perfectly when there are tied times).
n = 100
np.random.seed(3434)
x = np.random.normal(size=(n, 3))
time = np.random.uniform(0, 10, size=n)
status = np.random.randint(0, 2, size=n)
resultkm = SurvfuncRight(time, status)
result = SurvfuncRight(time, status, exog=x, bw_factor=10000)
assert_allclose(resultkm.surv_times, result.surv_times)
assert_allclose(resultkm.surv_prob, result.surv_prob, rtol=1e-6, atol=1e-6)
def test_simultaneous_cb():
# The exact numbers here are regression tests, but they are close
# to page 103 of Klein and Moeschberger.
df = bmt.loc[bmt["Group"] == "ALL", :]
sf = SurvfuncRight(df["T"], df["Status"])
lcb1, ucb1 = sf.simultaneous_cb(transform="log")
lcb2, ucb2 = sf.simultaneous_cb(transform="arcsin")
ti = sf.surv_times.tolist()
ix = [ti.index(x) for x in (110, 122, 129, 172)]
assert_allclose(lcb1[ix], np.r_[0.43590582, 0.42115592, 0.4035897, 0.38785927])
assert_allclose(ucb1[ix], np.r_[0.93491636, 0.89776803, 0.87922239, 0.85894181])
assert_allclose(lcb2[ix], np.r_[0.52115708, 0.48079378, 0.45595321, 0.43341115])
assert_allclose(ucb2[ix], np.r_[0.96465636, 0.92745068, 0.90885428, 0.88796708])
def test_survfunc2():
# Test where some times have no events.
sr = SurvfuncRight(ti2, st2)
assert_allclose(sr.surv_prob, surv_prob2, atol=1e-5, rtol=1e-5)
assert_allclose(sr.surv_prob_se, surv_prob_se2, atol=1e-5, rtol=1e-5)
assert_allclose(sr.surv_times, times2)
assert_allclose(sr.n_risk, n_risk2)
assert_allclose(sr.n_events, n_events2)
def test_survfunc1():
# Test where all times have at least 1 event.
sr = SurvfuncRight(ti1, st1)
assert_allclose(sr.surv_prob, surv_prob1, atol=1e-5, rtol=1e-5)
assert_allclose(sr.surv_prob_se, surv_prob_se1, atol=1e-5, rtol=1e-5)
assert_allclose(sr.surv_times, times1)
assert_allclose(sr.n_risk, n_risk1)
assert_allclose(sr.n_events, n_events1)
def test_plot_km():
if pdf_output:
from matplotlib.backends.backend_pdf import PdfPages
pdf = PdfPages("test_survfunc.pdf")
else:
pdf = None
sr1 = SurvfuncRight(ti1, st1)
sr2 = SurvfuncRight(ti2, st2)
fig = plot_survfunc(sr1)
close_or_save(pdf, fig)
fig = plot_survfunc(sr2)
close_or_save(pdf, fig)
fig = plot_survfunc([sr1, sr2])
close_or_save(pdf, fig)
# Plot the SAS BMT data
gb = bmt.groupby("Group")
sv = []
for g in gb:
s0 = SurvfuncRight(g[1]["T"], g[1]["Status"], title=g[0])
sv.append(s0)
fig = plot_survfunc(sv)
ax = fig.get_axes()[0]
ax.set_position([0.1, 0.1, 0.64, 0.8])
ha, lb = ax.get_legend_handles_labels()
fig.legend([ha[k] for k in (0, 2, 4)],
[lb[k] for k in (0, 2, 4)],
'center right')
close_or_save(pdf, fig)
# Simultaneous CB for BMT data
ii = bmt.Group == "ALL"
sf = SurvfuncRight(bmt.loc[ii, "T"], bmt.loc[ii, "Status"])
fig = sf.plot()
ax = fig.get_axes()[0]
ax.set_position([0.1, 0.1, 0.64, 0.8])
ha, lb = ax.get_legend_handles_labels()
lcb, ucb = sf.simultaneous_cb(transform="log")
plt.fill_between(sf.surv_times, lcb, ucb, color="lightgrey")
lcb, ucb = sf.simultaneous_cb(transform="arcsin")
plt.plot(sf.surv_times, lcb, color="darkgrey")
plt.plot(sf.surv_times, ucb, color="darkgrey")
plt.plot(sf.surv_times, sf.surv_prob - 2*sf.surv_prob_se, color="red")
plt.plot(sf.surv_times, sf.surv_prob + 2*sf.surv_prob_se, color="red")
plt.xlim(100, 600)
close_or_save(pdf, fig)
if pdf_output:
pdf.close()
def test_simultaneous_cb():
# The exact numbers here are regression tests, but they are close
# to page 103 of Klein and Moeschberger.
df = bmt.loc[bmt["Group"] == "ALL", :]
sf = SurvfuncRight(df["T"], df["Status"])
lcb1, ucb1 = sf.simultaneous_cb(transform="log")
lcb2, ucb2 = sf.simultaneous_cb(transform="arcsin")
ti = sf.surv_times.tolist()
ix = [ti.index(x) for x in (110, 122, 129, 172)]
assert_allclose(lcb1[ix], np.r_[0.43590582, 0.42115592, 0.4035897,
0.38785927])
assert_allclose(ucb1[ix], np.r_[0.93491636, 0.89776803, 0.87922239,
0.85894181])
assert_allclose(lcb2[ix], np.r_[0.52115708, 0.48079378, 0.45595321,
0.43341115])
assert_allclose(ucb2[ix], np.r_[0.96465636, 0.92745068, 0.90885428,
0.88796708])
def test_weights2():
# tm = c(1, 3, 5, 6, 7, 2, 4, 6, 8, 10)
# st = c(1, 1, 0, 1, 1, 1, 1, 0, 1, 1)
# wt = c(1, 1, 1, 1, 1, 2, 2, 2, 2, 2)
# library(survival)
# sf =s urvfit(Surv(tm, st) ~ 1, weights=wt, err='tsiatis')
tm = np.r_[1, 3, 5, 6, 7, 2, 4, 6, 8, 10]
st = np.r_[1, 1, 0, 1, 1, 1, 1, 0, 1, 1]
wt = np.r_[1, 1, 1, 1, 1, 2, 2, 2, 2, 2]
tm0 = np.r_[1, 3, 5, 6, 7, 2, 4, 6, 8, 10, 2, 4, 6, 8, 10]
st0 = np.r_[1, 1, 0, 1, 1, 1, 1, 0, 1, 1, 1, 1, 0, 1, 1]
sf0 = SurvfuncRight(tm, st, freq_weights=wt)
sf1 = SurvfuncRight(tm0, st0)
assert_allclose(sf0.surv_times, sf1.surv_times)
assert_allclose(sf0.surv_prob, sf1.surv_prob)
assert_allclose(
sf0.surv_prob_se,
np.r_[0.06666667, 0.1210311, 0.14694547, 0.19524829, 0.23183377,
0.30618115, 0.46770386, 0.84778942])
def test_kernel_survfunc1():
# Regression test
n = 100
np.random.seed(3434)
x = np.random.normal(size=(n, 3))
time = np.random.uniform(size=n)
status = np.random.randint(0, 2, size=n)
result = SurvfuncRight(time, status, exog=x)
timex = np.r_[0.30721103, 0.0515439, 0.69246897, 0.16446079, 0.31308528]
sprob = np.r_[0.98948277, 0.98162275, 0.97129237, 0.96044668, 0.95030368]
assert_allclose(result.time[0:5], timex)
assert_allclose(result.surv_prob[0:5], sprob)
def test_incidence2():
# Check that the cumulative incidence functions for all competing
# risks sum to the complementary survival function.
np.random.seed(2423)
n = 200
time = -np.log(np.random.uniform(size=n))
status = np.random.randint(0, 3, size=n)
ii = np.argsort(time)
time = time[ii]
status = status[ii]
ci = CumIncidenceRight(time, status)
statusa = 1 * (status >= 1)
sf = SurvfuncRight(time, statusa)
x = 1 - sf.surv_prob
y = (ci.cinc[0] + ci.cinc[1])[np.flatnonzero(statusa)]
assert_allclose(x, y)
def test_survfunc_entry_3():
# times = c(1, 2, 5, 6, 6, 6, 6, 6, 9)
# status = c(0, 0, 1, 1, 1, 0, 1, 1, 0)
# entry = c(0, 1, 1, 2, 2, 2, 3, 4, 4)
# sv = Surv(entry, times, event=status)
# sdf = survfit(coxph(sv ~ 1), type='kaplan-meier')
times = np.r_[1, 2, 5, 6, 6, 6, 6, 6, 9]
status = np.r_[0, 0, 1, 1, 1, 0, 1, 1, 0]
entry = np.r_[0, 1, 1, 2, 2, 2, 3, 4, 4]
sf = SurvfuncRight(times, status, entry=entry)
assert_allclose(sf.n_risk, np.r_[7, 6])
assert_allclose(sf.surv_times, np.r_[5, 6])
assert_allclose(sf.surv_prob, np.r_[0.857143, 0.285714], atol=1e-5)
assert_allclose(sf.surv_prob_se, np.r_[0.13226, 0.170747], atol=1e-5)
def test_weights1():
# tm = c(1, 3, 5, 6, 7, 8, 8, 9, 3, 4, 1, 3, 2)
# st = c(1, 1, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0)
# wt = c(1, 2, 3, 2, 3, 1, 2, 1, 1, 2, 2, 3, 1)
# library(survival)
# sf = survfit(Surv(tm, st) ~ 1, weights=wt, err='tsiatis')
tm = np.r_[1, 3, 5, 6, 7, 8, 8, 9, 3, 4, 1, 3, 2]
st = np.r_[1, 1, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0]
wt = np.r_[1, 2, 3, 2, 3, 1, 2, 1, 1, 2, 2, 3, 1]
sf = SurvfuncRight(tm, st, freq_weights=wt)
assert_allclose(sf.surv_times, np.r_[1, 3, 6, 7, 9])
assert_allclose(sf.surv_prob, np.r_[0.875, 0.65625, 0.51041667, 0.29166667,
0.])
assert_allclose(
sf.surv_prob_se, np.r_[0.07216878, 0.13307266, 0.20591185, 0.3219071,
1.05053519])
def test_survfunc_entry_1():
# times = c(1, 3, 3, 5, 5, 7, 7, 8, 8, 9, 10, 10)
# status = c(1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1)
# entry = c(0, 1, 1, 2, 2, 2, 3, 4, 4, 4, 4, 0)
# sv = Surv(entry, times, event=status)
# sdf = survfit(coxph(sv ~ 1), type='kaplan-meier')
times = np.r_[1, 3, 3, 5, 5, 7, 7, 8, 8, 9, 10, 10]
status = np.r_[1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1]
entry = np.r_[0, 1, 1, 2, 2, 2, 3, 4, 4, 4, 4, 0]
sf = SurvfuncRight(times, status, entry=entry)
assert_allclose(sf.n_risk, np.r_[2, 6, 9, 7, 5, 3, 2])
assert_allclose(sf.surv_times, np.r_[1, 3, 5, 7, 8, 9, 10])
assert_allclose(sf.surv_prob, np.r_[
0.5000, 0.4167, 0.3241, 0.2778, 0.2222, 0.1481, 0.0741],
atol=1e-4)
assert_allclose(sf.surv_prob_se, np.r_[
0.3536, 0.3043, 0.2436, 0.2132, 0.1776, 0.1330, 0.0846],
atol=1e-4)
def cli():
parser = argparse.ArgumentParser(
description='GAP - Git Activity Predictor')
parser.add_argument('paths',
metavar='PATH',
type=str,
nargs='*',
default=['.'],
help='Paths to one or more git repositories')
parser.add_argument(
'--date',
type=lambda d: dateutil.parser.parse(d).date(),
required=False,
default=datetime.date.today(),
help='Date used for predictions (default to current date)')
parser.add_argument('--obs',
type=int,
required=False,
default=20,
help='Number of observations to consider')
parser.add_argument('--probs',
metavar='PROB',
type=float,
nargs='*',
required=False,
default=[0.5, 0.6, 0.7, 0.8, 0.9],
help='Probabilities to output, strictly in [0,1].')
parser.add_argument(
'--limit',
type=int,
required=False,
default=30,
help=
'Limit contributors to the one that were active at least once during the last x days (default 30)'
)
parser.add_argument(
'--mapping',
type=str,
nargs='?',
help=
'Mapping file to merge identities. This file must be a csv file where each line contains two values: the name to be merged, and the corresponding identity. Use "IGNORE" as identity to ignore specific names.'
)
parser.add_argument('--branches',
metavar='BRANCH',
type=str,
nargs='*',
default=list(),
help='Git branches to analyse (default to all).')
parser.add_argument(
'--as-dates',
dest='as_dates',
action='store_true',
help=
'Express predictions using dates instead of time differences in days')
group = parser.add_mutually_exclusive_group()
group.add_argument('--text',
action='store_true',
help='Print results as text.')
group.add_argument('--csv',
action='store_true',
help='Print results as csv.')
group.add_argument('--json',
action='store_true',
help='Print results as json.')
group.add_argument(
'--plot',
nargs='?',
const=True,
help='Export results to a plot. Filepath can be optionaly specified.')
args = parser.parse_args()
# Default plot location
if args.plot is True:
args.plot = str(args.date) + '.pdf'
# Default to text if not other option is provided
if not args.csv and not args.json and not args.plot:
args.text = True
# Identity mapping
if args.mapping:
d = pandas.read_csv(args.mapping, names=['source', 'target'])
mapping = {r.source: r.target for r in d.itertuples()}
else:
mapping = {}
raw_data = dict() # author -> dates of activity
# Get data from git
for path in args.paths:
try:
repo = git.Repo(path)
except Exception as e: # Must be refined
print('Unable to access repository {} ({}:{})'.format(
path, e.__class__.__name__, e))
sys.exit()
# Default branches
if len(args.branches) == 0:
commits = repo.iter_commits('--all')
else:
commits = repo.iter_commits(' '.join(args.branches))
for commit in commits:
try:
author = commit.author.name
identity = mapping.get(author, author)
if author.lower() != 'ignore' and identity.lower() == 'ignore':
continue
date = datetime.date.fromtimestamp(commit.authored_date)
raw_data.setdefault(identity, []).append(date)
except Exception as e:
print('Unable to read commit ({}: {}): {}'.format(
e.__class__.__name__, e, commit))
# Compute durations and apply model
data = [] # (author, past activities, predicted durations)
for author, commits in raw_data.items():
commits = sorted([e for e in commits if e <= args.date])
durations = dates_to_duration(commits, window_size=args.obs)
if len(durations) >= args.obs:
# Currently implemented with no censor
surv = SurvfuncRight(durations, [1] * len(durations))
predictions = [surv.quantile(p) for p in args.probs]
last_day = commits[-1]
if last_day >= args.date - datetime.timedelta(args.limit):
data.append((
author,
commits,
predictions,
))
# Prepare dataframe
df = pandas.DataFrame(index=set([a for a, c, p in data]),
columns=['last'] + args.probs)
if len(df) == 0:
print(
'No author has {} observations and was active at least once during the last {} days'
.format(args.obs, args.limit))
sys.exit()
df.index.name = 'author'
if not args.plot:
for author, commits, predictions in data:
last = commits[-1]
if args.as_dates:
df.at[author, 'last'] = last
else:
df.at[author, 'last'] = (last - args.date).days
for prob, p in zip(args.probs, predictions):
if args.as_dates:
df.at[author,
prob] = last + datetime.timedelta(days=int(p))
else:
df.at[author,
prob] = (last + datetime.timedelta(days=int(p)) -
args.date).days
df = df.sort_values(['last'] + args.probs,
ascending=[False] + [True] * len(args.probs))
df = df.astype(str)
if args.text:
pandas.set_option('expand_frame_repr', False)
pandas.set_option('display.max_columns', 999)
print(df)
elif args.csv:
print(df.to_csv())
elif args.json:
print(df.to_json(orient='index'))
else:
# Because of plotnine's way of initializing matplotlib
import warnings
warnings.filterwarnings("ignore")
VIEW_LIMIT = 28
activities = [
] # List of (author, day) where day is a delta w.r.t. given date
forecasts = [
] # List of (author, from_day, to_day, p) where probability p
# applies between from_day and to_day (delta w.r.t. given date)
for author, commits, predictions in data:
last = (commits[-1] - args.date).days
for e in commits:
activities.append((author, (e - args.date).days))
previous = previous_previous = 0
for d, p in zip(predictions, args.probs):
if d > previous:
forecasts.append((author, last + previous, last + d, p))
previous_previous = previous
previous = d
else:
forecasts.append(
(author, last + previous_previous, last + d, p))
activities = pandas.DataFrame(columns=['author', 'day'],
data=activities)
forecasts = pandas.DataFrame(columns=['author', 'fromd', 'tod', 'p'],
data=forecasts)
plot = (p9.ggplot(p9.aes(y='author')) + p9.geom_segment(
p9.aes('day - 0.5', 'author', xend='day + 0.5', yend='author'),
data=activities,
size=4,
color='orange',
) + p9.geom_segment(
p9.aes('fromd + 0.5',
'author',
xend='tod + 0.5',
yend='author',
alpha='factor(p)'),
data=forecasts.sort_values('p').drop_duplicates(
['author', 'fromd', 'tod'], keep='last'),
size=4,
color='steelblue',
) + p9.geom_vline(
xintercept=0,
color='r', alpha=0.5, linetype='dashed') + p9.scale_x_continuous(
name=' << past days {:^20} future days >>'.format(
str(args.date)),
breaks=range(-VIEW_LIMIT // 7 * 7,
(VIEW_LIMIT // 7 * 7) + 1, 7),
minor_breaks=6) + p9.scale_y_discrete(
name='',
limits=activities.sort_values(
'day', ascending=False)['author'].unique()) +
p9.scale_alpha_discrete(range=(0.2, 1), name=' ') +
p9.coord_cartesian(xlim=(-VIEW_LIMIT, VIEW_LIMIT)) +
p9.theme_matplotlib() + p9.theme(
figure_size=(6, 4 * activities['author'].nunique() / 15)))
fig = plot.draw()
fig.savefig(args.plot, bbox_inches='tight')
print('Plot exported to {}'.format(args.plot))