def plotter(key, sims, ax, label='', ylabel='', low_q=0.05, high_q=0.95, startday=None):
color = cv.get_colors()[key.split('_')[1]]
ys = []
for s in sims:
ys.append(s.results[key].values)
yarr = np.array(ys)
best = pl.median(yarr, axis=0)
low = pl.quantile(yarr, q=low_q, axis=0)
high = pl.quantile(yarr, q=high_q, axis=0)
sim = sims[0]
tvec = np.arange(len(best))
fill_label = None
pl.fill_between(tvec, low, high, facecolor=color, alpha=0.2, label=fill_label)
pl.plot(tvec, best, c=color, label=label, lw=4, alpha=1.0)
sc.setylim()
datemarks = pl.array([sim.day('2020-03-01'),sim.day('2020-05-01'),sim.day('2020-07-01'),
sim.day('2020-09-01')])
ax.set_xticks(datemarks)
pl.ylabel(ylabel)
return
def plotter(key, sims, ax, label='', ylabel='', low_q=0.05, high_q=0.95, subsample=2):
which = key.split('_')[1]
try:
color = cv.get_colors()[which]
except:
color = [0.5,0.5,0.5]
ys = []
for s in sims:
ys.append(s.results[key].values)
yarr = np.array(ys)
best = pl.median(yarr, axis=0)
low = pl.quantile(yarr, q=low_q, axis=0)
high = pl.quantile(yarr, q=high_q, axis=0)
tvec = np.arange(len(best))
# tempsim = cv.Sim(datafile='../UK_Covid_cases_january03.xlsx')
# sim = sims[0]
# if key in tempsim.data:
# data_t = np.array((tempsim.data.index-sim['start_day'])/np.timedelta64(1,'D'))
# inds = np.arange(0, len(data_t), subsample)
# data = tempsim.data[key][inds]
# pl.plot(data_t[inds], data, 'd', c=color, markersize=10, alpha=0.5, label='Data')
fill_label = None
end = None
start = 2 if key == 'r_eff' else 0
pl.fill_between(tvec[start:end], low[start:end], high[start:end], facecolor=color, alpha=0.2, label=fill_label)
pl.plot(tvec[start:end], best[start:end], c=color, label=label, lw=4, alpha=1.0)
sc.setylim()
datemarks = pl.array([sim.day('2020-03-01'),sim.day('2020-06-01'),
sim.day('2020-09-01'),sim.day('2020-09-01'),
sim.day('2020-12-01'),sim.day('2021-03-01'),
sim.day('2021-05-01')])
ax.set_xticks(datemarks)
pl.ylabel(ylabel)
return
"75-79", "80-84", "85-89", "90+"
] #["0-29", "30-54", "55+"]
deaths = raw_deaths
pos = raw_pos
# From the model
mposlist, mdeathlist = [], []
for hists in agehists:
mposlist.append(hists['diagnosed'])
mdeathlist.append(hists['dead'])
mdeatharr = np.array(mdeathlist)
mposarr = np.array(mposlist)
low_q = 0.1
high_q = 0.9
raw_mdbest = pl.mean(mdeatharr, axis=0)
raw_mdlow = pl.quantile(mdeatharr, q=low_q, axis=0)
raw_mdhigh = pl.quantile(mdeatharr, q=high_q, axis=0)
raw_mpbest = pl.mean(mposarr, axis=0)
raw_mplow = pl.quantile(mposarr, q=low_q, axis=0)
raw_mphigh = pl.quantile(mposarr, q=high_q, axis=0)
if aggregate:
mpbest = [
raw_mpbest[0:6].sum(), raw_mpbest[6:13].sum(),
raw_mpbest[13:16].sum(), raw_mpbest[16:].sum()
]
mplow = [
raw_mplow[0:6].sum(), raw_mplow[6:13].sum(),
raw_mplow[13:16].sum(), raw_mplow[16:].sum()
]
mphigh = [
def run(self, debug=False, keep_people=False, verbose=None, **kwargs):
'''
Run the actual scenarios
Args:
debug (bool): if True, runs a single run instead of multiple, which makes debugging easier
verbose (int): level of detail to print, passed to sim.run()
kwargs (dict): passed to multi_run() and thence to sim.run()
Returns:
None (modifies Scenarios object in place)
'''
if verbose is None:
verbose = self['verbose']
def print_heading(string):
''' Choose whether to print a heading, regular text, or nothing '''
if verbose >= 2:
sc.heading(string)
elif verbose == 1:
print(string)
return
reskeys = self.result_keys() # Shorten since used extensively
# Loop over scenarios
for scenkey, scen in self.scenarios.items():
scenname = scen['name']
scenpars = scen['pars']
# This is necessary for plotting, and since self.npts is defined prior to run
if 'n_days' in scenpars.keys():
errormsg = 'Scenarios cannot be run with different numbers of days; set via basepars instead'
raise ValueError(errormsg)
# Create and run the simulations
print_heading(f'Multirun for {scenkey}')
scen_sim = sc.dcp(self.base_sim)
scen_sim.label = scenkey
scen_sim.update_pars(scenpars)
run_args = dict(n_runs=self['n_runs'],
noise=self['noise'],
noisepar=self['noisepar'],
keep_people=keep_people,
verbose=verbose)
if debug:
print('Running in debug mode (not parallelized)')
run_args.pop(
'n_runs',
None) # Remove n_runs argument, not used for a single run
scen_sims = [single_run(scen_sim, **run_args, **kwargs)]
else:
scen_sims = multi_run(
scen_sim, **run_args,
**kwargs) # This is where the sims actually get run
# Process the simulations
print_heading(f'Processing {scenkey}')
scenraw = {}
for reskey in reskeys:
scenraw[reskey] = pl.zeros((self.npts, len(scen_sims)))
for s, sim in enumerate(scen_sims):
scenraw[reskey][:, s] = sim.results[reskey].values
scenres = sc.objdict()
scenres.best = {}
scenres.low = {}
scenres.high = {}
for reskey in reskeys:
scenres.best[reskey] = pl.median(
scenraw[reskey],
axis=1) # Changed from median to mean for smoother plots
scenres.low[reskey] = pl.quantile(scenraw[reskey],
q=self['quantiles']['low'],
axis=1)
scenres.high[reskey] = pl.quantile(scenraw[reskey],
q=self['quantiles']['high'],
axis=1)
for reskey in reskeys:
self.results[reskey][scenkey]['name'] = scenname
for blh in ['best', 'low', 'high']:
self.results[reskey][scenkey][blh] = scenres[blh][reskey]
self.sims[scenkey] = scen_sims
#%% Print statistics
if verbose:
sc.heading('Results for last day in each scenario:')
x = defaultdict(dict)
scenkeys = list(self.scenarios.keys())
for scenkey in scenkeys:
for reskey in reskeys:
val = self.results[reskey][scenkey].best[-1]
if reskey not in ['r_eff', 'doubling_time']:
val = int(val)
x[scenkey][reskey] = val
df = pd.DataFrame.from_dict(x).astype(object)
print(df)
print()
# Save details about the run
self._kept_people = keep_people
return
def plotter(key, sims, ax, ys=None, calib=False, label='', ylabel='', low_q=0.1, high_q=0.9):
which = key.split('_')[1]
try:
color = cv.get_colors()[which]
except:
color = [0.5,0.5,0.5]
if which == 'deaths':
color = [0.5,0.0,0.0]
if ys is None:
ys = []
for s in sims:
ys.append(s.results[key].values)
yarr = np.array(ys)
best = pl.median(yarr, axis=0) # Changed from median to mean for smoother plots
low = pl.quantile(yarr, q=low_q, axis=0)
high = pl.quantile(yarr, q=high_q, axis=0)
sim = sims[0] # For having a sim to refer to
# Formatting parameters
plot_args = sc.mergedicts({'lw': 3, 'alpha': 0.8})
fill_args = sc.mergedicts({'alpha': 0.2})
tvec = np.arange(len(best))
if calib:
if key == 'r_eff':
end = -2
else:
end = -1
else:
end = None
pl.fill_between(tvec[:end], low[:end], high[:end], facecolor=color, **fill_args)
pl.plot(tvec[:end], best[:end], c=color, label=label, **plot_args)
if key in sim.data:
data_t = np.array((sim.data.index-sim['start_day'])/np.timedelta64(1,'D'))
pl.plot(data_t, sim.data[key], 'o', c=color, markersize=10, label='Data')
if calib:
xlims = pl.xlim()
pl.xlim([13, xlims[1]-1])
else:
pl.xlim([0,94])
sc.setylim()
xmin,xmax = ax.get_xlim()
if calib:
ax.set_xticks(pl.arange(xmin+2, xmax, 7))
else:
ax.set_xticks(pl.arange(xmin+2, xmax, 7))
pl.ylabel(ylabel)
pl.legend(loc='upper left')
return
def plot_calibration(sims, date, do_save=0):
sim = sims[0] # For having a sim to refer to
# Draw plots
fig1_path = f'calibration_{date}_fig1.png'
fig2_path = f'calibration_{date}_fig2.png'
fig_args = sc.mergedicts({'figsize': (16, 14)})
axis_args = sc.mergedicts({'left': 0.10, 'bottom': 0.05, 'right': 0.95, 'top': 0.93, 'wspace': 0.25, 'hspace': 0.40})
# Handle input arguments -- merge user input with defaults
low_q = 0.1
high_q = 0.9
# Figure 1: Calibration
pl.figure(**fig_args)
pl.subplots_adjust(**axis_args)
pl.figtext(0.42, 0.95, 'Model calibration', fontsize=30)
#%% Figure 1, panel 1
ax = pl.subplot(4,1,1)
format_ax(ax, sim)
plotter('new_tests', sims, ax, calib=True, label='Number of tests per day', ylabel='Tests')
plotter('new_diagnoses', sims, ax, calib=True, label='Number of diagnoses per day', ylabel='Tests')
#%% Figure 1, panel 2
ax = pl.subplot(4,1,2)
format_ax(ax, sim)
plotter('cum_diagnoses', sims, ax, calib=True, label='Cumulative diagnoses', ylabel='People')
#%% Figure 1, panel 3
ax = pl.subplot(4,1,3)
format_ax(ax, sim)
plotter('cum_deaths', sims, ax, calib=True, label='Cumulative deaths', ylabel='Deaths')
#%% Figure 1, panels 4A and 4B
agehists = []
for s,sim in enumerate(sims):
agehist = sim['analyzers'][0]
if s == 0:
age_data = agehist.data
agehists.append(agehist.hists[-1])
x = age_data['age'].values
pos = age_data['cum_diagnoses'].values
death = age_data['cum_deaths'].values
# From the model
mposlist = []
mdeathlist = []
for hists in agehists:
mposlist.append(hists['diagnosed'])
mdeathlist.append(hists['dead'])
mposarr = np.array(mposlist)
mdeatharr = np.array(mdeathlist)
mpbest = pl.median(mposarr, axis=0)
mplow = pl.quantile(mposarr, q=low_q, axis=0)
mphigh = pl.quantile(mposarr, q=high_q, axis=0)
mdbest = pl.median(mdeatharr, axis=0)
mdlow = pl.quantile(mdeatharr, q=low_q, axis=0)
mdhigh = pl.quantile(mdeatharr, q=high_q, axis=0)
# Plotting
w = 4
off = 2
bins = x.tolist() + [100]
ax = pl.subplot(4,2,7)
c1 = [0.3,0.3,0.6]
c2 = [0.6,0.7,0.9]
xx = x+w-off
pl.bar(x-off,pos, width=w, label='Data', facecolor=c1)
pl.bar(xx, mpbest, width=w, label='Model', facecolor=c2)
for i,ix in enumerate(xx):
pl.plot([ix,ix], [mplow[i], mphigh[i]], c='k')
ax.set_xticks(bins[:-1])
pl.title('Diagnosed cases by age')
pl.xlabel('Age')
pl.ylabel('Cases')
pl.legend()
ax = pl.subplot(4,2,8)
c1 = [0.5,0.0,0.0]
c2 = [0.9,0.4,0.3]
pl.bar(x-off,death, width=w, label='Data', facecolor=c1)
pl.bar(x+w-off, mdbest, width=w, label='Model', facecolor=c2)
for i,ix in enumerate(xx):
pl.plot([ix,ix], [mdlow[i], mdhigh[i]], c='k')
ax.set_xticks(bins[:-1])
pl.title('Deaths by age')
pl.xlabel('Age')
pl.ylabel('Deaths')
pl.legend()
# Tidy up
if do_save:
cv.savefig(fig1_path)
# Figure 2: Projections
pl.figure(**fig_args)
pl.subplots_adjust(**axis_args)
pl.figtext(0.42, 0.95, 'Model estimates', fontsize=30)
#%% Figure 2, panel 1
ax = pl.subplot(4,1,1)
format_ax(ax, sim)
plotter('cum_infections', sims, ax,calib=True, label='Cumulative infections', ylabel='People')
plotter('cum_recoveries', sims, ax,calib=True, label='Cumulative recoveries', ylabel='People')
#%% Figure 2, panel 2
ax = pl.subplot(4,1,2)
format_ax(ax, sim)
plotter('n_infectious', sims, ax,calib=True, label='Number of active infections', ylabel='People')
plot_intervs(sim, labels=True)
#%% Figure 2, panel 3
ax = pl.subplot(4,1,3)
format_ax(ax, sim)
plotter('new_infections', sims, ax,calib=True, label='Infections per day', ylabel='People')
plotter('new_recoveries', sims, ax,calib=True, label='Recoveries per day', ylabel='People')
plot_intervs(sim)
#%% Figure 2, panels 4
ax = pl.subplot(4,1,4)
format_ax(ax, sim)
plotter('r_eff', sims, ax, calib=True, label='Effective reproductive number', ylabel=r'$R_{eff}$')
ylims = [0,4]
pl.ylim(ylims)
xlims = pl.xlim()
pl.plot(xlims, [1, 1], 'k')
plot_intervs(sim)
# Tidy up
if do_save:
cv.savefig(fig2_path)
return
def plotter(key,
sims,
ax,
ys=None,
calib=False,
label='',
ylabel='',
low_q=0.025,
high_q=0.975,
flabel=True,
startday=None,
subsample=2,
chooseseed=None):
which = key.split('_')[1]
try:
color = cv.get_colors()[which]
except:
color = [0.5, 0.5, 0.5]
if which == 'diagnoses':
color = [0.03137255, 0.37401, 0.63813918, 1.]
elif which == '':
color = [0.82400815, 0., 0., 1.]
if ys is None:
ys = []
for s in sims:
ys.append(s.results[key].values)
yarr = np.array(ys)
if chooseseed is not None:
best = sims[chooseseed].results[key].values
else:
best = pl.median(yarr, axis=0)
low = pl.quantile(yarr, q=low_q, axis=0)
high = pl.quantile(yarr, q=high_q, axis=0)
sim = sims[0] # For having a sim to refer to
tvec = np.arange(len(best))
if key in sim.data:
data_t = np.array(
(sim.data.index - sim['start_day']) / np.timedelta64(1, 'D'))
inds = np.arange(0, len(data_t), subsample)
pl.plot(data_t[inds],
sim.data[key][inds],
'd',
c=color,
markersize=15,
alpha=0.75,
label='Data')
start = None
if startday is not None:
start = sim.day(startday)
end = sim.day(calibration_end)
if flabel:
if which == 'infections':
fill_label = '95% projected interval'
else:
fill_label = '95% projected interval'
else:
fill_label = None
pl.fill_between(tvec[startday:end],
low[startday:end],
high[startday:end],
facecolor=color,
alpha=0.2,
label=fill_label)
pl.plot(tvec[startday:end],
best[startday:end],
c=color,
label=label,
lw=4,
alpha=1.0)
# Print some stats
if key == 'cum_infections':
print(
f'Estimated {which} on July 25: {best[sim.day("2020-07-25")]} (95%: {low[sim.day("2020-07-25")]}-{high[sim.day("2020-07-25")]})'
)
print(
f'Estimated {which} overall: {best[sim.day(calibration_end)]} (95%: {low[sim.day(calibration_end)]}-{high[sim.day(calibration_end)]})'
)
elif key == 'n_infectious':
peakday = sc.findnearest(best, max(best))
peakval = max(best)
print(
f'Estimated peak {which} on {sim.date(peakday)}: {peakval} (95%: {low[peakday]}-{high[peakday]})'
)
print(
f'Estimated {which} on last day: {best[sim.day(calibration_end)]} (95%: {low[sim.day(calibration_end)]}-{high[sim.day(calibration_end)]})'
)
elif key == 'cum_diagnoses':
print(
f'Estimated {which} overall: {best[sim.day(calibration_end)]} (95%: {low[sim.day(calibration_end)]}-{high[sim.day(calibration_end)]})'
)
sc.setylim()
xmin, xmax = ax.get_xlim()
if calib:
ax.set_xticks(pl.arange(xmin + 2, xmax, 28))
else:
ax.set_xticks(pl.arange(xmin + 2, xmax, 28))
pl.ylabel(ylabel)
datemarks = pl.array([
sim.day('2020-07-01'),
sim.day('2020-08-01'),
sim.day('2020-09-01'),
sim.day('2020-10-01')
]) * 1.
ax.set_xticks(datemarks)
return
def run(self,
keep_sims=False,
debug=False,
healthsystems=True,
verbose=None):
'''
Run the actual scenarios
Args:
keep_sims (bool): whether or not to store the actual Sim objects in the Scenarios object (NB, very large)
debug (bool): if True, runs a single run instead of multiple, which makes debugging easier
healthsystems (bool): whether or not to run a health systems analysis on the results
verbose (int): level of detail to print, passed to sim.run()
Returns:
None (modifies Scenarios object in place)
'''
if verbose is None:
verbose = self['verbose']
def print_heading(string):
''' Choose whether to print a heading, regular text, or nothing '''
if verbose >= 2:
sc.heading(string)
elif verbose == 1:
print(string)
return
reskeys = self.reskeys # Shorten since used extensively
# Loop over scenarios
for scenkey, scen in self.scenarios.items():
scenname = scen['name']
scenpars = scen['pars']
# This is necessary for plotting, and since self.npts is defined prior to run
if 'n_days' in scenpars.keys():
errormsg = 'Scenarios cannot be run with different numbers of days; set via basepars instead'
raise ValueError(errormsg)
# Create and run the simulations
print_heading(f'Multirun for {scenkey}')
scen_sim = sc.dcp(self.base_sim)
scen_sim.update_pars(scenpars)
run_args = dict(n_runs=self['n_runs'],
noise=self['noise'],
noisepar=self['noisepar'],
verbose=verbose)
if debug:
print('Running in debug mode (not parallelized)')
run_args.pop(
'n_runs',
None) # Remove n_runs argument, not used for a single run
scen_sims = [single_run(scen_sim, **run_args)]
else:
scen_sims = multi_run(
scen_sim,
**run_args) # This is where the sims actually get run
# Process the simulations
print_heading(f'Processing {scenkey}')
scenraw = {}
for reskey in reskeys:
scenraw[reskey] = pl.zeros((self.npts, len(scen_sims)))
for s, sim in enumerate(scen_sims):
scenraw[reskey][:, s] = sim.results[reskey].values
scenres = sc.objdict()
scenres.best = {}
scenres.low = {}
scenres.high = {}
for reskey in reskeys:
scenres.best[reskey] = pl.mean(
scenraw[reskey],
axis=1) # Changed from median to mean for smoother plots
scenres.low[reskey] = pl.quantile(scenraw[reskey],
q=self['quantiles']['low'],
axis=1)
scenres.high[reskey] = pl.quantile(scenraw[reskey],
q=self['quantiles']['high'],
axis=1)
for reskey in reskeys:
self.allres[reskey][scenkey]['name'] = scenname
for blh in ['best', 'low', 'high']:
self.allres[reskey][scenkey][blh] = scenres[blh][reskey]
if keep_sims:
if 'sims' not in self.allres:
self.allres['sims'] = sc.objdict()
self.allres['sims'][scenkey] = scen_sims
print(
'Note: saving sims, which may produce a large file! Estimated to be:'
)
sc.checkmem(
self.allres
) # Print a warning about how big the file is likely to be
#%% Print statistics
if verbose:
print('\nResults for final time point in each scenario:')
for reskey in reskeys:
print(f'\n{reskey}')
for scenkey in list(self.scenarios.keys()):
print(
f' {scenkey}: {self.allres[reskey][scenkey].best[-1]:0.0f}'
)
print() # Add a blank space
# Perform health systems analysis
if healthsystems:
self.hsys = cvhs.HealthSystem(self.allres)
self.hsys.analyze()
return
def estimate_Gap_statistics(self, nrefs):
masknans = pl.ma.masked_not_equal(self._X[:, 0], 0).mask
minvals = self._X[masknans, :].min(axis=0)
maxvals = self._X[masknans, :].max(axis=0)
meanvals = self._X[masknans, :].mean(axis=0)
stdvals = self._X[masknans, :].std(axis=0)
ref_Affinity = []
Dref = []
# Compute a random uniform reference distribution of features
# precompute Distances and affinities.
for i in range(nrefs):
random_X = pl.ones_like(self._X)
# random_X [:,0 ] =np.random.uniform (low = minvals[0] , high=maxvals[0], size=pl.int_( self._X.shape[0]/10 ) )
random_X[:, 1] = np.random.uniform(
low=pl.quantile(q=0.16, a=self._X[masknans, 1]),
high=pl.quantile(q=0.16, a=self._X[masknans, 1]),
size=pl.int_(self._X.shape[0]),
)
random_X[:, 0] = np.random.normal(loc=meanvals[0],
scale=stdvals[0],
size=pl.int_(self._X.shape[0]))
ref_D = self._metric.pairwise(random_X)
ref_D = pl.ma.fix_invalid(ref_D, fill_value=1.0).data
Dref.append(ref_D)
ref_Affinity.append(pairwise_kernels(ref_D, metric="precomputed"))
self.Gaps = pl.zeros(len(self.Kvals))
self.sd = self.Gaps * 0.0
self.W = self.Gaps * 0.0 # KL index
p = self._nfeat
for j, K in enumerate(self.Kvals):
if self.verbose:
print(f"Running with K={K} clusters")
self.clusters = AgglomerativeClustering(
n_clusters=K,
affinity="precomputed",
linkage="average",
connectivity=self.connectivity,
)
self.clusters.fit_predict(self._Affinity)
# estimate WCSS for the samples
W = self.get_WCSS(K, self.clusters.labels_, self._distance_matr)
self.W[j] = W
# estimate WCSS for random samples
ref_W = pl.zeros(nrefs)
for i in range(nrefs):
ref_clusters = AgglomerativeClustering(
n_clusters=K,
affinity="precomputed",
linkage="average",
connectivity=self.connectivity,
)
ref_clusters.fit_predict(ref_Affinity[i])
ref_W[i] = self.get_WCSS(K, ref_clusters.labels_, Dref[i])
self.sd[j] = np.std(np.log(ref_W)) * np.sqrt(1 + 1.0 / nrefs)
self.Gaps[j] = np.mean(np.log(ref_W)) - np.log(W)
## see section 4 of Tibishrani et al. http://web.stanford.edu/~hastie/Papers/gap.pdf
gaps_criterion = pl.array(
[self.Kvals[:-1], self.Gaps[:-1] - self.Gaps[1:] + self.sd[1:]])
mask = pl.array(gaps_criterion[1, :] >= 0)
return pl.int_(gaps_criterion[0, mask][0])
def plotter(key,
sims,
ax,
ys=None,
calib=False,
label='',
ylabel='',
low_q=0.025,
high_q=0.975,
flabel=True,
subsample=2):
''' Plot a single time series with uncertainty '''
which = key.split('_')[1]
try:
color = cv.get_colors()[which]
except:
color = [0.5, 0.5, 0.5]
if which == 'deaths':
color = [0.5, 0.0, 0.0]
if ys is None:
ys = []
for i, s in enumerate(sims):
if i < sims_cutoff:
ys.append(s.results[key].values)
yarr = np.array(ys)
best = pl.median(yarr,
axis=0) # Changed from median to mean for smoother plots
low = pl.quantile(yarr, q=low_q, axis=0)
high = pl.quantile(yarr, q=high_q, axis=0)
sim = sims[0] # For having a sim to refer to
tvec = np.arange(len(best))
data, data_t = None, None
if key in sim.data:
data_t = np.array(
(sim.data.index - sim['start_day']) / np.timedelta64(1, 'D'))
inds = np.arange(0, len(data_t), subsample)
data = sim.data[key][inds]
pl.plot(data_t[inds],
data,
'd',
c=color,
markersize=10,
alpha=0.5,
label='Data')
end = None
if flabel:
if which == 'infections':
fill_label = '95% predic-\ntion interval'
else:
fill_label = '95% prediction\ninterval'
else:
fill_label = None
# Trim the beginning for r_eff and actually plot
start = 2 if key == 'r_eff' else 0
pl.fill_between(tvec[start:end],
low[start:end],
high[start:end],
facecolor=color,
alpha=0.2,
label=fill_label)
pl.plot(tvec[start:end],
best[start:end],
c=color,
label=label,
lw=4,
alpha=1.0)
sc.setylim()
xmin, xmax = ax.get_xlim()
ax.set_xticks(np.arange(xmin, xmax, day_stride))
pl.ylabel(ylabel)
plotres[key] = sc.objdict(
dict(tvec=tvec,
best=best,
low=low,
high=high,
data=data,
data_t=data_t))
return
def plot():
# Create the figure
fig = pl.figure(num='Fig. 1: Calibration', figsize=(24, 20))
tx1, ty1 = 0.005, 0.97
tx2, ty2 = 0.545, 0.66
ty3 = 0.34
fsize = 40
pl.figtext(tx1, ty1, 'a', fontsize=fsize)
pl.figtext(tx1, ty2, 'b', fontsize=fsize)
pl.figtext(tx2, ty1, 'c', fontsize=fsize)
pl.figtext(tx2, ty2, 'd', fontsize=fsize)
pl.figtext(tx1, ty3, 'e', fontsize=fsize)
pl.figtext(tx2, ty3, 'f', fontsize=fsize)
#%% Fig. 1A: diagnoses
x0, y0, dx, dy = 0.055, 0.73, 0.47, 0.24
ax1 = pl.axes([x0, y0, dx, dy])
format_ax(ax1, base_sim)
plotter('cum_diagnoses',
sims,
ax1,
calib=True,
label='Model',
ylabel='Cumulative diagnoses')
pl.legend(loc='lower right', frameon=False)
#%% Fig. 1B: deaths
y0b = 0.42
ax2 = pl.axes([x0, y0b, dx, dy])
format_ax(ax2, base_sim)
plotter('cum_deaths',
sims,
ax2,
calib=True,
label='Model',
ylabel='Cumulative deaths')
pl.legend(loc='lower right', frameon=False)
#%% Fig. 1A-B inserts (histograms)
agehists = []
for s, sim in enumerate(sims):
agehist = sim['analyzers'][0]
if s == 0:
age_data = agehist.data
agehists.append(agehist.hists[-1])
# Observed data
x = age_data['age'].values
pos = age_data['cum_diagnoses'].values
death = age_data['cum_deaths'].values
# Model outputs
mposlist = []
mdeathlist = []
for hists in agehists:
mposlist.append(hists['diagnosed'])
mdeathlist.append(hists['dead'])
mposarr = np.array(mposlist)
mdeatharr = np.array(mdeathlist)
low_q = 0.025
high_q = 0.975
mpbest = pl.median(mposarr, axis=0)
mplow = pl.quantile(mposarr, q=low_q, axis=0)
mphigh = pl.quantile(mposarr, q=high_q, axis=0)
mdbest = pl.median(mdeatharr, axis=0)
mdlow = pl.quantile(mdeatharr, q=low_q, axis=0)
mdhigh = pl.quantile(mdeatharr, q=high_q, axis=0)
w = 4
off = 2
# Insets
x0s, y0s, dxs, dys = 0.11, 0.84, 0.17, 0.13
ax1s = pl.axes([x0s, y0s, dxs, dys])
c1 = [0.3, 0.3, 0.6]
c2 = [0.6, 0.7, 0.9]
xx = x + w - off
pl.bar(x - off, pos, width=w, label='Data', facecolor=c1)
pl.bar(xx, mpbest, width=w, label='Model', facecolor=c2)
for i, ix in enumerate(xx):
pl.plot([ix, ix], [mplow[i], mphigh[i]], c='k')
ax1s.set_xticks(np.arange(0, 81, 20))
pl.xlabel('Age')
pl.ylabel('Cases')
sc.boxoff(ax1s)
pl.legend(frameon=False, bbox_to_anchor=(0.7, 1.1))
y0sb = 0.53
ax2s = pl.axes([x0s, y0sb, dxs, dys])
c1 = [0.5, 0.0, 0.0]
c2 = [0.9, 0.4, 0.3]
pl.bar(x - off, death, width=w, label='Data', facecolor=c1)
pl.bar(x + w - off, mdbest, width=w, label='Model', facecolor=c2)
for i, ix in enumerate(xx):
pl.plot([ix, ix], [mdlow[i], mdhigh[i]], c='k')
ax2s.set_xticks(np.arange(0, 81, 20))
pl.xlabel('Age')
pl.ylabel('Deaths')
sc.boxoff(ax2s)
pl.legend(frameon=False)
sc.boxoff(ax1s)
#%% Fig. 1C: infections
x0, dx = 0.60, 0.38
ax3 = pl.axes([x0, y0, dx, dy])
format_ax(ax3, sim)
# Plot SCAN data
pop_size = 2.25e6
scan = pd.read_csv(scan_file)
for i, r in scan.iterrows():
label = "Data" if i == 0 else None
ts = np.mean(sim.day(r['since'], r['to']))
low = r['lower'] * pop_size
high = r['upper'] * pop_size
mean = r['mean'] * pop_size
ax3.plot([ts] * 2, [low, high], alpha=1.0, color='k', zorder=1000)
ax3.plot(ts,
mean,
'o',
markersize=7,
color='k',
alpha=0.5,
label=label,
zorder=1000)
# Plot simulation
plotter('cum_infections',
sims,
ax3,
calib=True,
label='Cumulative\ninfections\n(modeled)',
ylabel='Infections')
plotter('n_infectious',
sims,
ax3,
calib=True,
label='Active\ninfections\n(modeled)',
ylabel='Infections',
flabel=False)
pl.legend(loc='upper left', frameon=False)
pl.ylim([0, 130e3])
plot_intervs(sim)
#%% Fig. 1C: R_eff
ax4 = pl.axes([x0, y0b, dx, dy])
format_ax(ax4, sim, key='r_eff')
plotter('r_eff',
sims,
ax4,
calib=True,
label='$R_{eff}$ (modeled)',
ylabel=r'Effective reproduction number')
pl.axhline(1, linestyle='--', lw=3, c='k', alpha=0.5)
pl.legend(loc='upper right', frameon=False)
plot_intervs(sim)
#%% Fig. 1E
# Do the plotting
pl.subplots_adjust(left=0.04,
right=0.52,
bottom=0.03,
top=0.35,
wspace=0.12,
hspace=0.50)
for i, k in enumerate(keys):
eax = pl.subplot(2, 2, i + 1)
c1 = [0.2, 0.5, 0.8]
c2 = [1.0, 0.5, 0.0]
c3 = [0.1, 0.6, 0.1]
sns.kdeplot(df1[k],
shade=1,
linewidth=3,
label='',
color=c1,
alpha=0.5)
sns.kdeplot(df2[k],
shade=0,
linewidth=3,
label='',
color=c2,
alpha=0.5)
pl.title(mapping[k])
pl.xlabel('')
pl.yticks([])
if not i % 4:
pl.ylabel('Density')
yfactor = 1.3
yl = pl.ylim()
pl.ylim([yl[0], yl[1] * yfactor])
m1 = np.median(df1[k])
m2 = np.median(df2[k])
m1std = df1[k].std()
m2std = df2[k].std()
pl.axvline(m1, c=c1, ymax=0.9, lw=3, linestyle='--')
pl.axvline(m2, c=c2, ymax=0.9, lw=3, linestyle='--')
def fmt(lab, val, std=-1):
if val < 0.1:
valstr = f'{lab} = {val:0.4f}'
elif val < 1.0:
valstr = f'{lab} = {val:0.2f}±{std:0.2f}'
else:
valstr = f'{lab} = {val:0.1f}±{std:0.1f}'
if std < 0:
valstr = valstr.split('±')[0] # Discard STD if not used
return valstr
if k.startswith('bc'):
pl.xlim([0, 100])
elif k == 'beta':
pl.xlim([3, 5])
elif k.startswith('tn'):
pl.xlim([0, 50])
else:
raise Exception(f'Please assign key {k}')
xl = pl.xlim()
xfmap = dict(
beta=0.15,
bc_wc1=0.30,
bc_lf=0.35,
tn=0.55,
)
xf = xfmap[k]
x0 = xl[0] + xf * (xl[1] - xl[0])
ypos1 = yl[1] * 0.97
ypos2 = yl[1] * 0.77
ypos3 = yl[1] * 0.57
if k == 'beta': # Use 2 s.f. instead of 1
pl.text(x0, ypos1, f'M: {m1:0.2f} ± {m1std:0.2f}', c=c1)
pl.text(x0, ypos2, f'N: {m2:0.2f} ± {m2std:0.2f}', c=c2)
pl.text(x0,
ypos3,
rf'$\Delta$: {(m2-m1):0.2f} ± {(m1std+m2std):0.2f}',
c=c3)
else:
pl.text(x0, ypos1, f'M: {m1:0.1f} ± {m1std:0.1f}', c=c1)
pl.text(x0, ypos2, f'N: {m2:0.1f} ± {m2std:0.1f}', c=c2)
pl.text(x0,
ypos3,
rf'$\Delta$: {(m2-m1):0.1f} ± {(m1std+m2std):0.1f}',
c=c3)
sc.boxoff(ax=eax)
#%% Fig. 1F: SafeGraph
x0, y0c, dyc = 0.60, 0.03, 0.30
ax5 = pl.axes([x0, y0c, dx, dyc])
format_ax(ax5, sim, key='r_eff')
fn = safegraph_file
df = pd.read_csv(fn)
week = df['week']
days = sim.day(week.values.tolist())
s = df['p.tot.schools'].values * 100
w = df['p.tot.no.schools'].values * 100
# From Fig. 2
colors = sc.gridcolors(5)
wcolor = colors[3] # Work color/community
scolor = colors[1] # School color
pl.plot(days,
w,
'd-',
c=wcolor,
markersize=15,
lw=3,
alpha=0.9,
label='Workplace and\ncommunity mobility data')
pl.plot(days,
s,
'd-',
c=scolor,
markersize=15,
lw=3,
alpha=0.9,
label='School mobility data')
sc.setylim()
xmin, xmax = ax5.get_xlim()
ax5.set_xticks(np.arange(xmin, xmax, day_stride))
pl.ylabel('Relative mobility (%)')
pl.legend(loc='upper right', frameon=False)
plot_intervs(sim)
return fig
def plotter(key,
sims,
ax,
ys=None,
calib=False,
label='',
ylabel='',
low_q=0.025,
high_q=0.975,
flabel=True,
startday=None,
subsample=2,
chooseseed=None):
which = key.split('_')[1]
try:
color = cv.get_colors()[which]
except:
color = [0.5, 0.5, 0.5]
if which == 'diagnoses':
color = [0.03137255, 0.37401, 0.63813918, 1.]
elif which == '':
color = [0.82400815, 0., 0., 1.]
if ys is None:
ys = []
for s in sims:
ys.append(s.results[key].values)
yarr = np.array(ys)
if chooseseed is not None:
best = sims[chooseseed].results[key].values
else:
best = pl.median(yarr, axis=0)
low = pl.quantile(yarr, q=low_q, axis=0)
high = pl.quantile(yarr, q=high_q, axis=0)
sim = sims[0] # For having a sim to refer to
tvec = np.arange(len(best))
if key in sim.data:
data_t = np.array(
(sim.data.index - sim['start_day']) / np.timedelta64(1, 'D'))
inds = np.arange(0, len(data_t), subsample)
pl.plot(data_t[inds],
sim.data[key][inds],
'd',
c=color,
markersize=10,
alpha=0.5,
label='Data')
start = None
if startday is not None:
start = sim.day(startday)
end = sim.day('2021-03-31')
if flabel:
if which == 'infections':
fill_label = '95% projected interval'
else:
fill_label = '95% projected interval'
else:
fill_label = None
pl.fill_between(tvec[startday:end],
low[startday:end],
high[startday:end],
facecolor=color,
alpha=0.2,
label=fill_label)
pl.plot(tvec[startday:end],
best[startday:end],
c=color,
label=label,
lw=4,
alpha=1.0)
#for s in sims:
# pl.plot(tvec[startday:end], s.results[key].values[startday:end], c=[0.4, 0.4, 0.4], label=label, lw=1, alpha=0.7)
#sc.setylim()
datemarks = pl.array([
sim.day('2020-07-01'),
sim.day('2020-09-01'),
sim.day('2020-11-01'),
sim.day('2021-01-01')
]) * 1.
ax.set_xticks(datemarks)
pl.ylabel(ylabel)
return
