def get_version_pars(version, verbose=True):
'''
Function for loading parameters from the specified version.
Parameters will be loaded for Covasim 'as at' the requested version i.e. the
most recent set of parameters that is <= the requested version. Available
parameter values are stored in the regression folder. If parameters are available
for versions 1.3, and 1.4, then this function will return the following
- If parameters for version '1.3' are requested, parameters will be returned from '1.3'
- If parameters for version '1.3.5' are requested, parameters will be returned from '1.3', since
Covasim at version 1.3.5 would have been using the parameters defined at version 1.3.
- If parameters for version '1.4' are requested, parameters will be returned from '1.4'
Args:
version (str): the version to load parameters from
Returns:
Dictionary of parameters from that version
'''
# Construct a sorted list of available parameters based on the files in the regression folder
regression_folder = sc.thisdir(__file__, 'regression', aspath=True)
available_versions = [
x.stem.replace('pars_v', '') for x in regression_folder.iterdir()
if x.suffix == '.json'
]
available_versions = sorted(available_versions, key=LooseVersion)
# Find the highest parameter version that is <= the requested version
version_comparison = [
sc.compareversions(version, v) >= 0 for v in available_versions
]
try:
target_version = available_versions[sc.findlast(version_comparison)]
except IndexError:
errormsg = f"Could not find a parameter version that was less than or equal to '{version}'. Available versions are {available_versions}"
raise ValueError(errormsg)
# Load the parameters
pars = sc.loadjson(filename=regression_folder /
f'pars_v{target_version}.json',
folder=regression_folder)
if verbose:
print(f'Loaded parameters from {target_version}')
return pars
def plot():
fig = pl.figure(num='Fig. 2: Transmission dynamics', figsize=(20,14))
piey, tsy, r3y = 0.68, 0.50, 0.07
piedx, tsdx, r3dx = 0.2, 0.9, 0.25
piedy, tsdy, r3dy = 0.2, 0.47, 0.35
pie1x, pie2x = 0.12, 0.65
tsx = 0.07
dispx, cumx, sympx = tsx, 0.33+tsx, 0.66+tsx
ts_ax = pl.axes([tsx, tsy, tsdx, tsdy])
pie_ax1 = pl.axes([pie1x, piey, piedx, piedy])
pie_ax2 = pl.axes([pie2x, piey, piedx, piedy])
symp_ax = pl.axes([sympx, r3y, r3dx, r3dy])
disp_ax = pl.axes([dispx, r3y, r3dx, r3dy])
cum_ax = pl.axes([cumx, r3y, r3dx, r3dy])
off = 0.06
txtdispx, txtcumx, txtsympx = dispx-off, cumx-off, sympx-off+0.02
tsytxt = tsy+tsdy
r3ytxt = r3y+r3dy
labelsize = 40-wf
pl.figtext(txtdispx, tsytxt, 'a', fontsize=labelsize)
pl.figtext(txtdispx, r3ytxt, 'b', fontsize=labelsize)
pl.figtext(txtcumx, r3ytxt, 'c', fontsize=labelsize)
pl.figtext(txtsympx, r3ytxt, 'd', fontsize=labelsize)
#%% Fig. 2A -- Time series plot
layer_keys = list(sim.layer_keys())
layer_mapping = {k:i for i,k in enumerate(layer_keys)}
n_layers = len(layer_keys)
colors = sc.gridcolors(n_layers)
layer_counts = np.zeros((sim.npts, n_layers))
for source_ind, target_ind in tt.count_transmissions():
dd = tt.detailed[target_ind]
date = dd['date']
layer_num = layer_mapping[dd['layer']]
layer_counts[date, layer_num] += sim.rescale_vec[date]
mar12 = cv.date('2020-03-12')
mar23 = cv.date('2020-03-23')
mar12d = sim.day(mar12)
mar23d = sim.day(mar23)
labels = ['Household', 'School', 'Workplace', 'Community', 'LTCF']
for l in range(n_layers):
ts_ax.plot(sim.datevec, layer_counts[:,l], c=colors[l], lw=3, label=labels[l])
sc.setylim(ax=ts_ax)
sc.boxoff(ax=ts_ax)
ts_ax.set_ylabel('Transmissions per day')
ts_ax.set_xlim([sc.readdate('2020-01-18'), sc.readdate('2020-06-09')])
ts_ax.xaxis.set_major_formatter(mdates.DateFormatter('%b-%d'))
ts_ax.set_xticks([sim.date(d, as_date=True) for d in np.arange(0, sim.day('2020-06-09'), 14)])
ts_ax.legend(frameon=False, bbox_to_anchor=(0.85,0.1))
color = [0.2, 0.2, 0.2]
ts_ax.axvline(mar12, c=color, linestyle='--', alpha=0.4, lw=3)
ts_ax.axvline(mar23, c=color, linestyle='--', alpha=0.4, lw=3)
yl = ts_ax.get_ylim()
labely = yl[1]*1.015
ts_ax.text(mar12, labely, 'Schools close ', color=color, alpha=0.9, style='italic', horizontalalignment='center')
ts_ax.text(mar23, labely, ' Stay-at-home', color=color, alpha=0.9, style='italic', horizontalalignment='center')
#%% Fig. 2A inset -- Pie charts
pre_counts = layer_counts[0:mar12d, :].sum(axis=0)
post_counts = layer_counts[mar23d:, :].sum(axis=0)
pre_counts = pre_counts/pre_counts.sum()*100
post_counts = post_counts/post_counts.sum()*100
lpre = [
f'Household\n{pre_counts[0]:0.1f}%',
f'School\n{pre_counts[1]:0.1f}% ',
f'Workplace\n{pre_counts[2]:0.1f}% ',
f'Community\n{pre_counts[3]:0.1f}%',
f'LTCF\n{pre_counts[4]:0.1f}%',
]
lpost = [
f'Household\n{post_counts[0]:0.1f}%',
f'School\n{post_counts[1]:0.1f}%',
f'Workplace\n{post_counts[2]:0.1f}%',
f'Community\n{post_counts[3]:0.1f}%',
f'LTCF\n{post_counts[4]:0.1f}%',
]
pie_ax1.pie(pre_counts, colors=colors, labels=lpre, **pieargs)
pie_ax2.pie(post_counts, colors=colors, labels=lpost, **pieargs)
pie_ax1.text(0, 1.75, 'Transmissions by layer\nbefore schools closed', style='italic', horizontalalignment='center')
pie_ax2.text(0, 1.75, 'Transmissions by layer\nafter stay-at-home', style='italic', horizontalalignment='center')
#%% Fig. 2B -- histogram by overdispersion
# Process targets
n_targets = tt.count_targets(end_day=mar12)
# Handle bins
max_infections = n_targets.max()
edges = np.arange(0, max_infections+2)
# Analysis
counts = np.histogram(n_targets, edges)[0]
bins = edges[:-1] # Remove last bin since it's an edge
norm_counts = counts/counts.sum()
raw_counts = counts*bins
total_counts = raw_counts/raw_counts.sum()*100
n_bins = len(bins)
index = np.linspace(0, 100, len(n_targets))
sorted_arr = np.sort(n_targets)
sorted_sum = np.cumsum(sorted_arr)
sorted_sum = sorted_sum/sorted_sum.max()*100
change_inds = sc.findinds(np.diff(sorted_arr) != 0)
pl.set_cmap('Spectral_r')
sscolors = sc.vectocolor(n_bins)
width = 1.0
for i in range(n_bins):
disp_ax.bar(bins[i], total_counts[i], width=width, facecolor=sscolors[i])
disp_ax.set_xlabel('Number of transmissions per case')
disp_ax.set_ylabel('Proportion of transmissions (%)')
sc.boxoff()
disp_ax.set_xlim([0.5, 32.5])
disp_ax.set_xticks(np.arange(0, 32.5, 4))
sc.boxoff(ax=disp_ax)
dpie_ax = pl.axes([dispx+0.05, 0.20, 0.2, 0.2])
trans1 = total_counts[1:3].sum()
trans2 = total_counts[3:5].sum()
trans3 = total_counts[5:8].sum()
trans4 = total_counts[8:].sum()
labels = [
f'1-2:\n{trans1:0.0f}%',
f' 3-4:\n {trans2:0.0f}%',
f'5-7: \n{trans3:0.0f}%\n',
f'>7: \n{trans4:0.0f}%\n',
]
dpie_args = sc.mergedicts(pieargs, dict(labeldistance=1.2)) # Slightly smaller label distance
dpie_ax.pie([trans1, trans2, trans3, trans4], labels=labels, colors=sscolors[[0,4,7,12]], **dpie_args)
#%% Fig. 2C -- cumulative distribution function
rev_ind = 100 - index
n_change_inds = len(change_inds)
change_bins = bins[counts>0][1:]
for i in range(n_change_inds):
ib = int(change_bins[i])
ci = change_inds[i]
ici = index[ci]
sci = sorted_sum[ci]
color = sscolors[ib]
if i>0:
cim1 = change_inds[i-1]
icim1 = index[cim1]
scim1 = sorted_sum[cim1]
cum_ax.plot([icim1, ici], [scim1, sci], lw=4, c=color)
cum_ax.scatter([ici], [sci], s=150, zorder=50-i, c=[color], edgecolor='w', linewidth=0.2)
if ib<=6 or ib in [8, 10, 25]:
xoff = 5 - 2*(ib==1) + 3*(ib>=10) + 1*(ib>=20)
yoff = 2*(ib==1)
cum_ax.text(ici-xoff, sci+yoff, ib, fontsize=18-wf, color=color)
cum_ax.set_xlabel('Proportion of primary infections (%)')
cum_ax.set_ylabel('Proportion of transmissions (%)')
xmin = -2
ymin = -2
cum_ax.set_xlim([xmin, 102])
cum_ax.set_ylim([ymin, 102])
sc.boxoff(ax=cum_ax)
# Draw horizontal lines and annotations
ancol1 = [0.2, 0.2, 0.2]
ancol2 = sscolors[0]
ancol3 = sscolors[6]
i01 = sc.findlast(sorted_sum==0)
i20 = sc.findlast(sorted_sum<=20)
i50 = sc.findlast(sorted_sum<=50)
cum_ax.plot([xmin, index[i01]], [0, 0], '--', lw=2, c=ancol1)
cum_ax.plot([xmin, index[i20], index[i20]], [20, 20, ymin], '--', lw=2, c=ancol2)
cum_ax.plot([xmin, index[i50], index[i50]], [50, 50, ymin], '--', lw=2, c=ancol3)
# Compute mean number of transmissions for 80% and 50% thresholds
q80 = sc.findfirst(np.cumsum(total_counts)>20) # Count corresponding to 80% of cumulative infections (100-80)
q50 = sc.findfirst(np.cumsum(total_counts)>50) # Count corresponding to 50% of cumulative infections
n80, n50 = [sum(bins[q:]*norm_counts[q:]/norm_counts[q:].sum()) for q in [q80, q50]]
# Plot annotations
kw = dict(bbox=dict(facecolor='w', alpha=0.9, lw=0), fontsize=20-wf)
cum_ax.text(2, 3, f'{index[i01]:0.0f}% of infections\ndo not transmit', c=ancol1, **kw)
cum_ax.text(8, 23, f'{rev_ind[i20]:0.0f}% of infections cause\n80% of transmissions\n(mean: {n80:0.1f} per infection)', c=ancol2, **kw)
cum_ax.text(14, 53, f'{rev_ind[i50]:0.0f}% of infections cause\n50% of transmissions\n(mean: {n50:0.1f} per infection)', c=ancol3, **kw)
#%% Fig. 2D -- histogram by date of symptom onset
# Calculate
asymp_count = 0
symp_counts = {}
minind = -5
maxind = 15
for _, target_ind in tt.transmissions:
dd = tt.detailed[target_ind]
date = dd['date']
delta = sim.rescale_vec[date] # Increment counts by this much
if dd['s']:
if tt.detailed[dd['source']]['date'] <= date: # Skip dynamical scaling reinfections
sdate = dd['s']['date_symptomatic']
if np.isnan(sdate):
asymp_count += delta
else:
ind = int(date - sdate)
if ind not in symp_counts:
symp_counts[ind] = 0
symp_counts[ind] += delta
# Convert to an array
xax = np.arange(minind-1, maxind+1)
sympcounts = np.zeros(len(xax))
for i,val in symp_counts.items():
if i<minind:
ind = 0
elif i>maxind:
ind = -1
else:
ind = sc.findinds(xax==i)[0]
sympcounts[ind] += val
# Plot
total_count = asymp_count + sympcounts.sum()
sympcounts = sympcounts/total_count*100
presymp = sc.findinds(xax<=0)[-1]
colors = ['#eed15b', '#ee943a', '#c3211a']
asymp_frac = asymp_count/total_count*100
pre_frac = sympcounts[:presymp].sum()
symp_frac = sympcounts[presymp:].sum()
symp_ax.bar(xax[0]-2, asymp_frac, label='Asymptomatic', color=colors[0])
symp_ax.bar(xax[:presymp], sympcounts[:presymp], label='Presymptomatic', color=colors[1])
symp_ax.bar(xax[presymp:], sympcounts[presymp:], label='Symptomatic', color=colors[2])
symp_ax.set_xlabel('Days since symptom onset')
symp_ax.set_ylabel('Proportion of transmissions (%)')
symp_ax.set_xticks([minind-3, 0, 5, 10, maxind])
symp_ax.set_xticklabels(['Asymp.', '0', '5', '10', f'>{maxind}'])
sc.boxoff(ax=symp_ax)
spie_ax = pl.axes([sympx+0.05, 0.20, 0.2, 0.2])
labels = [f'Asymp-\ntomatic\n{asymp_frac:0.0f}%', f' Presymp-\n tomatic\n {pre_frac:0.0f}%', f'Symp-\ntomatic\n{symp_frac:0.0f}%']
spie_ax.pie([asymp_frac, pre_frac, symp_frac], labels=labels, colors=colors, **pieargs)
return fig
def apply(self, sim):
''' Perform vaccination '''
t = sim.t
if t < self.start_day:
return
elif self.end_day is not None and t > self.end_day:
return
# Check that there are still vaccines
rel_t = t - self.start_day
if rel_t < len(self.daily_vaccines):
n_vaccines = sc.randround(
self.daily_vaccines[rel_t] / sim.rescale_vec[t]
) # Correct for scaling that may be applied by rounding to the nearest number of tests
if not (n_vaccines and np.isfinite(n_vaccines)
): # If there are no doses today, abort early
return
else:
if sim.rescale_vec[t] != sim['pop_scale']:
raise RuntimeError('bad rescale time')
else:
return
vacc_probs = np.ones(
sim.n
) # Begin by assigning equal vaccine weight (converted to a probability) to everyone
# age priority
ppl = sim.people
priority_stage = sc.findlast(self.priority_days <= t)
not_target_inds = cv.true(
np.logical_or(ppl.age < self.age_priority[priority_stage],
ppl.dead))
vacc_probs[not_target_inds] = 0.0
# Handle scheduling
# first dose:
vacc_probs[self.vaccinations == 0] *= self.dose_priority[0]
# time between first and second dose:
no_dose = [
sim.t < (d[0] + self.delay) if len(d) > 0 else False
for d in self.vaccination_dates
]
vacc_probs[no_dose] *= 0
# time available for second dose:
second_dose = [
sim.t >= (d[0] + self.delay) if len(d) > 0 else False
for d in self.vaccination_dates
]
vacc_probs[second_dose] *= self.dose_priority[1]
# Don't give dose 2 people who have had more than 1
vacc_inds = cvu.true(self.vaccinations > 1)
vacc_probs[vacc_inds] = 0.0
# Now choose who gets vaccinated and vaccinate them
n_vaccines = min(
n_vaccines, (vacc_probs != 0).sum()
) # Don't try to vaccinate more people than have nonzero vaccination probability
all_vacc_inds = cvu.choose_w(probs=vacc_probs,
n=n_vaccines,
unique=True) # Choose who actually tests
sim.results['new_doses'][t] += len(all_vacc_inds)
# Did the vaccine take?
vacc_take_inds = all_vacc_inds[np.logical_or(
cvu.n_binomial(self.take_prob, len(all_vacc_inds)) &
(self.vaccinations[all_vacc_inds] == 0),
self.vaccine_take[all_vacc_inds] &
(self.vaccinations[all_vacc_inds] == 1))]
self.vaccine_take[vacc_take_inds] = True
# Calculate the effect per person
vacc_doses = self.vaccinations[
vacc_take_inds] # Calculate current doses
eff_doses = np.minimum(vacc_doses,
len(self.cumulative) -
1) # Convert to a valid index
vacc_eff = self.cumulative[
eff_doses] # Pull out corresponding effect sizes
rel_trans_eff = (1.0 - vacc_eff) + vacc_eff * self.rel_trans
rel_symp_eff = (1.0 - vacc_eff) + vacc_eff * self.rel_symp
# Apply the vaccine to people
# schedule dose effect
for intv in sim['interventions']:
if isinstance(intv, dose_scheduler):
schedule = {
'inds': vacc_take_inds,
'rel_trans':
self.orig_rel_trans[vacc_take_inds] * rel_trans_eff,
'symp_prob':
self.orig_symp_prob[vacc_take_inds] * rel_symp_eff
}
schedule_day = sim.t + self.dose_delay
if schedule_day not in intv.scheduler:
intv.scheduler.update({schedule_day: [schedule]})
else:
intv.scheduler[schedule_day].append(schedule)
self.vaccinations[all_vacc_inds] += 1
for v_ind in all_vacc_inds:
self.vaccination_dates[v_ind].append(sim.t)
return