def plot_sea_onset_comparison(run_name,
var,
ax,
signature_types=signature_types):
if run_name == 'obs':
seadata['MPB'] = ['MPB ($nT^4$)', obs_mpb_v, obs_mpb_t]
seadata['al'] = ['AL (nT)', sml, supermag_times]
ylabel, data, times = seadata[var]
artists, labels = plot_onset_sea(obs_signatures,
obs_threshold,
data,
times,
ylabel,
ax,
signature_types=signature_types)
else:
names = [runprops['name'] for runprops in run_properties]
runprops = run_properties[names.index(run_name)]
run_signatures = get_model_signature_lists(runprops, datadir=datadir)
from spacepy import datamodel as dm
auroral_inds = dm.fromHDF5(
os.path.join(
datadir, runprops['name'].replace('/', '').replace(' ', '_') +
'_auroral_inds.h5'))
al_time = [
datetime(2005, 1, 1, tzinfo=UTC) + timedelta(seconds=60 * m)
for m in range(0, 1440 * 31)
]
#try:
# mpbdata=loadmat(os.path.join(datadir,'John Haiducek - '+runprops['name'].replace('/','').replace(' ','_')+'_mag_grid_lat=33_mpb.mat'))
#except:
# raise
#mpb_t=[datetime(2005,1,1,tzinfo=UTC)+timedelta(seconds=m*60) for m in range(0,31*24*60)]
#mpb_v=mpbdata['mpb']
mpb_t, mpb_v = parse_index(os.path.join(datadir, 'mpb_index.txt'))
seadata['MPB'] = ['MPB ($nT^4$)', mpb_v, mpb_t]
seadata['al'] = ['AL (nT)', auroral_inds['AL'], al_time]
ylabel, data, times = seadata[var]
artists, labels = plot_onset_sea(run_signatures,
model_threshold,
data,
times,
ylabel,
ax,
signature_types=signature_types)
return artists, labels
def plot_skill_score_v_count_ratio(runprops,obs_thresholds=None,model_thresholds=None,skip_styles=0,show_circle=True):
sweep_shapes=['o','s','v','d','^','>']
run_signatures=get_model_signature_lists(runprops,datadir=datadir)
obs_thresholds,model_thresholds,substorm_bins=get_sweep_substorm_bins(obs_signatures,run_signatures,obs_thresholds=obs_thresholds,model_thresholds=model_thresholds)
true_positive,false_positive,false_negative,true_negative=get_counts(substorm_bins[:,:,0,:],substorm_bins[:,:,1,:],axis=2)
hit_rates=hit_rate(true_positive,false_positive,false_negative,true_negative)
false_alarm_rates=false_alarm_rate(true_positive,false_positive,false_negative,true_negative)
total_model_substorms=(true_positive+false_positive).astype(int)
total_obs_substorms=(true_positive+false_negative).astype(int)
skillscores=heidke_skill(true_positive,false_positive,false_negative,true_negative)
skillscore_ci_lower,skillscore_ci_upper=heidke_ci(substorm_bins[:,:,0,:],substorm_bins[:,:,1,:],axis=2)
skillscore_err_upper=skillscore_ci_upper-skillscores
skillscore_err_lower=-(skillscore_ci_lower-skillscores)
fig=plt.figure(figsize=[5.5,3.5])
lines=[]
for i in range(skip_styles):
plt.plot([],[])
for i in range(total_obs_substorms.shape[0]):
if obs_thresholds[i]==2.5:
plot_kwargs={
'linewidth':2.5,
'markersize':8,
'zorder':10
}
else:
plot_kwargs={}
line,caps,bars=plt.errorbar(total_model_substorms[i].astype(float)/total_obs_substorms[i],skillscores[i],[skillscore_err_upper[i],skillscore_err_lower[i]],linestyle='',marker=sweep_shapes[i+skip_styles],**plot_kwargs)
lines.append(line)
if show_circle:
i=np.where(np.array(obs_thresholds)==2.5)
j=np.where(np.array(model_thresholds)==2.5)
plt.plot(total_model_substorms[i,j].astype(float)/total_obs_substorms[i,j],skillscores[i,j],marker='o',markersize=14,markerfacecolor='none',zorder=15,markeredgecolor='k',markeredgewidth=2)
labels=['Obs. threshold={0:0.1f} ({1:d} events)'.format(obs_thresholds[i],total_obs_substorms[i,0]) for i in range(len(lines))]
plt.legend(lines,labels,loc='upper left')
plt.xscale('log')
plt.xlabel('$n_{model}/n_{obs}$')
plt.ylabel('Heidke skill score')
plt.axvline(1,color='k',alpha=0.5,zorder=-1)
plt.axhline(0,color='k',alpha=0.5,zorder=-1)
plt.xlim(0.003,40)
plt.ylim(-0.03,0.3)
plt.tight_layout()
return fig
def plot_roc(runprops):
run_signatures=get_model_signature_lists(runprops,datadir=datadir)
obs_thresholds,model_thresholds,substorm_bins=get_sweep_substorm_bins(obs_signatures,run_signatures)
true_positive,false_positive,false_negative,true_negative=get_counts(substorm_bins[:,:,0,:],substorm_bins[:,:,1,:],axis=2)
hit_rates=hit_rate(true_positive,false_positive,false_negative,true_negative)
false_alarm_rates=false_alarm_rate(true_positive,false_positive,false_negative,true_negative)
total_model_substorms=(true_positive+false_positive).astype(int)
total_obs_substorms=(true_positive+false_negative).astype(int)
skillscores=heidke_skill(true_positive,false_positive,false_negative,true_negative)
skillscore_ci_lower,skillscore_ci_upper=heidke_ci(substorm_bins[:,:,0,:],substorm_bins[:,:,1,:],axis=2)
skillscore_err_upper=skillscore_ci_upper-skillscores
skillscore_err_lower=-(skillscore_ci_lower-skillscores)
fig=plt.figure(figsize=(4.5,3.5))
plt.plot([0,1],color='k',marker='',alpha=0.6)
lines=[]
for i in range(false_alarm_rates.shape[0]):
if obs_thresholds[i]==2.5:
plot_kwargs={
'linewidth':2.5,
'markersize':8,
'zorder':10
}
else:
plot_kwargs={}
lines.extend(plt.plot(false_alarm_rates[i],hit_rates[i],linestyle='-',clip_on=False,**plot_kwargs))
i=np.where(obs_thresholds==2.5)
j=np.where(model_thresholds==2.5)
plt.plot(false_alarm_rates[i,j],hit_rates[i,j],marker='o',markersize=14,markerfacecolor='none',zorder=15,markeredgecolor='k',markeredgewidth=2)
plt.xlabel('Probability of False Detection')
plt.ylabel('Probability of Detection')
labels=['Obs. threshold={0:0.1f} ({1:d})'.format(obs_thresholds[i],total_obs_substorms[i,0]) for i in range(len(lines))]
plt.legend(lines,labels,loc='lower right')
namestr=runprops['name'].replace('/','').replace(' ','_')
plt.xlim(0,1)
plt.ylim(0,1)
plt.tight_layout()
plt.savefig('roc_curves.svg')
plt.close(fig)
def isi_subplot(ax, run_names, onset_type, bin_max=15):
fills = []
lines = []
labels = []
for run_name in run_names:
if run_name == 'obs':
signatures = get_obs_signature_lists(datadir=datadir)
threshold = obs_threshold
else:
runprops, = [
runprops for runprops in run_properties
if runprops['name'] == run_name
]
signatures = get_model_signature_lists(runprops, datadir=datadir)
threshold = model_threshold
if onset_type == 'all':
onsets = find_convolution_onsets(signatures,
threshold,
bandwidth=timedelta(0, 60 * 10))
#onsets=[(onset-datetime(2005,1,1,tzinfo=UTC)).total_seconds() for onset in onsets]
#print onsets
#substorms,onsets=find_substorms(signatures,threshold=1,return_times=True,signature_filters=['AL'])
#onsets=onsets.compressed()
#print onsets
else:
onsets = signatures.get(onset_type, [])
bins = np.linspace(0, bin_max, 100)
bw = 0.2
if len(onsets) > 3:
if run_name == 'obs':
color = 'LightSteelBlue'
else:
all_runnames = [
runprops['name'] for runprops in run_properties
]
irun = all_runnames.index(run_name)
import matplotlib
run_colors = matplotlib.rcParams['axes.prop_cycle'].by_key(
)['color']
color = run_colors[irun]
if run_name == 'obs' or onset_type == 'plasmoids':
show_ci = True
else:
show_ci = False
line, fill = plot_isi(ax, bins, onsets, color, bw, show_ci=show_ci)
lines.append(line)
fills.append(fill)
if run_name == 'obs':
labels.append('Observations')
else:
if len(run_properties) > 1:
labels.append(run_name)
else:
labels.append('MHD')
return zip(lines, fills), labels
def get_table_data(table_signatures, nsamples=32000):
obs_signatures = get_obs_signature_lists(datadir=datadir)
obs_substorms = find_substorms_convolution(obs_signatures, obs_threshold)
table_data = []
for runprops in run_properties:
table_data.append([])
run_signatures = get_model_signature_lists(runprops, datadir=datadir)
for signature in table_signatures:
if signature == 'All':
signature_run_substorms = find_substorms_convolution(
run_signatures, model_threshold)
signature_obs_substorms = obs_substorms
else:
if signature in run_signatures:
signature_run_substorms, keys = make_grid(
{signature: run_signatures[signature]})
signature_run_substorms = signature_run_substorms.flatten()
if signature in obs_signatures:
signature_obs_substorms, keys = make_grid(
{signature: obs_signatures[signature]})
signature_obs_substorms = signature_obs_substorms.flatten()
if signature in obs_signatures and signature in run_signatures or signature == 'All':
counts = get_counts(signature_run_substorms,
signature_obs_substorms)
true_positive, false_positive, false_negative, true_negative = counts
obs_total = true_positive + false_negative
run_total = true_positive + false_positive
skill = heidke_skill(*counts)
skill_ci = heidke_ci(signature_run_substorms,
signature_obs_substorms, nsamples)
signature_hit_rate = hit_rate(*counts)
signature_hit_rate_ci = metric_ci(signature_run_substorms,
signature_obs_substorms,
hit_rate, nsamples)
signature_false_alarm_rate = false_alarm_rate(*counts)
signature_false_alarm_rate_ci = metric_ci(
signature_run_substorms, signature_obs_substorms,
false_alarm_rate, nsamples)
else:
if signature in run_signatures:
run_total = np.sum(signature_run_substorms)
else:
run_total = None
if signature in obs_signatures:
obs_total = np.sum(signature_obs_substorms)
else:
obs_total = None
skill = None
skill_ci = (None, None)
signature_hit_rate = None
signature_hit_rate_ci = (None, None)
signature_false_alarm_rate = None
signature_false_alarm_rate_ci = (None, None)
all_counts = get_counts(signature_run_substorms, obs_substorms)
all_skill = heidke_skill(*all_counts)
all_skill_ci = heidke_ci(signature_run_substorms, obs_substorms,
nsamples)
all_hit_rate = hit_rate(*all_counts)
all_hit_rate_ci = metric_ci(signature_run_substorms, obs_substorms,
hit_rate, nsamples)
all_false_alarm_rate = false_alarm_rate(*all_counts)
all_false_alarm_rate_ci = metric_ci(signature_run_substorms,
obs_substorms,
false_alarm_rate, nsamples)
table_data[-1].append([
signature, run_total, obs_total, skill, skill_ci,
signature_hit_rate, signature_hit_rate_ci,
signature_false_alarm_rate, signature_false_alarm_rate_ci,
all_skill, all_skill_ci, all_hit_rate, all_hit_rate_ci,
all_false_alarm_rate, all_false_alarm_rate_ci
])
return table_data
ax.axhline(0, color='k', linestyle=':')
ax.axvline(0, color='k', linestyle=':')
ax.set_ylabel(ylabel)
ax.set_xlabel('Time since onset (h)')
return zip(polycols,
lines), [signature_type_labels[key] for key in signature_types]
obs_substorms, obs_onsets = find_substorms_convolution(obs_signatures,
obs_threshold,
tstep=tstep,
return_times=True)
run_onsets = {}
run_signatures = get_model_signature_lists(run_properties[0], datadir=datadir)
signature_type_labels = {
'All': 'All',
'AL': 'AL',
'image': 'IMAGE/FUV',
'plasmoids': 'Plasmoids',
'dipolarizations': 'Dipolarizations',
'epdata': 'LANL',
'MPB': 'MPB'
}
signature_types = set(['All'] + obs_signatures.keys() + run_signatures.keys())
run_colors = {
run: color
for run, color in zip(