Skip to content
GitLab
Projects
Groups
Snippets
Help
Loading...
Help
What's new
10
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Sign in
Toggle navigation
Open sidebar
DavidWalz
airshower-deeplearning
Commits
e3d794ec
Commit
e3d794ec
authored
Sep 08, 2017
by
DavidWalz
Browse files
Options
Browse Files
Download
Email Patches
Plain Diff
add scripts for fitting arrival direction and energy
parent
543b050a
Changes
2
Hide whitespace changes
Inline
Side-by-side
Showing
2 changed files
with
146 additions
and
0 deletions
+146
-0
fitting/fit-energy.py
fitting/fit-energy.py
+90
-0
fitting/fit-planewave.py
fitting/fit-planewave.py
+56
-0
No files found.
fitting/fit-energy.py
0 → 100644
View file @
e3d794ec
import
numpy
as
np
import
matplotlib.pyplot
as
plt
# load data
with
np
.
load
(
'../RawData/showers-A1-0.npz'
)
as
data
:
vd
=
data
[
'detector'
]
vc
=
data
[
'showercore'
]
va
=
data
[
'showeraxis'
]
logE
=
data
[
'logE'
]
signal
=
data
[
'signal'
].
sum
(
axis
=-
1
)
# total signal
nb_events
=
len
(
vc
)
nb_stations
=
len
(
vd
)
# for each event, calculate distances of detectors to shower core, plane and axis
xd
=
np
.
repeat
(
np
.
expand_dims
(
vd
,
0
),
nb_events
,
axis
=
0
)
xc
=
np
.
repeat
(
np
.
expand_dims
(
vc
,
1
),
nb_stations
,
axis
=
1
)
xa
=
np
.
repeat
(
np
.
expand_dims
(
va
,
1
),
nb_stations
,
axis
=
1
)
d1
=
np
.
linalg
.
norm
(
xd
-
xc
,
axis
=-
1
)
# distance to shower core
d2
=
np
.
sum
((
xd
-
xc
)
*
xa
,
axis
=-
1
)
# distance to shower plane
d3
=
np
.
linalg
.
norm
(
xd
-
xc
-
xa
*
d2
[...,
np
.
newaxis
],
axis
=-
1
)
# distance to shower axis
# fit lateral distribution to each event, S(r) = S1000 * (r/1000)**a
logS1000
=
np
.
zeros
(
nb_events
)
for
i
in
range
(
nb_events
):
m
=
~
np
.
isnan
(
signal
[
i
])
# mask for triggered stations
m
*=
d3
[
i
]
>
200
# cut stations close to the shower core
x
=
np
.
log10
(
d3
[
i
][
m
]
/
1000
)
# log10(r/1000m) distance to shower axis
y
=
np
.
log10
(
signal
[
i
][
m
]
-
100
)
# log10(signal - noise level)
p
=
np
.
polyfit
(
x
,
y
,
1
)
# fit linear model
logS1000
[
i
]
=
p
[
1
]
# plt.figure()
# plt.scatter(x + 3, y)
# x = np.linspace(-1, 1)
# y = np.polyval(p, x)
# plt.plot(x + 3, y, 'r')
# plt.xlabel('$\log_{10}(r)$')
# plt.ylabel('$\log_{10}(S)$')
# plt.grid()
# fit zenith angle dependency: logS1000 = logS38 + p * (cos(zenith)**2 - cos(38)**2)
x
,
y
,
z
=
va
.
T
zenith
=
np
.
pi
/
2
-
np
.
arctan2
(
z
,
(
x
*
x
+
y
*
y
)
**
.
5
)
cz
=
np
.
cos
(
zenith
)
**
2
-
np
.
cos
(
np
.
deg2rad
(
38
))
**
2
pcal1
=
np
.
polyfit
(
cz
,
logS1000
,
1
)
cic
=
np
.
polyval
(
pcal1
,
cz
)
logS38
=
logS1000
/
cic
# remove the zenith angle dependency
plt
.
figure
()
plt
.
scatter
(
cz
,
logS1000
)
plt
.
plot
(
cz
,
cic
,
'r'
)
plt
.
xlabel
(
r
'$\cos(\theta)^2 - \cos(38)^2$'
)
plt
.
ylabel
(
r
'$\log_{10}(S_{1000})$'
)
plt
.
savefig
(
'energy-calibration1.png'
)
# fit energy dependency: logE = p0 + p1 * logS
pcal2
=
np
.
polyfit
(
logS38
,
logE
,
1
)
logE_rec
=
np
.
polyval
(
pcal2
,
logS38
)
plt
.
figure
()
plt
.
scatter
(
logS38
,
logE
)
x
=
np
.
linspace
(
min
(
logS38
),
max
(
logS38
))
y
=
np
.
polyval
(
pcal2
,
x
)
plt
.
plot
(
x
,
y
,
'r'
)
plt
.
xlabel
(
'$\log_{10}(S_{38})$'
)
plt
.
ylabel
(
'$\log_{10}(E / \mathrm{eV})$'
)
plt
.
grid
()
plt
.
savefig
(
'energy-calibration2.png'
)
# evaluate resolution
r
=
10
**
(
logE_rec
-
logE
)
-
1
# (E_rec - E_true) / E_true
plt
.
figure
()
plt
.
hist
(
r
,
bins
=
np
.
linspace
(
-
0.6
,
0.6
,
20
))
plt
.
xlabel
(
'$(E_\mathrm{rec} - E_\mathrm{true}) / E_\mathrm{true}$'
)
plt
.
ylabel
(
'\#'
)
plt
.
grid
()
plt
.
text
(
0.95
,
0.95
,
'mean = %.3f
\n
std = %.3f'
%
(
np
.
mean
(
r
),
np
.
std
(
r
)),
transform
=
plt
.
gca
().
transAxes
,
ha
=
'right'
,
va
=
'top'
,
bbox
=
dict
(
boxstyle
=
'round'
,
facecolor
=
'white'
))
plt
.
savefig
(
'energy-resolution.png'
)
fitting/fit-planewave.py
0 → 100644
View file @
e3d794ec
import
numpy
as
np
import
matplotlib.pyplot
as
plt
# load data
with
np
.
load
(
'../RawData/showers-A1-0.npz'
)
as
data
:
vd
=
data
[
'detector'
]
va
=
data
[
'showeraxis'
]
time
=
data
[
'time'
]
nb_events
=
len
(
va
)
# reconstruct shower directions with plane wave fit
# cf. ftp://ftp.mi.ingv.it/download/augliera/PER-MARA/article_del_pezzo.pdf
va_rec
=
np
.
zeros
((
nb_events
,
3
))
for
i
in
range
(
nb_events
):
m
=
~
np
.
isnan
(
time
[
i
])
# mask for triggered stations
t
=
time
[
i
][
m
]
x
,
y
,
z
=
vd
[
m
].
T
# distances between all pairs of stations
idx
=
np
.
triu_indices
(
len
(
t
),
k
=
1
)
dt
=
np
.
subtract
.
outer
(
t
,
t
)[
idx
]
dx
=
np
.
subtract
.
outer
(
x
,
x
)[
idx
]
dy
=
np
.
subtract
.
outer
(
y
,
y
)[
idx
]
# all stations on a plane --> neglect z-position
dv
=
np
.
stack
((
dx
,
dy
),
axis
=-
1
)
a1
=
np
.
linalg
.
inv
(
np
.
dot
(
dv
.
T
,
dv
))
# # for using z-position as well
# dz = np.subtract.outer(z, z)[idx]
# dv = np.stack((dx, dy, dz), axis=-1)
# a1 = np.linalg.pinv(np.dot(dv.T, dv))
a2
=
np
.
einsum
(
'ij,i->j'
,
dv
,
dt
)
a3
=
np
.
einsum
(
'ij,j->i'
,
a1
,
a2
)
# determine z-component using known speed of propagation = c
px
,
py
=
-
a3
*
3E8
pz
=
(
1
-
px
**
2
-
py
**
2
)
**
.
5
va_rec
[
i
]
=
[
px
,
py
,
pz
]
# plot angular separation between true and reconstructed angle
ang
=
np
.
arccos
(
np
.
clip
(
np
.
sum
(
va
*
va_rec
,
axis
=
1
),
-
1
,
1
))
*
180
/
np
.
pi
r68
=
np
.
percentile
(
ang
,
68
)
plt
.
figure
()
plt
.
hist
(
ang
,
bins
=
np
.
linspace
(
0
,
5
,
41
))
plt
.
axvline
(
r68
,
c
=
'r'
)
plt
.
xlabel
(
'Angular distance [deg]'
)
plt
.
ylabel
(
'\#'
)
plt
.
grid
()
plt
.
savefig
(
'fit-planewave.png'
,
bbox_inches
=
'tight'
)
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment