mirror of
https://gitlab.science.ru.nl/mthesis-edeboone/m-thesis-introduction.git
synced 2024-12-30 15:13:31 +01:00
PDFs: revamped as used in Thesis
This commit is contained in:
parent
373164f1b0
commit
b9314a2800
1 changed files with 43 additions and 20 deletions
|
@ -13,6 +13,7 @@ import scipy.stats as stat
|
|||
from scipy import special
|
||||
|
||||
from lib.util import MethodMappingProxy as MethodProxy
|
||||
from lib.ft_plot import axis_pi_ticker
|
||||
|
||||
def expectation(x,pdfx):
|
||||
dx = x[1]-x[0]
|
||||
|
@ -48,6 +49,23 @@ def amplitude_distribution_gauss(a,sigma,s):
|
|||
k=s/sigma
|
||||
return (2*np.pi)**-0.5*np.exp(-((a-s)/sigma)**2/2)
|
||||
|
||||
|
||||
figsize = (8,6)
|
||||
if True:
|
||||
from matplotlib import rcParams
|
||||
#rcParams["text.usetex"] = True
|
||||
rcParams["font.family"] = "serif"
|
||||
plt.rc('lines',lw=2)
|
||||
if True:# small
|
||||
figsize = (6, 4)
|
||||
rcParams["font.size"] = "15" # 15 at 6,4 looks fine
|
||||
elif True: # large
|
||||
figsize = (9, 6)
|
||||
rcParams["font.size"] = "16" # 15 at 9,6 looks fine
|
||||
rcParams["grid.linestyle"] = 'dotted'
|
||||
rcParams["figure.figsize"] = figsize
|
||||
|
||||
|
||||
signal_max= 4
|
||||
amp_max = signal_max*2
|
||||
thetas = np.linspace(-np.pi,np.pi,500)
|
||||
|
@ -58,14 +76,14 @@ sigma = 1
|
|||
|
||||
## figure 1
|
||||
if True:
|
||||
fig, ax = plt.subplots(1,2,figsize=(2*8,1*8))
|
||||
_fig1, _ax0 =plt.subplots(1,1, figsize=(1*8, 1*8))
|
||||
_fig2, _ax1 =plt.subplots(1,1, figsize=(1*8, 1*8))
|
||||
fig, ax = plt.subplots(1,2,figsize=(2*figsize[0],figsize[1]))
|
||||
_fig1, _ax0 =plt.subplots(1,1)
|
||||
_fig2, _ax1 =plt.subplots(1,1)
|
||||
|
||||
ax0 = MethodProxy(ax[0], _ax0)
|
||||
ax1 = MethodProxy(ax[1], _ax1)
|
||||
for s in signals:
|
||||
pdfs_label='s/$\sigma$ ='+str(s)
|
||||
pdfs_label='s = '+str(int(s))
|
||||
phase_vals= phase_distribution(thetas,sigma,s)
|
||||
amp_vals= amplitude_distribution(amplitudes,sigma,s)
|
||||
phase_vals_g = phase_distribution_gauss(thetas,sigma,s)
|
||||
|
@ -73,13 +91,16 @@ if True:
|
|||
ax0.plot(amplitudes,amp_vals, label=pdfs_label)
|
||||
ax1.plot(thetas,phase_vals, label=pdfs_label)
|
||||
ax0.legend()
|
||||
ax0.set_xlabel(r'Amplitude $a$')
|
||||
ax0.set_ylabel(r'$p(a)$')
|
||||
|
||||
ax1.set_xlabel(r'Phase $\varphi$')
|
||||
ax1.set_ylabel(r'$p(\varphi)$')
|
||||
_ax1.legend()# only in the separate figure
|
||||
ax0.set_xlabel(r'$a$')
|
||||
ax0.set_xlabel(r'$\theta$')
|
||||
ax1.set_ylabel(r'$p(a)$')
|
||||
ax1.set_ylabel(r'$p(\theta)$')
|
||||
# ax[0].grid()
|
||||
# ax[1].grid()
|
||||
[ axis_pi_ticker(ax.xaxis, major_divider=3) for ax in ax1.elements ]
|
||||
|
||||
for a in [ax0, ax1]:
|
||||
a.grid()
|
||||
MethodProxy(fig, _fig1, _fig2).tight_layout()
|
||||
|
||||
fig.savefig('pdfs.pdf')
|
||||
|
@ -92,7 +113,7 @@ if True:
|
|||
## figure 2
|
||||
amplitudes = np.linspace(0,amp_max*5,500)
|
||||
signals = np.linspace(0.1,signal_max*5,101)
|
||||
if False:
|
||||
if True:
|
||||
V_theta = [variance(thetas,phase_distribution(thetas,sigma,s)) for s in signals ]
|
||||
E_theta=[expectation(thetas,phase_distribution(thetas,sigma,s)) for s in signals ]
|
||||
V_theta_g = [variance(thetas,phase_distribution_gauss(thetas,sigma,s)) for s in signals ]
|
||||
|
@ -102,17 +123,17 @@ if False:
|
|||
V_a_g = [variance(amplitudes,amplitude_distribution_gauss(amplitudes,sigma,s)) for s in signals ]
|
||||
E_a_g=[expectation(amplitudes,amplitude_distribution_gauss(amplitudes,sigma,s)) for s in signals ]
|
||||
|
||||
fig2, _ax2 = plt.subplots(2,2,figsize=(2*8,2*8))
|
||||
fig2, _ax2 = plt.subplots(2,2,figsize=(2*figsize[0],2*figsize[1]))
|
||||
ax2 = fig2.get_axes()
|
||||
if True:
|
||||
_figs = []
|
||||
_axs = []
|
||||
for i, ax in enumerate(_ax2):
|
||||
_f, _a = plt.subplots(1,1, figsize=(1*8, 1*8))
|
||||
for i, ax in enumerate(ax2):
|
||||
_f, _a = plt.subplots(1,1)
|
||||
_figs.append(_f)
|
||||
_axs.append(_a)
|
||||
|
||||
ax2[i] = MethodProxy(ax2[0], _a)
|
||||
ax2[i] = MethodProxy(ax, _a)
|
||||
|
||||
ax2[0].plot(signals,E_a,label='$p(a)$')
|
||||
ax2[0].plot(signals,E_a_g,ls='dashed',label='Gaussian approx.')
|
||||
|
@ -125,17 +146,17 @@ if False:
|
|||
ax2[1].set_xscale('log')
|
||||
ax2[1].set_ylabel('$\sigma_a^2$')
|
||||
|
||||
ax2[2].plot(signals,E_theta,label=r'$p(\theta)$')
|
||||
ax2[2].plot(signals,E_theta,label=r'$p(\varphi)$')
|
||||
ax2[2].plot(signals,E_theta_g,ls='dashed',label='Gaussian approx.')
|
||||
ax2[2].set_xscale('log')
|
||||
ax2[2].set_ylim(-1.1,1.1)
|
||||
ax2[2].set_ylabel(r'$\mu_\theta$')
|
||||
ax2[2].set_ylabel(r'$\mu_\varphi$')
|
||||
|
||||
ax2[3].plot(signals,V_theta,label=r'$p(\theta)$')
|
||||
ax2[3].plot(signals,V_theta,label=r'$p(\varphi)$')
|
||||
ax2[3].plot(signals,V_theta_g,ls='dashed',label='Gaussian approx.')
|
||||
ax2[3].set_xscale('log')
|
||||
ax2[3].set_yscale('log')
|
||||
ax2[3].set_ylabel(r'$\sigma_\theta^2$')
|
||||
ax2[3].set_ylabel(r'$\sigma_\varphi^2$')
|
||||
for a in ax2:
|
||||
a.grid(which='both')
|
||||
a.set_xlabel(r'$s/\sigma$')
|
||||
|
@ -150,12 +171,14 @@ if False:
|
|||
'phase_mean',
|
||||
'phase_sigma',
|
||||
][i]
|
||||
|
||||
_f.tight_layout()
|
||||
_f.savefig(fnames+'.pdf')
|
||||
plt.close(_f)
|
||||
|
||||
## figure 3, beacon timing accuracy
|
||||
if True:
|
||||
fig3, ax3 = plt.subplots(1,1,figsize=(1*8,1*8))
|
||||
fig3, ax3 = plt.subplots(1,1)
|
||||
ax3 = fig3.get_axes()
|
||||
sigma_t = [variance(thetas,phase_distribution(thetas,sigma,s)) for s in signals ]
|
||||
lfs=np.linspace(np.log10(50.),4,1)
|
||||
|
|
Loading…
Reference in a new issue