{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Import"
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"C:\\Users\\sa_li\\.conda\\envs\\rmd\\lib\\site-packages\\zfit\\util\\execution.py:57: UserWarning: Not running on Linux. Determining available cpus for thread can failand be overestimated. Workaround (only if too many cpus are used):`zfit.run.set_n_cpu(your_cpu_number)`\n",
" warnings.warn(\"Not running on Linux. Determining available cpus for thread can fail\"\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\n",
"WARNING: The TensorFlow contrib module will not be included in TensorFlow 2.0.\n",
"For more information, please see:\n",
" * https://github.com/tensorflow/community/blob/master/rfcs/20180907-contrib-sunset.md\n",
" * https://github.com/tensorflow/addons\n",
"If you depend on functionality not listed there, please file an issue.\n",
"\n"
]
}
],
"source": [
"import os\n",
"\n",
"# os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"-1\"\n",
"import random\n",
"import numpy as np\n",
"from pdg_const1 import pdg\n",
"import matplotlib\n",
"import matplotlib.pyplot as plt\n",
"import pickle as pkl\n",
"import sys\n",
"import time\n",
"from helperfunctions import display_time, prepare_plot\n",
"import cmath as c\n",
"import scipy.integrate as integrate\n",
"from scipy.optimize import fminbound\n",
"from array import array as arr\n",
"import collections\n",
"from itertools import compress\n",
"import tensorflow as tf\n",
"import zfit\n",
"from zfit import ztf\n",
"# from IPython.display import clear_output\n",
"import os\n",
"import tensorflow_probability as tfp\n",
"tfd = tfp.distributions"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [],
"source": [
"# chunksize = 10000\n",
"# zfit.run.chunking.active = True\n",
"# zfit.run.chunking.max_n_points = chunksize"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Build model and graphs\n",
"## Create graphs"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [],
"source": [
"def formfactor(q2, subscript, b0_0, b0_1, b0_2, bplus_0, bplus_1, bplus_2, bT_0, bT_1, bT_2): #returns real value\n",
" #check if subscript is viable\n",
"\n",
" if subscript != \"0\" and subscript != \"+\" and subscript != \"T\":\n",
" raise ValueError('Wrong subscript entered, choose either 0, + or T')\n",
"\n",
" #get constants\n",
"\n",
" mK = ztf.constant(pdg['Ks_M'])\n",
" mbstar0 = ztf.constant(pdg[\"mbstar0\"])\n",
" mbstar = ztf.constant(pdg[\"mbstar\"])\n",
"\n",
"\n",
" mmu = ztf.constant(pdg['muon_M'])\n",
" mb = ztf.constant(pdg['bquark_M'])\n",
" ms = ztf.constant(pdg['squark_M'])\n",
" mB = ztf.constant(pdg['Bplus_M'])\n",
"\n",
" #N comes from derivation in paper\n",
"\n",
" N = 3\n",
"\n",
" #some helperfunctions\n",
"\n",
" tpos = (mB - mK)**2\n",
" tzero = (mB + mK)*(ztf.sqrt(mB)-ztf.sqrt(mK))**2\n",
"\n",
" z_oben = ztf.sqrt(tpos - q2) - ztf.sqrt(tpos - tzero)\n",
" z_unten = ztf.sqrt(tpos - q2) + ztf.sqrt(tpos - tzero)\n",
" z = tf.divide(z_oben, z_unten)\n",
"\n",
" #calculate f0\n",
"\n",
" if subscript == \"0\":\n",
" prefactor = 1/(1 - q2/(mbstar0**2))\n",
" _sum = 0\n",
" b0 = [b0_0, b0_1, b0_2]\n",
"\n",
" for i in range(N):\n",
" _sum += b0[i]*(tf.pow(z,i))\n",
"\n",
" return ztf.to_complex(prefactor * _sum)\n",
"\n",
" #calculate f+ or fT\n",
"\n",
" else:\n",
" prefactor = 1/(1 - q2/(mbstar**2))\n",
" _sum = 0\n",
"\n",
" if subscript == \"T\":\n",
" bT = [bT_0, bT_1, bT_2]\n",
" for i in range(N):\n",
" _sum += bT[i] * (tf.pow(z, i) - ((-1)**(i-N)) * (i/N) * tf.pow(z, N))\n",
" else:\n",
" bplus = [bplus_0, bplus_1, bplus_2]\n",
" for i in range(N):\n",
" _sum += bplus[i] * (tf.pow(z, i) - ((-1)**(i-N)) * (i/N) * tf.pow(z, N))\n",
"\n",
" return ztf.to_complex(prefactor * _sum)\n",
"\n",
"def resonance(q, _mass, width, phase, scale):\n",
"\n",
" q2 = tf.pow(q, 2)\n",
"\n",
" mmu = ztf.constant(pdg['muon_M'])\n",
"\n",
" p = 0.5 * ztf.sqrt(q2 - 4*(mmu**2))\n",
"\n",
" p0 = 0.5 * ztf.sqrt(_mass**2 - 4*mmu**2)\n",
"\n",
" gamma_j = tf.divide(p, q) * _mass * width / p0\n",
"\n",
" #Calculate the resonance\n",
"\n",
" _top = tf.complex(_mass * width, ztf.constant(0.0))\n",
"\n",
" _bottom = tf.complex(_mass**2 - q2, -_mass*gamma_j)\n",
"\n",
" com = _top/_bottom\n",
"\n",
" #Rotate by the phase\n",
"\n",
" r = ztf.to_complex(scale*tf.abs(com))\n",
"\n",
" _phase = tf.angle(com)\n",
"\n",
" _phase += phase\n",
"\n",
" com = r * tf.exp(tf.complex(ztf.constant(0.0), _phase))\n",
"\n",
" return com\n",
"\n",
"\n",
"def axiv_nonres(q, b0_0, b0_1, b0_2, bplus_0, bplus_1, bplus_2, bT_0, bT_1, bT_2):\n",
"\n",
" GF = ztf.constant(pdg['GF'])\n",
" alpha_ew = ztf.constant(pdg['alpha_ew'])\n",
" Vtb = ztf.constant(pdg['Vtb'])\n",
" Vts = ztf.constant(pdg['Vts'])\n",
" C10eff = ztf.constant(pdg['C10eff'])\n",
"\n",
" mmu = ztf.constant(pdg['muon_M'])\n",
" mb = ztf.constant(pdg['bquark_M'])\n",
" ms = ztf.constant(pdg['squark_M'])\n",
" mK = ztf.constant(pdg['Ks_M'])\n",
" mB = ztf.constant(pdg['Bplus_M'])\n",
"\n",
" q2 = tf.pow(q, 2)\n",
"\n",
" #Some helperfunctions\n",
"\n",
" beta = 1. - 4. * mmu**2. / q2\n",
"\n",
" kabs = ztf.sqrt(mB**2. + tf.pow(q2, 2)/mB**2. + mK**4./mB**2. - 2. * (mB**2. * mK**2. + mK**2. * q2 + mB**2. * q2) / mB**2.)\n",
"\n",
" #prefactor in front of whole bracket\n",
"\n",
" prefactor1 = GF**2. *alpha_ew**2. * (tf.abs(Vtb*Vts))**2. * kabs * beta / (128. * np.pi**5.)\n",
"\n",
" #left term in bracket\n",
"\n",
" bracket_left = 2./3. * tf.pow(kabs,2) * tf.pow(beta,2) * tf.pow(tf.abs(ztf.to_complex(C10eff)*formfactor(q2, \"+\", b0_0, b0_1, b0_2, bplus_0, bplus_1, bplus_2, bT_0, bT_1, bT_2)),2)\n",
"\n",
" #middle term in bracket\n",
"\n",
" _top = 4. * mmu**2. * (mB**2. - mK**2.) * (mB**2. - mK**2.)\n",
"\n",
" _under = q2 * mB**2.\n",
"\n",
" bracket_middle = _top/_under *tf.pow(tf.abs(ztf.to_complex(C10eff) * formfactor(q2, \"0\", b0_0, b0_1, b0_2, bplus_0, bplus_1, bplus_2, bT_0, bT_1, bT_2)), 2)\n",
" \n",
" #Note sqrt(q2) comes from derivation as we use q2 and plot q\n",
"\n",
" return prefactor1 * (bracket_left + bracket_middle) * 2 * q\n",
"\n",
"def vec(q, funcs, b0_0, b0_1, b0_2, bplus_0, bplus_1, bplus_2, bT_0, bT_1, bT_2):\n",
" \n",
" q2 = tf.pow(q, 2)\n",
"\n",
" GF = ztf.constant(pdg['GF'])\n",
" alpha_ew = ztf.constant(pdg['alpha_ew'])\n",
" Vtb = ztf.constant(pdg['Vtb'])\n",
" Vts = ztf.constant(pdg['Vts'])\n",
" C7eff = ztf.constant(pdg['C7eff'])\n",
"\n",
" mmu = ztf.constant(pdg['muon_M'])\n",
" mb = ztf.constant(pdg['bquark_M'])\n",
" ms = ztf.constant(pdg['squark_M'])\n",
" mK = ztf.constant(pdg['Ks_M'])\n",
" mB = ztf.constant(pdg['Bplus_M'])\n",
"\n",
" #Some helperfunctions\n",
"\n",
" beta = 1. - 4. * mmu**2. / q2\n",
"\n",
" kabs = ztf.sqrt(mB**2. + tf.pow(q2, 2)/mB**2. + mK**4./mB**2. - 2 * (mB**2 * mK**2 + mK**2 * q2 + mB**2 * q2) / mB**2)\n",
" \n",
" #prefactor in front of whole bracket\n",
"\n",
" prefactor1 = GF**2. *alpha_ew**2. * (tf.abs(Vtb*Vts))**2 * kabs * beta / (128. * np.pi**5.)\n",
"\n",
" #right term in bracket\n",
"\n",
" prefactor2 = tf.pow(kabs,2) * (1. - 1./3. * beta)\n",
"\n",
" abs_bracket = tf.pow(tf.abs(c9eff(q, funcs) * formfactor(q2, \"+\", b0_0, b0_1, b0_2, bplus_0, bplus_1, bplus_2, bT_0, bT_1, bT_2) + ztf.to_complex(2.0 * C7eff * (mb + ms)/(mB + mK)) * formfactor(q2, \"T\", b0_0, b0_1, b0_2, bplus_0, bplus_1, bplus_2, bT_0, bT_1, bT_2)),2)\n",
"\n",
" bracket_right = prefactor2 * abs_bracket\n",
"\n",
" #Note sqrt(q2) comes from derivation as we use q2 and plot q\n",
"\n",
" return prefactor1 * bracket_right * 2 * q\n",
"\n",
"def c9eff(q, funcs):\n",
"\n",
" C9eff_nr = ztf.to_complex(ztf.constant(pdg['C9eff']))\n",
"\n",
" c9 = C9eff_nr + funcs\n",
"\n",
" return c9"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [],
"source": [
"def G(y):\n",
" \n",
" def inner_rect_bracket(q):\n",
" return tf.log(ztf.to_complex((1+tf.sqrt(q))/(1-tf.sqrt(q)))-tf.complex(ztf.constant(0), -1*ztf.constant(np.pi))) \n",
" \n",
" def inner_right(q):\n",
" return ztf.to_complex(2 * tf.atan(1/tf.sqrt(tf.math.real(-q))))\n",
" \n",
" big_bracket = tf.where(tf.math.real(y) > ztf.constant(0.0), inner_rect_bracket(y), inner_right(y))\n",
" \n",
" return ztf.to_complex(tf.sqrt(tf.abs(y))) * big_bracket\n",
"\n",
"def h_S(m, q):\n",
" \n",
" return ztf.to_complex(2) - G(ztf.to_complex(1) - ztf.to_complex(4*tf.pow(m, 2)) / ztf.to_complex(tf.pow(q, 2)))\n",
"\n",
"def h_P(m, q):\n",
" \n",
" return ztf.to_complex(2/3) + (ztf.to_complex(1) - ztf.to_complex(4*tf.pow(m, 2)) / ztf.to_complex(tf.pow(q, 2))) * h_S(m,q)\n",
"\n",
"def two_p_ccbar(mD, m_D_bar, m_D_star, q):\n",
" \n",
" \n",
" #Load constants\n",
" nu_D_bar = ztf.to_complex(pdg[\"nu_D_bar\"])\n",
" nu_D = ztf.to_complex(pdg[\"nu_D\"])\n",
" nu_D_star = ztf.to_complex(pdg[\"nu_D_star\"])\n",
" \n",
" phase_D_bar = ztf.to_complex(pdg[\"phase_D_bar\"])\n",
" phase_D = ztf.to_complex(pdg[\"phase_D\"])\n",
" phase_D_star = ztf.to_complex(pdg[\"phase_D_star\"])\n",
" \n",
" #Calculation\n",
" left_part = nu_D_bar * tf.exp(tf.complex(ztf.constant(0.0), phase_D_bar)) * h_S(m_D_bar, q) \n",
" \n",
" right_part_D = nu_D * tf.exp(tf.complex(ztf.constant(0.0), phase_D)) * h_P(m_D, q) \n",
" \n",
" right_part_D_star = nu_D_star * tf.exp(tf.complex(ztf.constant(0.0), phase_D_star)) * h_P(m_D_star, q) \n",
"\n",
" return left_part + right_part_D + right_part_D_star"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Build pdf"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [],
"source": [
"class total_pdf_cut(zfit.pdf.ZPDF):\n",
" _N_OBS = 1 # dimension, can be omitted\n",
" _PARAMS = ['b0_0', 'b0_1', 'b0_2', \n",
" 'bplus_0', 'bplus_1', 'bplus_2', \n",
" 'bT_0', 'bT_1', 'bT_2', \n",
" 'rho_mass', 'rho_scale', 'rho_phase', 'rho_width',\n",
" 'jpsi_mass', 'jpsi_scale', 'jpsi_phase', 'jpsi_width',\n",
" 'psi2s_mass', 'psi2s_scale', 'psi2s_phase', 'psi2s_width',\n",
" 'p3770_mass', 'p3770_scale', 'p3770_phase', 'p3770_width',\n",
" 'p4040_mass', 'p4040_scale', 'p4040_phase', 'p4040_width',\n",
" 'p4160_mass', 'p4160_scale', 'p4160_phase', 'p4160_width',\n",
" 'p4415_mass', 'p4415_scale', 'p4415_phase', 'p4415_width',\n",
" 'omega_mass', 'omega_scale', 'omega_phase', 'omega_width',\n",
" 'phi_mass', 'phi_scale', 'phi_phase', 'phi_width',\n",
" 'Dbar_mass', 'Dbar_scale', 'Dbar_phase',\n",
" 'Dstar_mass', 'DDstar_scale', 'DDstar_phase', 'D_mass',\n",
" 'tau_mass', 'C_tt']\n",
"# the name of the parameters\n",
"\n",
" def _unnormalized_pdf(self, x):\n",
" \n",
" x = x.unstack_x()\n",
" \n",
" b0 = [self.params['b0_0'], self.params['b0_1'], self.params['b0_2']]\n",
" bplus = [self.params['bplus_0'], self.params['bplus_1'], self.params['bplus_2']]\n",
" bT = [self.params['bT_0'], self.params['bT_1'], self.params['bT_2']]\n",
" \n",
" def rho_res(q):\n",
" return resonance(q, _mass = self.params['rho_mass'], scale = self.params['rho_scale'],\n",
" phase = self.params['rho_phase'], width = self.params['rho_width'])\n",
" \n",
" def omega_res(q):\n",
" return resonance(q, _mass = self.params['omega_mass'], scale = self.params['omega_scale'],\n",
" phase = self.params['omega_phase'], width = self.params['omega_width'])\n",
" \n",
" def phi_res(q):\n",
" return resonance(q, _mass = self.params['phi_mass'], scale = self.params['phi_scale'],\n",
" phase = self.params['phi_phase'], width = self.params['phi_width'])\n",
"\n",
" def jpsi_res(q):\n",
" return ztf.to_complex(tf.pow(q, 2) / tf.pow(self.params['jpsi_mass'], 2)) * resonance(q, _mass = self.params['jpsi_mass'], \n",
" scale = self.params['jpsi_scale'],\n",
" phase = self.params['jpsi_phase'], \n",
" width = self.params['jpsi_width'])\n",
" def psi2s_res(q):\n",
" return ztf.to_complex(tf.pow(q, 2) / tf.pow(self.params['psi2s_mass'], 2)) * resonance(q, _mass = self.params['psi2s_mass'], \n",
" scale = self.params['psi2s_scale'],\n",
" phase = self.params['psi2s_phase'], \n",
" width = self.params['psi2s_width'])\n",
" def p3770_res(q):\n",
" return ztf.to_complex(tf.pow(q, 2) / tf.pow(self.params['p3770_mass'], 2)) * resonance(q, _mass = self.params['p3770_mass'], \n",
" scale = self.params['p3770_scale'],\n",
" phase = self.params['p3770_phase'], \n",
" width = self.params['p3770_width'])\n",
" \n",
" def p4040_res(q):\n",
" return ztf.to_complex(tf.pow(q, 2) / tf.pow(self.params['p4040_mass'], 2)) * resonance(q, _mass = self.params['p4040_mass'], \n",
" scale = self.params['p4040_scale'],\n",
" phase = self.params['p4040_phase'], \n",
" width = self.params['p4040_width'])\n",
" \n",
" def p4160_res(q):\n",
" return ztf.to_complex(tf.pow(q, 2) / tf.pow(self.params['p4160_mass'], 2)) * resonance(q, _mass = self.params['p4160_mass'], \n",
" scale = self.params['p4160_scale'],\n",
" phase = self.params['p4160_phase'], \n",
" width = self.params['p4160_width'])\n",
" \n",
" def p4415_res(q):\n",
" return ztf.to_complex(tf.pow(q, 2) / tf.pow(self.params['p4415_mass'], 2)) * resonance(q, _mass = self.params['p4415_mass'], \n",
" scale = self.params['p4415_scale'],\n",
" phase = self.params['p4415_phase'], \n",
" width = self.params['p4415_width'])\n",
" \n",
" def P2_D(q):\n",
" Dbar_contrib = ztf.to_complex(self.params['Dbar_scale'])*tf.exp(tf.complex(ztf.constant(0.0), self.params['Dbar_phase']))*ztf.to_complex(h_S(self.params['Dbar_mass'], q))\n",
" DDstar_contrib = ztf.to_complex(self.params['DDstar_scale'])*tf.exp(tf.complex(ztf.constant(0.0), self.params['DDstar_phase']))*(ztf.to_complex(h_P(self.params['Dstar_mass'], q)) + ztf.to_complex(h_P(self.params['D_mass'], q)))\n",
" return Dbar_contrib + DDstar_contrib\n",
" \n",
" def ttau_cusp(q):\n",
" return ztf.to_complex(self.params['C_tt'])*(ztf.to_complex((h_S(self.params['tau_mass'], q))) - ztf.to_complex(h_P(self.params['tau_mass'], q)))\n",
" \n",
"\n",
" funcs = rho_res(x) + omega_res(x) + phi_res(x) + jpsi_res(x) + psi2s_res(x) + p3770_res(x) + p4040_res(x)+ p4160_res(x) + p4415_res(x) + P2_D(x) + ttau_cusp(x)\n",
"\n",
" vec_f = vec(x, funcs, b0_0, b0_1, b0_2, bplus_0, bplus_1, bplus_2, bT_0, bT_1, bT_2)\n",
"\n",
" axiv_nr = axiv_nonres(x, b0_0, b0_1, b0_2, bplus_0, bplus_1, bplus_2, bT_0, bT_1, bT_2)\n",
"\n",
" tot = vec_f + axiv_nr\n",
" \n",
" #Cut out jpsi and psi2s\n",
" \n",
" tot = tf.where(tf.math.logical_or(x < ztf.constant(jpsi_mass-60.), x > ztf.constant(jpsi_mass+70.)), tot, 0.0*tot)\n",
" \n",
" tot = tf.where(tf.math.logical_or(x < ztf.constant(psi2s_mass-50.), x > ztf.constant(psi2s_mass+50.)), tot, 0.0*tot)\n",
" \n",
" return tot\n",
" \n",
"class total_pdf_full(zfit.pdf.ZPDF):\n",
" _N_OBS = 1 # dimension, can be omitted\n",
" _PARAMS = ['b0_0', 'b0_1', 'b0_2', \n",
" 'bplus_0', 'bplus_1', 'bplus_2', \n",
" 'bT_0', 'bT_1', 'bT_2', \n",
" 'rho_mass', 'rho_scale', 'rho_phase', 'rho_width',\n",
" 'jpsi_mass', 'jpsi_scale', 'jpsi_phase', 'jpsi_width',\n",
" 'psi2s_mass', 'psi2s_scale', 'psi2s_phase', 'psi2s_width',\n",
" 'p3770_mass', 'p3770_scale', 'p3770_phase', 'p3770_width',\n",
" 'p4040_mass', 'p4040_scale', 'p4040_phase', 'p4040_width',\n",
" 'p4160_mass', 'p4160_scale', 'p4160_phase', 'p4160_width',\n",
" 'p4415_mass', 'p4415_scale', 'p4415_phase', 'p4415_width',\n",
" 'omega_mass', 'omega_scale', 'omega_phase', 'omega_width',\n",
" 'phi_mass', 'phi_scale', 'phi_phase', 'phi_width',\n",
" 'Dbar_mass', 'Dbar_scale', 'Dbar_phase',\n",
" 'Dstar_mass', 'DDstar_scale', 'DDstar_phase', 'D_mass',\n",
" 'tau_mass', 'C_tt']\n",
"# the name of the parameters\n",
"\n",
" def _unnormalized_pdf(self, x):\n",
" \n",
" x = x.unstack_x()\n",
" \n",
" b0 = [self.params['b0_0'], self.params['b0_1'], self.params['b0_2']]\n",
" bplus = [self.params['bplus_0'], self.params['bplus_1'], self.params['bplus_2']]\n",
" bT = [self.params['bT_0'], self.params['bT_1'], self.params['bT_2']]\n",
" \n",
" def rho_res(q):\n",
" return resonance(q, _mass = self.params['rho_mass'], scale = self.params['rho_scale'],\n",
" phase = self.params['rho_phase'], width = self.params['rho_width'])\n",
" \n",
" def omega_res(q):\n",
" return resonance(q, _mass = self.params['omega_mass'], scale = self.params['omega_scale'],\n",
" phase = self.params['omega_phase'], width = self.params['omega_width'])\n",
" \n",
" def phi_res(q):\n",
" return resonance(q, _mass = self.params['phi_mass'], scale = self.params['phi_scale'],\n",
" phase = self.params['phi_phase'], width = self.params['phi_width'])\n",
"\n",
" def jpsi_res(q):\n",
" return ztf.to_complex(tf.pow(q, 2) / tf.pow(self.params['jpsi_mass'], 2)) * resonance(q, _mass = self.params['jpsi_mass'], \n",
" scale = self.params['jpsi_scale'],\n",
" phase = self.params['jpsi_phase'], \n",
" width = self.params['jpsi_width'])\n",
" def psi2s_res(q):\n",
" return ztf.to_complex(tf.pow(q, 2) / tf.pow(self.params['psi2s_mass'], 2)) * resonance(q, _mass = self.params['psi2s_mass'], \n",
" scale = self.params['psi2s_scale'],\n",
" phase = self.params['psi2s_phase'], \n",
" width = self.params['psi2s_width'])\n",
" def p3770_res(q):\n",
" return ztf.to_complex(tf.pow(q, 2) / tf.pow(self.params['p3770_mass'], 2)) * resonance(q, _mass = self.params['p3770_mass'], \n",
" scale = self.params['p3770_scale'],\n",
" phase = self.params['p3770_phase'], \n",
" width = self.params['p3770_width'])\n",
" \n",
" def p4040_res(q):\n",
" return ztf.to_complex(tf.pow(q, 2) / tf.pow(self.params['p4040_mass'], 2)) * resonance(q, _mass = self.params['p4040_mass'], \n",
" scale = self.params['p4040_scale'],\n",
" phase = self.params['p4040_phase'], \n",
" width = self.params['p4040_width'])\n",
" \n",
" def p4160_res(q):\n",
" return ztf.to_complex(tf.pow(q, 2) / tf.pow(self.params['p4160_mass'], 2)) * resonance(q, _mass = self.params['p4160_mass'], \n",
" scale = self.params['p4160_scale'],\n",
" phase = self.params['p4160_phase'], \n",
" width = self.params['p4160_width'])\n",
" \n",
" def p4415_res(q):\n",
" return ztf.to_complex(tf.pow(q, 2) / tf.pow(self.params['p4415_mass'], 2)) * resonance(q, _mass = self.params['p4415_mass'], \n",
" scale = self.params['p4415_scale'],\n",
" phase = self.params['p4415_phase'], \n",
" width = self.params['p4415_width'])\n",
" \n",
" def P2_D(q):\n",
" Dbar_contrib = ztf.to_complex(self.params['Dbar_scale'])*tf.exp(tf.complex(ztf.constant(0.0), self.params['Dbar_phase']))*ztf.to_complex(h_S(self.params['Dbar_mass'], q))\n",
" DDstar_contrib = ztf.to_complex(self.params['DDstar_scale'])*tf.exp(tf.complex(ztf.constant(0.0), self.params['DDstar_phase']))*(ztf.to_complex(h_P(self.params['Dstar_mass'], q)) + ztf.to_complex(h_P(self.params['D_mass'], q)))\n",
" return Dbar_contrib + DDstar_contrib\n",
" \n",
" def ttau_cusp(q):\n",
" return ztf.to_complex(self.params['C_tt'])*(ztf.to_complex((h_S(self.params['tau_mass'], q))) - ztf.to_complex(h_P(self.params['tau_mass'], q)))\n",
" \n",
"\n",
" funcs = rho_res(x) + omega_res(x) + phi_res(x) + jpsi_res(x) + psi2s_res(x) + p3770_res(x) + p4040_res(x)+ p4160_res(x) + p4415_res(x) + P2_D(x) + ttau_cusp(x)\n",
"\n",
" vec_f = vec(x, funcs, b0_0, b0_1, b0_2, bplus_0, bplus_1, bplus_2, bT_0, bT_1, bT_2)\n",
"\n",
" axiv_nr = axiv_nonres(x, b0_0, b0_1, b0_2, bplus_0, bplus_1, bplus_2, bT_0, bT_1, bT_2)\n",
"\n",
" tot = vec_f + axiv_nr\n",
" \n",
" #Cut out jpsi and psi2s\n",
" \n",
"# tot = tf.where(tf.math.logical_or(x < ztf.constant(jpsi_mass-60.), x > ztf.constant(jpsi_mass+70.)), tot, 0.0*tot)\n",
" \n",
"# tot = tf.where(tf.math.logical_or(x < ztf.constant(psi2s_mass-50.), x > ztf.constant(psi2s_mass+50.)), tot, 0.0*tot)\n",
" \n",
" return tot"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Setup parameters"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"WARNING:tensorflow:From C:\\Users\\sa_li\\.conda\\envs\\rmd\\lib\\site-packages\\tensorflow\\python\\ops\\resource_variable_ops.py:435: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.\n",
"Instructions for updating:\n",
"Colocations handled automatically by placer.\n"
]
}
],
"source": [
"# formfactors\n",
"\n",
"b0_0 = zfit.Parameter(\"b0_0\", ztf.constant(0.292), floating = False) #, lower_limit = -2.0, upper_limit= 2.0)\n",
"b0_1 = zfit.Parameter(\"b0_1\", ztf.constant(0.281), floating = False) #, lower_limit = -2.0, upper_limit= 2.0)\n",
"b0_2 = zfit.Parameter(\"b0_2\", ztf.constant(0.150), floating = False) #, lower_limit = -2.0, upper_limit= 2.0)\n",
"\n",
"bplus_0 = zfit.Parameter(\"bplus_0\", ztf.constant(0.466), lower_limit = -2.0, upper_limit= 2.0)\n",
"bplus_1 = zfit.Parameter(\"bplus_1\", ztf.constant(-0.885), lower_limit = -2.0, upper_limit= 2.0)\n",
"bplus_2 = zfit.Parameter(\"bplus_2\", ztf.constant(-0.213), lower_limit = -2.0, upper_limit= 2.0)\n",
"\n",
"bT_0 = zfit.Parameter(\"bT_0\", ztf.constant(0.460), floating = False) #, lower_limit = -2.0, upper_limit= 2.0)\n",
"bT_1 = zfit.Parameter(\"bT_1\", ztf.constant(-1.089), floating = False) #, lower_limit = -2.0, upper_limit= 2.0)\n",
"bT_2 = zfit.Parameter(\"bT_2\", ztf.constant(-1.114), floating = False) #, lower_limit = -2.0, upper_limit= 2.0)\n",
"\n",
"\n",
"#rho\n",
"\n",
"rho_mass, rho_width, rho_phase, rho_scale = pdg[\"rho\"]\n",
"\n",
"rho_m = zfit.Parameter(\"rho_m\", ztf.constant(rho_mass), floating = False) #lower_limit = rho_mass - rho_width, upper_limit = rho_mass + rho_width)\n",
"rho_w = zfit.Parameter(\"rho_w\", ztf.constant(rho_width), floating = False)\n",
"rho_p = zfit.Parameter(\"rho_p\", ztf.constant(rho_phase), lower_limit=-2*np.pi, upper_limit=2*np.pi)\n",
"rho_s = zfit.Parameter(\"rho_s\", ztf.constant(rho_scale), lower_limit=rho_scale-np.sqrt(rho_scale), upper_limit=rho_scale+np.sqrt(rho_scale))\n",
"\n",
"#omega\n",
"\n",
"omega_mass, omega_width, omega_phase, omega_scale = pdg[\"omega\"]\n",
"\n",
"omega_m = zfit.Parameter(\"omega_m\", ztf.constant(omega_mass), floating = False)\n",
"omega_w = zfit.Parameter(\"omega_w\", ztf.constant(omega_width), floating = False)\n",
"omega_p = zfit.Parameter(\"omega_p\", ztf.constant(omega_phase), lower_limit=-2*np.pi, upper_limit=2*np.pi)\n",
"omega_s = zfit.Parameter(\"omega_s\", ztf.constant(omega_scale), lower_limit=omega_scale-np.sqrt(omega_scale), upper_limit=omega_scale+np.sqrt(omega_scale))\n",
"\n",
"\n",
"#phi\n",
"\n",
"phi_mass, phi_width, phi_phase, phi_scale = pdg[\"phi\"]\n",
"\n",
"phi_m = zfit.Parameter(\"phi_m\", ztf.constant(phi_mass), floating = False)\n",
"phi_w = zfit.Parameter(\"phi_w\", ztf.constant(phi_width), floating = False)\n",
"phi_p = zfit.Parameter(\"phi_p\", ztf.constant(phi_phase), lower_limit=-2*np.pi, upper_limit=2*np.pi)\n",
"phi_s = zfit.Parameter(\"phi_s\", ztf.constant(phi_scale), lower_limit=phi_scale-np.sqrt(phi_scale), upper_limit=phi_scale+np.sqrt(phi_scale))\n",
"\n",
"#jpsi\n",
"\n",
"jpsi_mass, jpsi_width, jpsi_phase, jpsi_scale = pdg[\"jpsi\"]\n",
"\n",
"jpsi_m = zfit.Parameter(\"jpsi_m\", ztf.constant(jpsi_mass), floating = False)\n",
"jpsi_w = zfit.Parameter(\"jpsi_w\", ztf.constant(jpsi_width), floating = False)\n",
"jpsi_p = zfit.Parameter(\"jpsi_p\", ztf.constant(jpsi_phase), lower_limit=-2*np.pi, upper_limit=2*np.pi)\n",
"jpsi_s = zfit.Parameter(\"jpsi_s\", ztf.constant(jpsi_scale), floating = False) #, lower_limit=jpsi_scale-np.sqrt(jpsi_scale), upper_limit=jpsi_scale+np.sqrt(jpsi_scale))\n",
"\n",
"#psi2s\n",
"\n",
"psi2s_mass, psi2s_width, psi2s_phase, psi2s_scale = pdg[\"psi2s\"]\n",
"\n",
"psi2s_m = zfit.Parameter(\"psi2s_m\", ztf.constant(psi2s_mass), floating = False)\n",
"psi2s_w = zfit.Parameter(\"psi2s_w\", ztf.constant(psi2s_width), floating = False)\n",
"psi2s_p = zfit.Parameter(\"psi2s_p\", ztf.constant(psi2s_phase), lower_limit=-2*np.pi, upper_limit=2*np.pi)\n",
"psi2s_s = zfit.Parameter(\"psi2s_s\", ztf.constant(psi2s_scale), floating = False) #, lower_limit=psi2s_scale-np.sqrt(psi2s_scale), upper_limit=psi2s_scale+np.sqrt(psi2s_scale))\n",
"\n",
"#psi(3770)\n",
"\n",
"p3770_mass, p3770_width, p3770_phase, p3770_scale = pdg[\"p3770\"]\n",
"\n",
"p3770_m = zfit.Parameter(\"p3770_m\", ztf.constant(p3770_mass), floating = False)\n",
"p3770_w = zfit.Parameter(\"p3770_w\", ztf.constant(p3770_width), floating = False)\n",
"p3770_p = zfit.Parameter(\"p3770_p\", ztf.constant(p3770_phase), lower_limit=-2*np.pi, upper_limit=2*np.pi)\n",
"p3770_s = zfit.Parameter(\"p3770_s\", ztf.constant(p3770_scale), lower_limit=p3770_scale-np.sqrt(p3770_scale), upper_limit=p3770_scale+np.sqrt(p3770_scale))\n",
"\n",
"#psi(4040)\n",
"\n",
"p4040_mass, p4040_width, p4040_phase, p4040_scale = pdg[\"p4040\"]\n",
"\n",
"p4040_m = zfit.Parameter(\"p4040_m\", ztf.constant(p4040_mass), floating = False)\n",
"p4040_w = zfit.Parameter(\"p4040_w\", ztf.constant(p4040_width), floating = False)\n",
"p4040_p = zfit.Parameter(\"p4040_p\", ztf.constant(p4040_phase), lower_limit=-2*np.pi, upper_limit=2*np.pi)\n",
"p4040_s = zfit.Parameter(\"p4040_s\", ztf.constant(p4040_scale), lower_limit=p4040_scale-np.sqrt(p4040_scale), upper_limit=p4040_scale+np.sqrt(p4040_scale))\n",
"\n",
"#psi(4160)\n",
"\n",
"p4160_mass, p4160_width, p4160_phase, p4160_scale = pdg[\"p4160\"]\n",
"\n",
"p4160_m = zfit.Parameter(\"p4160_m\", ztf.constant(p4160_mass), floating = False)\n",
"p4160_w = zfit.Parameter(\"p4160_w\", ztf.constant(p4160_width), floating = False)\n",
"p4160_p = zfit.Parameter(\"p4160_p\", ztf.constant(p4160_phase), lower_limit=-2*np.pi, upper_limit=2*np.pi)\n",
"p4160_s = zfit.Parameter(\"p4160_s\", ztf.constant(p4160_scale), lower_limit=p4160_scale-np.sqrt(p4160_scale), upper_limit=p4160_scale+np.sqrt(p4160_scale))\n",
"\n",
"#psi(4415)\n",
"\n",
"p4415_mass, p4415_width, p4415_phase, p4415_scale = pdg[\"p4415\"]\n",
"\n",
"p4415_m = zfit.Parameter(\"p4415_m\", ztf.constant(p4415_mass), floating = False)\n",
"p4415_w = zfit.Parameter(\"p4415_w\", ztf.constant(p4415_width), floating = False)\n",
"p4415_p = zfit.Parameter(\"p4415_p\", ztf.constant(p4415_phase), lower_limit=-2*np.pi, upper_limit=2*np.pi)\n",
"p4415_s = zfit.Parameter(\"p4415_s\", ztf.constant(p4415_scale), lower_limit=p4415_scale-np.sqrt(p4415_scale), upper_limit=p4415_scale+np.sqrt(p4415_scale))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Dynamic generation of 2 particle contribution"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {},
"outputs": [],
"source": [
"m_c = 1300\n",
"\n",
"Dbar_phase = 0.0\n",
"DDstar_phase = 0.0\n",
"Dstar_mass = pdg['Dst_M']\n",
"Dbar_mass = pdg['D0_M']\n",
"D_mass = pdg['D0_M']\n",
"\n",
"Dbar_s = zfit.Parameter(\"Dbar_s\", ztf.constant(0.0), lower_limit=-0.3, upper_limit=0.3)\n",
"Dbar_m = zfit.Parameter(\"Dbar_m\", ztf.constant(Dbar_mass), floating = False)\n",
"Dbar_p = zfit.Parameter(\"Dbar_p\", ztf.constant(Dbar_phase), lower_limit=-2*np.pi, upper_limit=2*np.pi)#, floating = False)\n",
"DDstar_s = zfit.Parameter(\"DDstar_s\", ztf.constant(0.0), lower_limit=-0.3, upper_limit=0.3)#, floating = False)\n",
"Dstar_m = zfit.Parameter(\"Dstar_m\", ztf.constant(Dstar_mass), floating = False)\n",
"D_m = zfit.Parameter(\"D_m\", ztf.constant(D_mass), floating = False)\n",
"DDstar_p = zfit.Parameter(\"DDstar_p\", ztf.constant(DDstar_phase), lower_limit=-2*np.pi, upper_limit=2*np.pi)#, floating = False)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Tau parameters"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [],
"source": [
"tau_m = zfit.Parameter(\"tau_m\", ztf.constant(pdg['tau_M']), floating = False)\n",
"Ctt = zfit.Parameter(\"Ctt\", ztf.constant(0.0), lower_limit=-2.5, upper_limit=2.5)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Load data"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {},
"outputs": [],
"source": [
"x_min = 2*pdg['muon_M']\n",
"x_max = (pdg[\"Bplus_M\"]-pdg[\"Ks_M\"]-0.1)\n",
"\n",
"# # Full spectrum\n",
"\n",
"obs_toy = zfit.Space('q', limits = (x_min, x_max))\n",
"\n",
"# Jpsi and Psi2s cut out\n",
"\n",
"obs1 = zfit.Space('q', limits = (x_min, jpsi_mass - 60.))\n",
"obs2 = zfit.Space('q', limits = (jpsi_mass + 70., psi2s_mass - 50.))\n",
"obs3 = zfit.Space('q', limits = (psi2s_mass + 50., x_max))\n",
"\n",
"obs_fit = obs1 + obs2 + obs3\n",
"\n",
"# with open(r\"./data/slim_points/slim_points_toy_0_range({0}-{1}).pkl\".format(int(x_min), int(x_max)), \"rb\") as input_file:\n",
"# part_set = pkl.load(input_file)\n",
"\n",
"# x_part = part_set['x_part']\n",
"\n",
"# x_part = x_part.astype('float64')\n",
"\n",
"# data = zfit.data.Data.from_numpy(array=x_part, obs=obs)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Setup pdf"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {},
"outputs": [],
"source": [
"total_f = total_pdf_cut(obs=obs_toy, jpsi_mass = jpsi_m, jpsi_scale = jpsi_s, jpsi_phase = jpsi_p, jpsi_width = jpsi_w,\n",
" psi2s_mass = psi2s_m, psi2s_scale = psi2s_s, psi2s_phase = psi2s_p, psi2s_width = psi2s_w,\n",
" p3770_mass = p3770_m, p3770_scale = p3770_s, p3770_phase = p3770_p, p3770_width = p3770_w,\n",
" p4040_mass = p4040_m, p4040_scale = p4040_s, p4040_phase = p4040_p, p4040_width = p4040_w,\n",
" p4160_mass = p4160_m, p4160_scale = p4160_s, p4160_phase = p4160_p, p4160_width = p4160_w,\n",
" p4415_mass = p4415_m, p4415_scale = p4415_s, p4415_phase = p4415_p, p4415_width = p4415_w,\n",
" rho_mass = rho_m, rho_scale = rho_s, rho_phase = rho_p, rho_width = rho_w,\n",
" omega_mass = omega_m, omega_scale = omega_s, omega_phase = omega_p, omega_width = omega_w,\n",
" phi_mass = phi_m, phi_scale = phi_s, phi_phase = phi_p, phi_width = phi_w,\n",
" Dstar_mass = Dstar_m, DDstar_scale = DDstar_s, DDstar_phase = DDstar_p, D_mass = D_m,\n",
" Dbar_mass = Dbar_m, Dbar_scale = Dbar_s, Dbar_phase = Dbar_p,\n",
" tau_mass = tau_m, C_tt = Ctt, b0_0 = b0_0, b0_1 = b0_1, b0_2 = b0_2,\n",
" bplus_0 = bplus_0, bplus_1 = bplus_1, bplus_2 = bplus_2,\n",
" bT_0 = bT_0, bT_1 = bT_1, bT_2 = bT_2)\n",
"\n",
"total_f_fit = total_pdf_full(obs=obs_fit, jpsi_mass = jpsi_m, jpsi_scale = jpsi_s, jpsi_phase = jpsi_p, jpsi_width = jpsi_w,\n",
" psi2s_mass = psi2s_m, psi2s_scale = psi2s_s, psi2s_phase = psi2s_p, psi2s_width = psi2s_w,\n",
" p3770_mass = p3770_m, p3770_scale = p3770_s, p3770_phase = p3770_p, p3770_width = p3770_w,\n",
" p4040_mass = p4040_m, p4040_scale = p4040_s, p4040_phase = p4040_p, p4040_width = p4040_w,\n",
" p4160_mass = p4160_m, p4160_scale = p4160_s, p4160_phase = p4160_p, p4160_width = p4160_w,\n",
" p4415_mass = p4415_m, p4415_scale = p4415_s, p4415_phase = p4415_p, p4415_width = p4415_w,\n",
" rho_mass = rho_m, rho_scale = rho_s, rho_phase = rho_p, rho_width = rho_w,\n",
" omega_mass = omega_m, omega_scale = omega_s, omega_phase = omega_p, omega_width = omega_w,\n",
" phi_mass = phi_m, phi_scale = phi_s, phi_phase = phi_p, phi_width = phi_w,\n",
" Dstar_mass = Dstar_m, DDstar_scale = DDstar_s, DDstar_phase = DDstar_p, D_mass = D_m,\n",
" Dbar_mass = Dbar_m, Dbar_scale = Dbar_s, Dbar_phase = Dbar_p,\n",
" tau_mass = tau_m, C_tt = Ctt, b0_0 = b0_0, b0_1 = b0_1, b0_2 = b0_2,\n",
" bplus_0 = bplus_0, bplus_1 = bplus_1, bplus_2 = bplus_2,\n",
" bT_0 = bT_0, bT_1 = bT_1, bT_2 = bT_2)\n",
" \n",
"# print(total_pdf.obs)\n",
"\n",
"# print(calcs_test)\n",
"\n",
"# for param in total_f.get_dependents():\n",
"# print(zfit.run(param))"
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {},
"outputs": [],
"source": [
"# total_f_fit.normalization(obs_fit)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Test if graphs actually work and compute values"
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {},
"outputs": [],
"source": [
"# def total_test_tf(xq):\n",
"\n",
"# def jpsi_res(q):\n",
"# return resonance(q, jpsi_m, jpsi_s, jpsi_p, jpsi_w)\n",
"\n",
"# def psi2s_res(q):\n",
"# return resonance(q, psi2s_m, psi2s_s, psi2s_p, psi2s_w)\n",
"\n",
"# def cusp(q):\n",
"# return bifur_gauss(q, cusp_m, sig_L, sig_R, cusp_s)\n",
"\n",
"# funcs = jpsi_res(xq) + psi2s_res(xq) + cusp(xq)\n",
"\n",
"# vec_f = vec(xq, funcs)\n",
"\n",
"# axiv_nr = axiv_nonres(xq)\n",
"\n",
"# tot = vec_f + axiv_nr\n",
" \n",
"# return tot\n",
"\n",
"# def jpsi_res(q):\n",
"# return resonance(q, jpsi_m, jpsi_s, jpsi_p, jpsi_w)\n",
"\n",
"# calcs = zfit.run(total_test_tf(x_part))\n",
"\n",
"\n",
"\n",
"test_q = np.linspace(x_min, x_max, int(2e6))\n",
"\n",
"probs = total_f_fit.pdf(test_q, norm_range=False)\n",
"\n",
"calcs_test = zfit.run(probs)\n",
"\n",
"Ctt.set_value(0.5)\n",
"\n",
"calcs_test1 = zfit.run(probs)\n",
"\n",
"Ctt.set_value(0.0)\n",
"\n",
"Dbar_s.set_value(0.3)\n",
"\n",
"DDstar_s.set_value(0.3)\n",
"\n",
"calcs_test2 = zfit.run(probs)\n",
"# res_y = zfit.run(jpsi_res(test_q))\n",
"# b0 = [b0_0, b0_1, b0_2]\n",
"# bplus = [bplus_0, bplus_1, bplus_2]\n",
"# bT = [bT_0, bT_1, bT_2]\n",
"# f0_y = zfit.run(tf.math.real(formfactor(test_q,\"0\", b0, bplus, bT)))\n",
"# fplus_y = zfit.run(tf.math.real(formfactor(test_q,\"+\", b0, bplus, bT)))\n",
"# fT_y = zfit.run(tf.math.real(formfactor(test_q,\"T\", b0, bplus, bT)))"
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"C:\\Users\\sa_li\\.conda\\envs\\rmd\\lib\\site-packages\\ipykernel_launcher.py:14: UserWarning: Creating legend with loc=\"best\" can be slow with large amounts of data.\n",
" \n",
"C:\\Users\\sa_li\\.conda\\envs\\rmd\\lib\\site-packages\\IPython\\core\\pylabtools.py:128: UserWarning: Creating legend with loc=\"best\" can be slow with large amounts of data.\n",
" fig.canvas.print_figure(bytes_io, **kw)\n"
]
},
{
"data": {
"image/png": "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\n",
"text/plain": [
"<Figure size 432x288 with 1 Axes>"
]
},
"metadata": {
"needs_background": "light"
},
"output_type": "display_data"
}
],
"source": [
"plt.clf()\n",
"# plt.plot(x_part, calcs, '.')\n",
"plt.plot(test_q, calcs_test, label = 'pdf (Ctt = 0.0)')\n",
"plt.plot(test_q, calcs_test1, label = 'pdf (Ctt = 0.5)')\n",
"plt.plot(test_q, calcs_test2, label = 'pdf (D-contribs = 0.3)')\n",
"# plt.plot(test_q, f0_y, label = '0')\n",
"# plt.plot(test_q, fT_y, label = 'T')\n",
"# plt.plot(test_q, fplus_y, label = '+')\n",
"# plt.plot(test_q, res_y, label = 'res')\n",
"plt.legend()\n",
"plt.ylim(0.0, 1.5e-6)\n",
"# plt.yscale('log')\n",
"# plt.xlim(770, 785)\n",
"plt.savefig('test.png')\n",
"# print(jpsi_width)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"\n",
"\n",
"# probs = mixture.prob(test_q)\n",
"# probs_np = zfit.run(probs)\n",
"# probs_np *= np.max(calcs_test) / np.max(probs_np)\n",
"# plt.figure()\n",
"# plt.semilogy(test_q, probs_np,label=\"importance sampling\")\n",
"# plt.semilogy(test_q, calcs_test, label = 'pdf')\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# 0.213/(0.00133+0.213+0.015)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Adjust scaling of different parts"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"total_f.update_integration_options(draws_per_dim=2000000, mc_sampler=None)\n",
"# inte = total_f.integrate(limits = (950., 1050.), norm_range=False)\n",
"# inte_fl = zfit.run(inte)\n",
"# print(inte_fl/4500)\n",
"# print(pdg[\"jpsi_BR\"]/pdg[\"NR_BR\"], inte_fl*pdg[\"psi2s_auc\"]/pdg[\"NR_auc\"])"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# # print(\"jpsi:\", inte_fl)\n",
"# # print(\"Increase am by factor:\", np.sqrt(pdg[\"jpsi_BR\"]/pdg[\"NR_BR\"]*pdg[\"NR_auc\"]/inte_fl))\n",
"# # print(\"New amp:\", pdg[\"jpsi\"][3]*np.sqrt(pdg[\"jpsi_BR\"]/pdg[\"NR_BR\"]*pdg[\"NR_auc\"]/inte_fl))\n",
"\n",
"# # print(\"psi2s:\", inte_fl)\n",
"# # print(\"Increase am by factor:\", np.sqrt(pdg[\"psi2s_BR\"]/pdg[\"NR_BR\"]*pdg[\"NR_auc\"]/inte_fl))\n",
"# # print(\"New amp:\", pdg[\"psi2s\"][3]*np.sqrt(pdg[\"psi2s_BR\"]/pdg[\"NR_BR\"]*pdg[\"NR_auc\"]/inte_fl))\n",
"\n",
"# name = \"phi\"\n",
"\n",
"# print(name+\":\", inte_fl)\n",
"# print(\"Increase am by factor:\", np.sqrt(pdg[name+\"_BR\"]/pdg[\"NR_BR\"]*pdg[\"NR_auc\"]/inte_fl))\n",
"# print(\"New amp:\", pdg[name][0]*np.sqrt(pdg[name+\"_BR\"]/pdg[\"NR_BR\"]*pdg[\"NR_auc\"]/inte_fl))\n",
"\n",
"\n",
"# print(x_min)\n",
"# print(x_max)\n",
"# # total_f.update_integration_options(draws_per_dim=2000000, mc_sampler=None)\n",
"# total_f.update_integration_options(mc_sampler=lambda dim, num_results,\n",
"# dtype: tf.random_uniform(maxval=1., shape=(num_results, dim), dtype=dtype),\n",
"# draws_per_dim=1000000)\n",
"# # _ = []\n",
"\n",
"# # for i in range(10):\n",
"\n",
"# # inte = total_f.integrate(limits = (x_min, x_max))\n",
"# # inte_fl = zfit.run(inte)\n",
"# # print(inte_fl)\n",
"# # _.append(inte_fl)\n",
"\n",
"# # print(\"mean:\", np.mean(_))\n",
"\n",
"# _ = time.time()\n",
"\n",
"# inte = total_f.integrate(limits = (x_min, x_max))\n",
"# inte_fl = zfit.run(inte)\n",
"# print(inte_fl)\n",
"# print(\"Time taken: {}\".format(display_time(int(time.time() - _))))\n",
"\n",
"# print(pdg['NR_BR']/pdg['NR_auc']*inte_fl)\n",
"# print(0.25**2*4.2/1000)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Sampling\n",
"## Mixture distribution for sampling"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"\n",
" \n",
"# print(list_of_borders[:9])\n",
"# print(list_of_borders[-9:])\n",
"\n",
"\n",
"class UniformSampleAndWeights(zfit.util.execution.SessionHolderMixin):\n",
" def __call__(self, limits, dtype, n_to_produce):\n",
" # n_to_produce = tf.cast(n_to_produce, dtype=tf.int32)\n",
" low, high = limits.limit1d\n",
" low = tf.cast(low, dtype=dtype)\n",
" high = tf.cast(high, dtype=dtype)\n",
"# uniform = tfd.Uniform(low=low, high=high)\n",
"# uniformjpsi = tfd.Uniform(low=tf.constant(3080, dtype=dtype), high=tf.constant(3112, dtype=dtype))\n",
"# uniformpsi2s = tfd.Uniform(low=tf.constant(3670, dtype=dtype), high=tf.constant(3702, dtype=dtype))\n",
"\n",
"# list_of_borders = []\n",
"# _p = []\n",
"# splits = 10\n",
"\n",
"# _ = np.linspace(x_min, x_max, splits)\n",
"\n",
"# for i in range(splits):\n",
"# list_of_borders.append(tf.constant(_[i], dtype=dtype))\n",
"# _p.append(tf.constant(1/splits, dtype=dtype))\n",
" \n",
"# mixture = tfd.MixtureSameFamily(mixture_distribution=tfd.Categorical(probs=_p[:(splits-1)]),\n",
"# components_distribution=tfd.Uniform(low=list_of_borders[:(splits-1)], \n",
"# high=list_of_borders[-(splits-1):]))\n",
" mixture = tfd.MixtureSameFamily(mixture_distribution=tfd.Categorical(probs=[tf.constant(0.05, dtype=dtype),\n",
" tf.constant(0.93, dtype=dtype),\n",
" tf.constant(0.05, dtype=dtype),\n",
" tf.constant(0.065, dtype=dtype),\n",
" tf.constant(0.04, dtype=dtype),\n",
" tf.constant(0.05, dtype=dtype)]),\n",
" components_distribution=tfd.Uniform(low=[tf.constant(x_min, dtype=dtype), \n",
" tf.constant(3090, dtype=dtype),\n",
" tf.constant(3681, dtype=dtype), \n",
" tf.constant(3070, dtype=dtype),\n",
" tf.constant(1000, dtype=dtype),\n",
" tf.constant(3660, dtype=dtype)], \n",
" high=[tf.constant(x_max, dtype=dtype),\n",
" tf.constant(3102, dtype=dtype), \n",
" tf.constant(3691, dtype=dtype),\n",
" tf.constant(3110, dtype=dtype),\n",
" tf.constant(1040, dtype=dtype),\n",
" tf.constant(3710, dtype=dtype)]))\n",
"# dtype = tf.float64\n",
"# mixture = tfd.MixtureSameFamily(mixture_distribution=tfd.Categorical(probs=[tf.constant(0.04, dtype=dtype),\n",
"# tf.constant(0.90, dtype=dtype),\n",
"# tf.constant(0.02, dtype=dtype),\n",
"# tf.constant(0.07, dtype=dtype),\n",
"# tf.constant(0.02, dtype=dtype)]),\n",
"# components_distribution=tfd.Uniform(low=[tf.constant(x_min, dtype=dtype), \n",
"# tf.constant(3089, dtype=dtype),\n",
"# tf.constant(3103, dtype=dtype), \n",
"# tf.constant(3681, dtype=dtype),\n",
"# tf.constant(3691, dtype=dtype)], \n",
"# high=[tf.constant(3089, dtype=dtype),\n",
"# tf.constant(3103, dtype=dtype), \n",
"# tf.constant(3681, dtype=dtype),\n",
"# tf.constant(3691, dtype=dtype), \n",
"# tf.constant(x_max, dtype=dtype)]))\n",
"# mixture = tfd.Uniform(tf.constant(x_min, dtype=dtype), tf.constant(x_max, dtype=dtype))\n",
"# sample = tf.random.uniform((n_to_produce, 1), dtype=dtype)\n",
" sample = mixture.sample((n_to_produce, 1))\n",
"# sample = tf.random.uniform((n_to_produce, 1), dtype=dtype)\n",
" weights = mixture.prob(sample)[:,0]\n",
"# weights = tf.broadcast_to(tf.constant(1., dtype=dtype), shape=(n_to_produce,))\n",
" # sample = tf.expand_dims(sample, axis=-1)\n",
"# print(sample, weights)\n",
" \n",
"# weights = tf.ones(shape=(n_to_produce,), dtype=dtype)\n",
" weights_max = None\n",
" thresholds = tf.random_uniform(shape=(n_to_produce,), dtype=dtype)\n",
" return sample, thresholds, weights, weights_max, n_to_produce"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# total_f._sample_and_weights = UniformSampleAndWeights"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# 0.00133/(0.00133+0.213+0.015)*(x_max-3750)/(x_max-x_min)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# zfit.settings.set_verbosity(10)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"scrolled": false
},
"outputs": [],
"source": [
"# # zfit.run.numeric_checks = False \n",
"\n",
"# nr_of_toys = 1\n",
"# nevents = int(pdg[\"number_of_decays\"])\n",
"# nevents = pdg[\"number_of_decays\"]\n",
"# event_stack = 1000000\n",
"# # zfit.settings.set_verbosity(10)\n",
"# calls = int(nevents/event_stack + 1)\n",
"\n",
"# total_samp = []\n",
"\n",
"# start = time.time()\n",
"\n",
"# sampler = total_f.create_sampler(n=event_stack)\n",
"\n",
"# for toy in range(nr_of_toys):\n",
" \n",
"# dirName = 'data/zfit_toys/toy_{0}'.format(toy)\n",
" \n",
"# if not os.path.exists(dirName):\n",
"# os.mkdir(dirName)\n",
"# print(\"Directory \" , dirName , \" Created \")\n",
"\n",
"# for call in range(calls):\n",
"\n",
"# sampler.resample(n=event_stack)\n",
"# s = sampler.unstack_x()\n",
"# sam = zfit.run(s)\n",
"# # clear_output(wait=True)\n",
"\n",
"# c = call + 1\n",
" \n",
"# print(\"{0}/{1} of Toy {2}/{3}\".format(c, calls, toy+1, nr_of_toys))\n",
"# print(\"Time taken: {}\".format(display_time(int(time.time() - start))))\n",
"# print(\"Projected time left: {}\".format(display_time(int((time.time() - start)/(c+calls*(toy))*((nr_of_toys-toy)*calls-c)))))\n",
"\n",
"# with open(\"data/zfit_toys/toy_{0}/{1}.pkl\".format(toy, call), \"wb\") as f:\n",
"# pkl.dump(sam, f, pkl.HIGHEST_PROTOCOL)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# with open(r\"data/zfit_toys/toy_0/0.pkl\", \"rb\") as input_file:\n",
"# sam = pkl.load(input_file)\n",
"# print(sam[:10])\n",
"\n",
"# with open(r\"data/zfit_toys/toy_0/1.pkl\", \"rb\") as input_file:\n",
"# sam2 = pkl.load(input_file)\n",
"# print(sam2[:10])\n",
"\n",
"# print(np.sum(sam-sam2))"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# print(\"Time to generate full toy: {} s\".format(int(time.time()-start)))\n",
"\n",
"# total_samp = []\n",
"\n",
"# for call in range(calls):\n",
"# with open(r\"data/zfit_toys/toy_0/{}.pkl\".format(call), \"rb\") as input_file:\n",
"# sam = pkl.load(input_file)\n",
"# total_samp = np.append(total_samp, sam)\n",
"\n",
"# total_samp = total_samp.astype('float64')\n",
"\n",
"# data2 = zfit.data.Data.from_numpy(array=total_samp[:int(nevents)], obs=obs)\n",
"\n",
"# data3 = zfit.data.Data.from_numpy(array=total_samp, obs=obs)\n",
"\n",
"# print(total_samp[:nevents].shape)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# plt.clf()\n",
"\n",
"# bins = int((x_max-x_min)/7)\n",
"\n",
"# # calcs = zfit.run(total_test_tf(samp))\n",
"# print(total_samp[:nevents].shape)\n",
"\n",
"# plt.hist(total_samp[:nevents], bins = bins, range = (x_min,x_max), label = 'data')\n",
"# # plt.plot(test_q, calcs_test*nevents , label = 'pdf')\n",
"\n",
"# # plt.plot(sam, calcs, '.')\n",
"# # plt.plot(test_q, calcs_test)\n",
"# # plt.yscale('log')\n",
"# plt.ylim(0, 200)\n",
"# # plt.xlim(3080, 3110)\n",
"\n",
"# plt.legend()\n",
"\n",
"# plt.savefig('test2.png')"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# sampler = total_f.create_sampler(n=nevents)\n",
"# nll = zfit.loss.UnbinnedNLL(model=total_f, data=sampler, fit_range = (x_min, x_max))\n",
"\n",
"# # for param in pdf.get_dependents():\n",
"# # param.set_value(initial_value)\n",
"\n",
"# sampler.resample(n=nevents)\n",
"\n",
"# # Randomise initial values\n",
"# # for param in pdf.get_dependents():\n",
"# # param.set_value(random value here)\n",
"\n",
"# # Minimise the NLL\n",
"# minimizer = zfit.minimize.MinuitMinimizer(verbosity = 10)\n",
"# minimum = minimizer.minimize(nll)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# jpsi_width"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# plt.hist(sample, weights=1 / prob(sample))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Fitting"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# start = time.time()\n",
"\n",
"# for param in total_f.get_dependents():\n",
"# param.randomize()\n",
" \n",
"# # for param in total_f.get_dependents():\n",
"# # print(zfit.run(param))\n",
" \n",
"# nll = zfit.loss.UnbinnedNLL(model=total_f, data=data2, fit_range = (x_min, x_max))\n",
"\n",
"# minimizer = zfit.minimize.MinuitMinimizer(verbosity = 5)\n",
"# # minimizer._use_tfgrad = False\n",
"# result = minimizer.minimize(nll)\n",
"\n",
"# # param_errors = result.error()\n",
"\n",
"# # for var, errors in param_errors.items():\n",
"# # print('{}: ^{{+{}}}_{{{}}}'.format(var.name, errors['upper'], errors['lower']))\n",
"\n",
"# print(\"Function minimum:\", result.fmin)\n",
"# # print(\"Results:\", result.params)\n",
"# print(\"Hesse errors:\", result.hesse())"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# print(\"Time taken for fitting: {}\".format(display_time(int(time.time()-start))))\n",
"\n",
"# # probs = total_f.pdf(test_q)\n",
"\n",
"# calcs_test = zfit.run(probs)\n",
"# res_y = zfit.run(jpsi_res(test_q))"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# plt.clf()\n",
"# # plt.plot(x_part, calcs, '.')\n",
"# plt.plot(test_q, calcs_test, label = 'pdf')\n",
"# # plt.plot(test_q, res_y, label = 'res')\n",
"# plt.legend()\n",
"# plt.ylim(0.0, 10e-6)\n",
"# # plt.yscale('log')\n",
"# # plt.xlim(3080, 3110)\n",
"# plt.savefig('test3.png')\n",
"# # print(jpsi_width)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# _tot = 4.37e-7+6.02e-5+4.97e-6\n",
"# _probs = []\n",
"# _probs.append(6.02e-5/_tot)\n",
"# _probs.append(4.97e-6/_tot)\n",
"# _probs.append(4.37e-7/_tot)\n",
"# print(_probs)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# dtype = 'float64'\n",
"# # mixture = tfd.Uniform(tf.constant(x_min, dtype=dtype), tf.constant(x_max, dtype=dtype))\n",
"# mixture = tfd.MixtureSameFamily(mixture_distribution=tfd.Categorical(probs=[tf.constant(0.007, dtype=dtype),\n",
"# tf.constant(0.917, dtype=dtype),\n",
"# tf.constant(0.076, dtype=dtype)]),\n",
"# components_distribution=tfd.Uniform(low=[tf.constant(x_min, dtype=dtype), \n",
"# tf.constant(3080, dtype=dtype),\n",
"# tf.constant(3670, dtype=dtype)], \n",
"# high=[tf.constant(x_max, dtype=dtype),\n",
"# tf.constant(3112, dtype=dtype), \n",
"# tf.constant(3702, dtype=dtype)]))\n",
"# # for i in range(10):\n",
"# # print(zfit.run(mixture.prob(mixture.sample((10, 1)))))"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# print((zfit.run(jpsi_p)%(2*np.pi))/np.pi)\n",
"# print((zfit.run(psi2s_p)%(2*np.pi))/np.pi)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# def jpsi_res(q):\n",
"# return resonance(q, _mass = jpsi_mass, scale = jpsi_scale,\n",
"# phase = jpsi_phase, width = jpsi_width)\n",
"\n",
"# def psi2s_res(q):\n",
"# return resonance(q, _mass = psi2s_mass, scale = psi2s_scale,\n",
"# phase = psi2s_phase, width = psi2s_width)\n",
" \n",
"# def p3770_res(q):\n",
"# return resonance(q, _mass = p3770_mass, scale = p3770_scale,\n",
"# phase = p3770_phase, width = p3770_width)\n",
" \n",
"# def p4040_res(q):\n",
"# return resonance(q, _mass = p4040_mass, scale = p4040_scale,\n",
"# phase = p4040_phase, width = p4040_width)\n",
" \n",
"# def p4160_res(q):\n",
"# return resonance(q, _mass = p4160_mass, scale = p4160_scale,\n",
"# phase = p4160_phase, width = p4160_width)\n",
" \n",
"# def p4415_res(q):\n",
"# return resonance(q, _mass = p4415_mass, scale = p4415_scale,\n",
"# phase = p4415_phase, width = p4415_width)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# 0.15**2*4.2/1000\n",
"# result.hesse()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Constraints"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# 1. Constraint - Real part of sum of Psi contrib and D contribs\n",
"\n",
"sum_list = []\n",
"\n",
"sum_list.append(ztf.to_complex(jpsi_s) * tf.exp(tf.complex(ztf.constant(0.0), jpsi_p)) * ztf.to_complex(jpsi_w / (tf.pow(jpsi_m,3))))\n",
"sum_list.append(ztf.to_complex(psi2s_s) * tf.exp(tf.complex(ztf.constant(0.0), psi2s_p)) * ztf.to_complex(psi2s_w / (tf.pow(psi2s_m,3))))\n",
"sum_list.append(ztf.to_complex(p3770_s) * tf.exp(tf.complex(ztf.constant(0.0), p3770_p)) * ztf.to_complex(p3770_w / (tf.pow(p3770_m,3))))\n",
"sum_list.append(ztf.to_complex(p4040_s) * tf.exp(tf.complex(ztf.constant(0.0), p4040_p)) * ztf.to_complex(p4040_w / (tf.pow(p4040_m,3))))\n",
"sum_list.append(ztf.to_complex(p4160_s) * tf.exp(tf.complex(ztf.constant(0.0), p4160_p)) * ztf.to_complex(p4160_w / (tf.pow(p4160_m,3))))\n",
"sum_list.append(ztf.to_complex(p4415_s) * tf.exp(tf.complex(ztf.constant(0.0), p4415_p)) * ztf.to_complex(p4415_w / (tf.pow(p4415_m,3))))\n",
"sum_list.append(ztf.to_complex(DDstar_s) * tf.exp(tf.complex(ztf.constant(0.0), DDstar_p)) * (ztf.to_complex(1.0 / (10.0*tf.pow(Dstar_m,2)) + 1.0 / (10.0*tf.pow(D_m,2)))))\n",
"sum_list.append(ztf.to_complex(Dbar_s) * tf.exp(tf.complex(ztf.constant(0.0), Dbar_p)) * ztf.to_complex(1.0 / (6.0*tf.pow(Dbar_m,2))))\n",
"\n",
"sum_ru_1 = ztf.to_complex(ztf.constant(0.0))\n",
"\n",
"for part in sum_list:\n",
" sum_ru_1 += part\n",
"\n",
"sum_1 = tf.math.real(sum_ru_1)\n",
"# constraint1 = zfit.constraint.GaussianConstraint(params = sum_1, mu = ztf.constant(1.7*10**-8), \n",
"# sigma = ztf.constant(2.2*10**-8))\n",
"\n",
"constraint1 = tf.pow((sum_1-ztf.constant(1.7*10**-8))/ztf.constant(2.2*10**-8),2)/ztf.constant(2.)\n",
"\n",
"# 2. Constraint - Abs. of sum of Psi contribs and D contribs\n",
"\n",
"sum_2 = tf.abs(sum_ru_1)\n",
"constraint2 = tf.cond(tf.greater_equal(sum_2, 5.0e-8), lambda: 100000., lambda: 0.)\n",
"\n",
"# 3. Constraint - Maximum eta of D contribs\n",
"\n",
"constraint3_0 = tf.cond(tf.greater_equal(tf.abs(Dbar_s), 0.2), lambda: 100000., lambda: 0.)\n",
"\n",
"constraint3_1 = tf.cond(tf.greater_equal(tf.abs(DDstar_s), 0.2), lambda: 100000., lambda: 0.)\n",
"\n",
"# 4. Constraint - Formfactor multivariant gaussian covariance fplus\n",
"\n",
"Cov_matrix = [[ztf.constant( 1.), ztf.constant( 0.45), ztf.constant( 0.19), ztf.constant(0.857), ztf.constant(0.598), ztf.constant(0.531), ztf.constant(0.752), ztf.constant(0.229), ztf.constant(0,117)],\n",
" [ztf.constant( 0.45), ztf.constant( 1.), ztf.constant(0.677), ztf.constant(0.708), ztf.constant(0.958), ztf.constant(0.927), ztf.constant(0.227), ztf.constant(0.443), ztf.constant(0.287)],\n",
" [ztf.constant( 0.19), ztf.constant(0.677), ztf.constant( 1.), ztf.constant(0.595), ztf.constant(0.770), ztf.constant(0.819),ztf.constant(-0.023), ztf.constant( 0.07), ztf.constant(0.196)],\n",
" [ztf.constant(0.857), ztf.constant(0.708), ztf.constant(0.595), ztf.constant( 1.), ztf.constant( 0.83), ztf.constant(0.766), ztf.constant(0.582), ztf.constant(0.237), ztf.constant(0.192)],\n",
" [ztf.constant(0.598), ztf.constant(0.958), ztf.constant(0.770), ztf.constant( 0.83), ztf.constant( 1.), ztf.constant(0.973), ztf.constant(0.324), ztf.constant(0.372), ztf.constant(0.272)],\n",
" [ztf.constant(0.531), ztf.constant(0.927), ztf.constant(0.819), ztf.constant(0.766), ztf.constant(0.973), ztf.constant( 1.), ztf.constant(0.268), ztf.constant(0.332), ztf.constant(0.269)],\n",
" [ztf.constant(0.752), ztf.constant(0.227),ztf.constant(-0.023), ztf.constant(0.582), ztf.constant(0.324), ztf.constant(0.268), ztf.constant( 1.), ztf.constant( 0.59), ztf.constant(0.515)],\n",
" [ztf.constant(0.229), ztf.constant(0.443), ztf.constant( 0.07), ztf.constant(0.237), ztf.constant(0.372), ztf.constant(0.332), ztf.constant( 0.59), ztf.constant( 1.), ztf.constant(0.897)],\n",
" [ztf.constant(0.117), ztf.constant(0.287), ztf.constant(0.196), ztf.constant(0.192), ztf.constant(0.272), ztf.constant(0.269), ztf.constant(0.515), ztf.constant(0.897), ztf.constant( 1.)]]\n",
"\n",
"def triGauss(val1,val2,val3,m = Cov_matrix):\n",
"\n",
" mean1 = ztf.constant(0.466)\n",
" mean2 = ztf.constant(-0.885)\n",
" mean3 = ztf.constant(-0.213)\n",
" sigma1 = ztf.constant(0.014)\n",
" sigma2 = ztf.constant(0.128)\n",
" sigma3 = ztf.constant(0.548)\n",
" x1 = (val1-mean1)/sigma1\n",
" x2 = (val2-mean2)/sigma2\n",
" x3 = (val3-mean3)/sigma3\n",
" rho12 = m[0][1]\n",
" rho13 = m[0][2]\n",
" rho23 = m[1][2]\n",
" w = x1*x1*(rho23*rho23-1) + x2*x2*(rho13*rho13-1)+x3*x3*(rho12*rho12-1)+2*(x1*x2*(rho12-rho13*rho23)+x1*x3*(rho13-rho12*rho23)+x2*x3*(rho23-rho12*rho13))\n",
" d = 2*(rho12*rho12+rho13*rho13+rho23*rho23-2*rho12*rho13*rho23-1)\n",
" \n",
" fcn = -w/d\n",
" chisq = -2*fcn\n",
" return chisq\n",
"\n",
"constraint4 = triGauss(bplus_0, bplus_1, bplus_2)\n",
"\n",
"# mean1 = ztf.constant(0.466)\n",
"# mean2 = ztf.constant(-0.885)\n",
"# mean3 = ztf.constant(-0.213)\n",
"# sigma1 = ztf.constant(0.014)\n",
"# sigma2 = ztf.constant(0.128)\n",
"# sigma3 = ztf.constant(0.548)\n",
"# constraint4_0 = tf.pow((bplus_0-mean1)/sigma1,2)/ztf.constant(2.)\n",
"# constraint4_1 = tf.pow((bplus_1-mean2)/sigma2,2)/ztf.constant(2.)\n",
"# constraint4_2 = tf.pow((bplus_2-mean3)/sigma3,2)/ztf.constant(2.)\n",
"\n",
"# 5. Constraint - Abs. of sum of light contribs\n",
"\n",
"sum_list_5 = []\n",
"\n",
"sum_list_5.append(rho_s*rho_w/rho_m)\n",
"sum_list_5.append(omega_s*omega_w/omega_m)\n",
"sum_list_5.append(phi_s*phi_w/phi_m)\n",
"\n",
"\n",
"sum_ru_5 = ztf.constant(0.0)\n",
"\n",
"for part in sum_list_5:\n",
" sum_ru_5 += part\n",
"\n",
"constraint5 = tf.cond(tf.greater_equal(tf.abs(sum_ru_5), ztf.constant(0.02)), lambda: 100000., lambda: 0.)\n",
"\n",
"# 6. Constraint on phases of Jpsi and Psi2s for cut out fit\n",
"\n",
"\n",
"# constraint6_0 = zfit.constraint.GaussianConstraint(params = jpsi_p, mu = ztf.constant(pdg[\"jpsi_phase_unc\"]),\n",
"# sigma = ztf.constant(jpsi_phase))\n",
"# constraint6_1 = zfit.constraint.GaussianConstraint(params = psi2s_p, mu = ztf.constant(pdg[\"psi2s_phase_unc\"]),\n",
"# sigma = ztf.constant(psi2s_phase))\n",
"\n",
"constraint6_0 = tf.pow((jpsi_p-ztf.constant(jpsi_phase))/ztf.constant(pdg[\"jpsi_phase_unc\"]),2)/ztf.constant(2.)\n",
"constraint6_1 = tf.pow((psi2s_p-ztf.constant(psi2s_phase))/ztf.constant(pdg[\"psi2s_phase_unc\"]),2)/ztf.constant(2.)\n",
"\n",
"# 7. Constraint on Ctt with higher limits\n",
"\n",
"constraint7 = tf.cond(tf.greater_equal(Ctt*Ctt, 0.25), lambda: 100000., lambda: 0.)\n",
"\n",
"constraint7dtype = tf.float64\n",
"\n",
"# zfit.run(constraint6_0)\n",
"\n",
"# ztf.convert_to_tensor(constraint6_0)\n",
"\n",
"#List of all constraints\n",
"\n",
"constraints = [constraint1, constraint2, constraint3_0, constraint3_1,# constraint4, #constraint4_0, constraint4_1, constraint4_2,\n",
" constraint6_0, constraint6_1]#, constraint7]"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Reset params"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"param_values_dic = {\n",
" 'jpsi_m': jpsi_mass,\n",
" 'jpsi_s': jpsi_scale,\n",
" 'jpsi_p': jpsi_phase,\n",
" 'jpsi_w': jpsi_width,\n",
" 'psi2s_m': psi2s_mass,\n",
" 'psi2s_s': psi2s_scale,\n",
" 'psi2s_p': psi2s_phase,\n",
" 'psi2s_w': psi2s_width,\n",
" 'p3770_m': p3770_mass,\n",
" 'p3770_s': p3770_scale,\n",
" 'p3770_p': p3770_phase,\n",
" 'p3770_w': p3770_width,\n",
" 'p4040_m': p4040_mass,\n",
" 'p4040_s': p4040_scale,\n",
" 'p4040_p': p4040_phase,\n",
" 'p4040_w': p4040_width,\n",
" 'p4160_m': p4160_mass,\n",
" 'p4160_s': p4160_scale,\n",
" 'p4160_p': p4160_phase,\n",
" 'p4160_w': p4160_width,\n",
" 'p4415_m': p4415_mass,\n",
" 'p4415_s': p4415_scale,\n",
" 'p4415_p': p4415_phase,\n",
" 'p4415_w': p4415_width,\n",
" 'rho_m': rho_mass,\n",
" 'rho_s': rho_scale,\n",
" 'rho_p': rho_phase,\n",
" 'rho_w': rho_width,\n",
" 'omega_m': omega_mass,\n",
" 'omega_s': omega_scale,\n",
" 'omega_p': omega_phase,\n",
" 'omega_w': omega_width,\n",
" 'phi_m': phi_mass,\n",
" 'phi_s': phi_scale,\n",
" 'phi_p': phi_phase,\n",
" 'phi_w': phi_width,\n",
" 'Dstar_m': Dstar_mass,\n",
" 'DDstar_s': 0.0,\n",
" 'DDstar_p': 0.0,\n",
" 'D_m': D_mass,\n",
" 'Dbar_m': Dbar_mass,\n",
" 'Dbar_s': 0.0,\n",
" 'Dbar_p': 0.0,\n",
" 'tau_m': pdg['tau_M'],\n",
" 'Ctt': 0.0,\n",
" 'b0_0': 0.292,\n",
" 'b0_1': 0.281,\n",
" 'b0_2': 0.150,\n",
" 'bplus_0': 0.466,\n",
" 'bplus_1': -0.885,\n",
" 'bplus_2': -0.213,\n",
" 'bT_0': 0.460,\n",
" 'bT_1': -1.089,\n",
" 'bT_2': -1.114}\n",
"\n",
"\n",
"def reset_param_values(variation = 0.05):\n",
" for param in total_f_fit.get_dependents():\n",
" if param.floating:\n",
" param.set_value(param_values_dic[param.name] + random.uniform(-1, 1) * param_values_dic[param.name]* variation)\n",
"# print(param.name)\n",
"# for param in totalf.get_dependents():\n",
"# param.set_value()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Analysis"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"scrolled": false
},
"outputs": [],
"source": [
"# # zfit.run.numeric_checks = False \n",
"\n",
"# fitting_range = 'cut'\n",
"# total_BR = 1.7e-10 + 4.9e-10 + 2.5e-9 + 6.02e-5 + 4.97e-6 + 1.38e-9 + 4.2e-10 + 2.6e-9 + 6.1e-10 + 4.37e-7\n",
"# cut_BR = 1.0 - (6.02e-5 + 4.97e-6)/total_BR\n",
"\n",
"# Ctt_list = []\n",
"# Ctt_error_list = []\n",
"\n",
"# nr_of_toys = 1\n",
"# if fitting_range == 'cut':\n",
"# nevents = int(pdg[\"number_of_decays\"]*cut_BR)\n",
"# else:\n",
"# nevents = int(pdg[\"number_of_decays\"])\n",
"# # nevents = pdg[\"number_of_decays\"]\n",
"# event_stack = 1000000\n",
"# # nevents *= 41\n",
"# # zfit.settings.set_verbosity(10)\n",
"# calls = int(nevents/event_stack + 1)\n",
"\n",
"# total_samp = []\n",
"\n",
"# start = time.time()\n",
"\n",
"# sampler = total_f.create_sampler(n=event_stack)\n",
"\n",
"# for toy in range(nr_of_toys):\n",
" \n",
"# ### Generate data\n",
" \n",
"# # clear_output(wait=True)\n",
" \n",
"# print(\"Toy {}: Generating data...\".format(toy))\n",
" \n",
"# dirName = 'data/zfit_toys/toy_{0}'.format(toy)\n",
" \n",
"# if not os.path.exists(dirName):\n",
"# os.mkdir(dirName)\n",
"# print(\"Directory \" , dirName , \" Created \")\n",
" \n",
"# reset_param_values()\n",
" \n",
"# if fitting_range == 'cut':\n",
" \n",
"# sampler.resample(n=nevents)\n",
"# s = sampler.unstack_x()\n",
"# sam = zfit.run(s)\n",
"# calls = 0\n",
"# c = 1\n",
" \n",
"# else: \n",
"# for call in range(calls):\n",
"\n",
"# sampler.resample(n=event_stack)\n",
"# s = sampler.unstack_x()\n",
"# sam = zfit.run(s)\n",
"\n",
"# c = call + 1\n",
"\n",
"# with open(\"data/zfit_toys/toy_{0}/{1}.pkl\".format(toy, call), \"wb\") as f:\n",
"# pkl.dump(sam, f, pkl.HIGHEST_PROTOCOL)\n",
" \n",
"# print(\"Toy {}: Data generation finished\".format(toy))\n",
" \n",
"# ### Load data\n",
" \n",
"# print(\"Toy {}: Loading data...\".format(toy))\n",
" \n",
"# if fitting_range == 'cut':\n",
" \n",
"# total_samp = sam\n",
" \n",
"# else:\n",
" \n",
"# for call in range(calls):\n",
"# with open(r\"data/zfit_toys/toy_0/{}.pkl\".format(call), \"rb\") as input_file:\n",
"# sam = pkl.load(input_file)\n",
"# total_samp = np.append(total_samp, sam)\n",
"\n",
"# total_samp = total_samp.astype('float64')\n",
" \n",
"# if fitting_range == 'full':\n",
"\n",
"# data = zfit.data.Data.from_numpy(array=total_samp[:int(nevents)], obs=obs)\n",
" \n",
"# print(\"Toy {}: Loading data finished\".format(toy))\n",
"\n",
"# ### Fit data\n",
"\n",
"# print(\"Toy {}: Fitting pdf...\".format(toy))\n",
"\n",
"# for param in total_f.get_dependents():\n",
"# param.randomize()\n",
"\n",
"# nll = zfit.loss.UnbinnedNLL(model=total_f, data=data, fit_range = (x_min, x_max), constraints = constraints)\n",
"\n",
"# minimizer = zfit.minimize.MinuitMinimizer(verbosity = 5)\n",
"# # minimizer._use_tfgrad = False\n",
"# result = minimizer.minimize(nll)\n",
"\n",
"# print(\"Toy {}: Fitting finished\".format(toy))\n",
"\n",
"# print(\"Function minimum:\", result.fmin)\n",
"# print(\"Hesse errors:\", result.hesse())\n",
"\n",
"# params = result.params\n",
"# Ctt_list.append(params[Ctt]['value'])\n",
"# Ctt_error_list.append(params[Ctt]['minuit_hesse']['error'])\n",
"\n",
"# #plotting the result\n",
"\n",
"# plotdirName = 'data/plots'.format(toy)\n",
"\n",
"# if not os.path.exists(plotdirName):\n",
"# os.mkdir(plotdirName)\n",
"# # print(\"Directory \" , dirName , \" Created \")\n",
" \n",
"# probs = total_f.pdf(test_q, norm_range=False)\n",
"# calcs_test = zfit.run(probs)\n",
"# plt.clf()\n",
"# plt.plot(test_q, calcs_test, label = 'pdf')\n",
"# plt.legend()\n",
"# plt.ylim(0.0, 6e-6)\n",
"# plt.savefig(plotdirName + '/toy_fit_full_range{}.png'.format(toy))\n",
"\n",
"# print(\"Toy {0}/{1}\".format(toy+1, nr_of_toys))\n",
"# print(\"Time taken: {}\".format(display_time(int(time.time() - start))))\n",
"# print(\"Projected time left: {}\".format(display_time(int((time.time() - start)/(c+calls*(toy))*((nr_of_toys-toy)*calls-c)))))\n",
" \n",
"# if fitting_range == 'cut':\n",
" \n",
"# _1 = np.where((total_samp >= x_min) & (total_samp <= (jpsi_mass - 60.)))\n",
" \n",
"# tot_sam_1 = total_samp[_1]\n",
" \n",
"# _2 = np.where((total_samp >= (jpsi_mass + 70.)) & (total_samp <= (psi2s_mass - 50.)))\n",
" \n",
"# tot_sam_2 = total_samp[_2]\n",
"\n",
"# _3 = np.where((total_samp >= (psi2s_mass + 50.)) & (total_samp <= x_max))\n",
" \n",
"# tot_sam_3 = total_samp[_3]\n",
"\n",
"# tot_sam = np.append(tot_sam_1, tot_sam_2)\n",
"# tot_sam = np.append(tot_sam, tot_sam_3)\n",
" \n",
"# data = zfit.data.Data.from_numpy(array=tot_sam[:int(nevents)], obs=obs_fit)\n",
" \n",
"# print(\"Toy {}: Loading data finished\".format(toy))\n",
" \n",
"# ### Fit data\n",
"\n",
"# print(\"Toy {}: Fitting pdf...\".format(toy))\n",
"\n",
"# for param in total_f_fit.get_dependents():\n",
"# param.randomize()\n",
"\n",
"# nll = zfit.loss.UnbinnedNLL(model=total_f_fit, data=data, constraints = constraints)\n",
"\n",
"# minimizer = zfit.minimize.MinuitMinimizer(verbosity = 5)\n",
"# # minimizer._use_tfgrad = False\n",
"# result = minimizer.minimize(nll)\n",
"\n",
"# print(\"Function minimum:\", result.fmin)\n",
"# print(\"Hesse errors:\", result.hesse())\n",
"\n",
"# params = result.params\n",
" \n",
"# if result.converged:\n",
"# Ctt_list.append(params[Ctt]['value'])\n",
"# Ctt_error_list.append(params[Ctt]['minuit_hesse']['error'])\n",
"\n",
"# #plotting the result\n",
"\n",
"# plotdirName = 'data/plots'.format(toy)\n",
"\n",
"# if not os.path.exists(plotdirName):\n",
"# os.mkdir(plotdirName)\n",
"# # print(\"Directory \" , dirName , \" Created \")\n",
" \n",
"# plt.clf()\n",
"# plt.hist(tot_sam, bins = int((x_max-x_min)/7.), label = 'toy data')\n",
"# plt.savefig(plotdirName + '/toy_histo_cut_region{}.png'.format(toy))\n",
"\n",
" \n",
"# probs = total_f_fit.pdf(test_q, norm_range=False)\n",
"# calcs_test = zfit.run(probs)\n",
"# plt.clf()\n",
"# plt.plot(test_q, calcs_test, label = 'pdf')\n",
"# plt.axvline(x=jpsi_mass-60.,color='red', linewidth=0.7, linestyle = 'dotted')\n",
"# plt.axvline(x=jpsi_mass+70.,color='red', linewidth=0.7, linestyle = 'dotted')\n",
"# plt.axvline(x=psi2s_mass-50.,color='red', linewidth=0.7, linestyle = 'dotted')\n",
"# plt.axvline(x=psi2s_mass+50.,color='red', linewidth=0.7, linestyle = 'dotted')\n",
"# plt.legend()\n",
"# plt.ylim(0.0, 1.5e-6)\n",
"# plt.savefig(plotdirName + '/toy_fit_cut_region{}.png'.format(toy))\n",
" \n",
"# print(\"Toy {0}/{1}\".format(toy+1, nr_of_toys))\n",
"# print(\"Time taken: {}\".format(display_time(int(time.time() - start))))\n",
"# print(\"Projected time left: {}\".format(display_time(int((time.time() - start)/(toy+1))*((nr_of_toys-toy-1)))))\n",
" "
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# with open(\"data/results/Ctt_list.pkl\", \"wb\") as f:\n",
"# pkl.dump(Ctt_list, f, pkl.HIGHEST_PROTOCOL)\n",
"# with open(\"data/results/Ctt_error_list.pkl\", \"wb\") as f:\n",
"# pkl.dump(Ctt_error_list, f, pkl.HIGHEST_PROTOCOL)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# print('{0}/{1} fits converged'.format(len(Ctt_list), nr_of_toys))\n",
"# print('Mean Ctt value = {}'.format(np.mean(Ctt_list)))\n",
"# print('Mean Ctt error = {}'.format(np.mean(Ctt_error_list)))\n",
"# print('95 Sensitivy = {}'.format(((2*np.mean(Ctt_error_list))**2)*4.2/1000))"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# plt.hist(tot_sam, bins = int((x_max-x_min)/7.))\n",
"\n",
"# plt.show()\n",
"\n",
"# # _ = np.where((total_samp >= x_min) & (total_samp <= (jpsi_mass - 50.)))\n",
"\n",
"# tot_sam.shape"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Ctt.floating = False"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# zfit.run(nll.value())"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# result.fmin"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# BR_steps = np.linspace(0.0, 1e-3, 11)\n",
"pull_dic = {}\n",
"\n",
"mi = 0.0\n",
"ma = 1e-2\n",
"ste = 11\n",
"\n",
"for param in total_f_fit.get_dependents():\n",
" if param.floating:\n",
" pull_dic[param.name] = []\n",
" for step in range(2*ste):\n",
" pull_dic[param.name].append([])\n",
" \n",
"\n",
"\n",
"def save_pulls(step):\n",
" for param in total_f_fit.get_dependents():\n",
" if param.floating:\n",
" pull_dic[param.name][step].append((params[param]['value'] - param_values_dic[param.name])/params[param]['minuit_hesse']['error'])\n",
"\n",
"\n",
"\n",
"# for key in pull_dic.keys():\n",
"# print(np.shape(pull_dic[key]))\n",
"# save_pulls(New_step=True)\n",
"# params[Ctt]['value']"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# for param in total_f_fit.get_dependents():\n",
"# if param.floating:\n",
"# print(param.name)\n",
"\n",
"# print(params[Ctt])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# CLS Code"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"scrolled": true
},
"outputs": [],
"source": [
"# zfit.run.numeric_checks = False \n",
"\n",
"load = True\n",
"\n",
"bo = True\n",
"\n",
"bo_set = 1\n",
"\n",
"fitting_range = 'cut'\n",
"total_BR = 1.7e-10 + 4.9e-10 + 2.5e-9 + 6.02e-5 + 4.97e-6 + 1.38e-9 + 4.2e-10 + 2.6e-9 + 6.1e-10 + 4.37e-7\n",
"cut_BR = 1.0 - (6.02e-5 + 4.97e-6)/total_BR\n",
"\n",
"Ctt_list = []\n",
"Ctt_error_list = []\n",
"\n",
"nr_of_toys = 1\n",
"nevents = int(pdg[\"number_of_decays\"]*cut_BR)\n",
"# nevents = pdg[\"number_of_decays\"]\n",
"event_stack = 1000000\n",
"nevents *= 41\n",
"# zfit.settings.set_verbosity(10)\n",
"\n",
"# mi = 0.0\n",
"# ma = 1e-3\n",
"# ste = 11\n",
"\n",
"BR_steps = np.linspace(mi, ma, ste)\n",
"\n",
"Ctt_steps = np.sqrt(BR_steps/4.2*1000)\n",
"\n",
"print(Ctt_steps)\n",
"\n",
"# total_samp = []\n",
"\n",
"start = time.time()\n",
"\n",
"Nll_list = []\n",
"\n",
"sampler = total_f.create_sampler(n=nevents)\n",
"sampler.set_data_range(obs_fit)\n",
"\n",
"__ = -1\n",
"\n",
"#-----------------------------------------------------\n",
"\n",
"if not load:\n",
" for Ctt_step in Ctt_steps:\n",
" \n",
" __ += 1\n",
" \n",
" for i in range(2):\n",
" Ctt_list.append([])\n",
" Ctt_error_list.append([])\n",
" Nll_list.append([])\n",
"\n",
" for param in total_f_fit.get_dependents():\n",
" if param.floating:\n",
" pull_dic[param.name].append([])\n",
" \n",
" for toy in range(nr_of_toys): \n",
" \n",
" newset = True\n",
" \n",
" while newset:\n",
" \n",
" for floaty in [True, False]:\n",
" Ctt.floating = floaty\n",
" \n",
" for bo_step in range(bo_set):\n",
"\n",
" print('Step: {0}/{1}'.format(int(__), ste))\n",
" print('Current Ctt: {0}'.format(Ctt_step))\n",
" print('Ctt floating: {0}'.format(floaty))\n",
" \n",
" reset_param_values(variation = 0.0)\n",
"\n",
" if floaty:\n",
" print('Toy {0}/{1} - Fit {2}/{3}'.format(toy, nr_of_toys, bo_step, bo_set))\n",
" Ctt.set_value(Ctt_step)\n",
"\n",
" else:\n",
" Ctt.set_value(0.0)\n",
" print('Toy {0}/{1} - Fit {2}/{3}'.format(toy, nr_of_toys, bo_step, bo_set))\n",
"\n",
" if newset:\n",
" sampler.resample(n=nevents)\n",
" data = sampler\n",
" newset = False\n",
"\n",
" ### Fit data\n",
"\n",
" nll = zfit.loss.UnbinnedNLL(model=total_f_fit, data=data, constraints = constraints)\n",
"\n",
" minimizer = zfit.minimize.MinuitMinimizer(verbosity = 5)\n",
" # minimizer._use_tfgrad = False\n",
" result = minimizer.minimize(nll)\n",
"\n",
" print(\"Function minimum:\", result.fmin)\n",
" print(\"Hesse errors:\", result.hesse())\n",
"\n",
" params = result.params\n",
"\n",
" if result.converged:\n",
" if floaty:\n",
" _step = 2*__\n",
" else:\n",
" _step = 2*__+1\n",
" \n",
" save_pulls(step = _step)\n",
"\n",
" if floaty:\n",
" Nll_list[-2].append(result.fmin)\n",
" Ctt_list[-2].append(params[Ctt]['value'])\n",
" Ctt_error_list[-2].append(params[Ctt]['minuit_hesse']['error'])\n",
"\n",
" else:\n",
" Nll_list[-1].append(result.fmin)\n",
" Ctt_list[-1].append(0.0)\n",
" Ctt_error_list[-1].append(0.0)\n",
" \n",
"\n",
" else:\n",
" for _ in [1,2]:\n",
" del Nll_list[-_][toy*bo_set:]\n",
"# print(np.shape(Nll_list[-_]))\n",
" del Ctt_list[-_][toy*bo_set:]\n",
" del Ctt_error_list[-_][toy*bo_set:]\n",
" for param in total_f_fit.get_dependents():\n",
" if param.floating:\n",
" del pull_dic[param.name][_step][toy*bo_set:]\n",
" newset = True\n",
" break\n",
" \n",
" if not result.converged:\n",
" break\n",
" \n",
" print()\n",
" print('Time taken: {}'.format(display_time(int(time.time()-start))))\n",
" print('Estimated time left: {}'.format(display_time(int((time.time()-start)/(__+(toy+1)/nr_of_toys)*(ste-__-(nr_of_toys-toy-1)/nr_of_toys)))))"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"if load:\n",
" \n",
" phase_combi = '-+'\n",
"\n",
" _dir = 'data/CLs/finished/f1d1/{}/'.format(phase_combi)\n",
" \n",
" jobs = os.listdir(_dir)\n",
" \n",
" First = True\n",
" \n",
"# print(jobs)\n",
" \n",
" for job in jobs:\n",
" \n",
" dirName = _dir + str(job) + '/data/CLs'\n",
" \n",
" if not os.path.exists(\"{}/{}-{}_{}s{}b{}t--CLs_Nll_list.pkl\".format(dirName, mi,ma,ste,bo_set,nr_of_toys)):\n",
"# print(job)\n",
" continue\n",
" \n",
" with open(r\"{}/variab.pkl\".format(dirName), \"rb\") as input_file:\n",
" variab = pkl.load(input_file)\n",
"# print(variab)\n",
" \n",
" ### sanity check:\n",
" if variab['mi'] != mi or variab['ma'] != ma or variab['ste'] != ste or bo_set != bo_set:\n",
" print('Fitting parameters of data dont equal the ones given -- Job {} skipped!'.format(job))\n",
" \n",
" with open(r\"{}/{}-{}_{}s{}b{}t--CLs_Nll_list.pkl\".format(dirName, mi,ma,ste,bo_set,nr_of_toys), \"rb\") as input_file:\n",
" _Nll_list = pkl.load(input_file)\n",
" \n",
" with open(r\"{}/{}-{}_{}s{}b{}t--Ctt_list.pkl\".format(dirName, mi,ma,ste,bo_set,nr_of_toys), \"rb\") as input_file:\n",
" _Ctt_list = pkl.load(input_file)\n",
" \n",
" with open(r\"{}/{}-{}_{}s{}b{}t--Ctt_error_list.pkl\".format(dirName, mi,ma,ste,bo_set,nr_of_toys), \"rb\") as input_file:\n",
" _Ctt_error_list = pkl.load(input_file)\n",
" \n",
" with open(r\"{}/{}-{}_{}s{}b{}t--pull_dic.pkl\".format(dirName, mi,ma,ste,bo_set,nr_of_toys), \"rb\") as input_file:\n",
" _pull_dic = pkl.load(input_file)\n",
" \n",
" with open(r\"{}/{}-{}_{}s--CLs_list.pkl\".format(dirName, mi,ma,ste), \"rb\") as input_file:\n",
" _CLs_list = pkl.load(input_file)\n",
" \n",
" \n",
" if First:\n",
" Nll_list = _Nll_list\n",
" Ctt_list = _Ctt_list\n",
" Ctt_error_list = _Ctt_error_list\n",
" pull_dic = _pull_dic\n",
"# print(_pull_dic)\n",
" CLs_list = _CLs_list\n",
" First = False\n",
" else:\n",
" for step in range(2*ste):\n",
"# print(Nll_list[step], step)\n",
" Nll_list[step].extend(_Nll_list[step])\n",
" Ctt_list[step].extend(_Ctt_list[step])\n",
" Ctt_error_list[step].extend(_Ctt_error_list[step])\n",
" for key in pull_dic.keys():\n",
"# print(key, np.shape(pull_dic[key]))\n",
" pull_dic[key][step].extend(_pull_dic[key][step])\n",
" for step in range(ste):\n",
" CLs_list[step].extend(_CLs_list[step])\n",
"\n",
"# print('----------------------')"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"dirName = 'data/CLs'\n",
"\n",
"# if bo and not load:\n",
"# for s in range(2*ste):\n",
"# Nll_list[s] = [np.min(Nll_list[s])]\n",
"\n",
"\n",
"if not load:\n",
" \n",
" if not os.path.exists(dirName):\n",
" os.mkdir(dirName)\n",
" print(\"Directory \" , dirName , \" Created \")\n",
"\n",
" with open(\"{}/{}-{}_{}s{}b{}t--CLs_Nll_list.pkl\".format(dirName, mi,ma,ste,bo_set,nr_of_toys), \"wb\") as f:\n",
" pkl.dump(Nll_list, f, pkl.HIGHEST_PROTOCOL)\n",
" \n",
" with open(\"{}/{}-{}_{}s{}b{}t--Ctt_list.pkl\".format(dirName, mi,ma,ste,bo_set,nr_of_toys), \"wb\") as f:\n",
" pkl.dump(Ctt_list, f, pkl.HIGHEST_PROTOCOL)\n",
" \n",
" with open(\"{}/{}-{}_{}s{}b{}t--Ctt_error_list.pkl\".format(dirName, mi,ma,ste,bo_set,nr_of_toys), \"wb\") as f:\n",
" pkl.dump(Ctt_error_list, f, pkl.HIGHEST_PROTOCOL)\n",
" \n",
" with open(\"{}/{}-{}_{}s{}b{}t--pull_dic.pkl\".format(dirName, mi,ma,ste,bo_set,nr_of_toys), \"wb\") as f:\n",
" pkl.dump(pull_dic, f, pkl.HIGHEST_PROTOCOL)\n",
" \n",
" variab = {'mi': mi,\n",
" 'ma': ma,\n",
" 'ste': ste,\n",
" 'bo_set': bo_set,\n",
" 'nr_of_toys': nr_of_toys}\n",
" \n",
" with open(\"{}/variab.pkl\".format(dirName), \"wb\") as f:\n",
" pkl.dump(variab, f, pkl.HIGHEST_PROTOCOL)\n",
" \n",
" CLs_values = []\n",
" \n",
" toy_size = bo_set\n",
"\n",
" print(np.shape(Nll_list))\n",
" print(Nll_list[0:1])\n",
" \n",
" for step in range(ste):\n",
" CLs_values.append([])\n",
" for toy in range(nr_of_toys):\n",
" float_min = np.min(Nll_list[2*step][toy*bo_set:(toy+1)*bo_set])\n",
" fix_min = np.min(Nll_list[2*step+1][toy*bo_set:(toy+1)*bo_set])\n",
" CLs_values[step].append(float_min-fix_min)\n",
" \n",
" \n",
" print(np.shape(CLs_values))\n",
" \n",
" with open(\"{}/{}-{}_{}s--CLs_list.pkl\".format(dirName, mi,ma,ste), \"wb\") as f:\n",
" pkl.dump(CLs_values, f, pkl.HIGHEST_PROTOCOL)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# print(variab['mi'] != mi)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Plot"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# l = []\n",
"\n",
"# CLs_values = CLs_list\n",
"\n",
"# if not os.path.exists('data/CLs/plots'):\n",
"# os.mkdir('data/CLs/plots')\n",
"# print(\"Directory \" , 'data/CLs/plots' , \" Created \")\n",
"\n",
"# print(np.shape(CLs_values))\n",
" \n",
"# for step in range(1,ste):\n",
"# plt.clf()\n",
"# plt.title('Ctt value: {:.2f}'.format(Ctt_steps[step]))\n",
"# plt.hist(CLs_values[0], bins = 100, range = (-25, 25), label = 'Ctt fixed to 0')\n",
"# plt.hist(CLs_values[step], bins = 100, range = (-25, 25), label = 'Ctt floating')\n",
"# plt.axvline(x=np.mean(CLs_values[0]),color='red', linewidth=1.0, linestyle = 'dotted')\n",
"# plt.legend()\n",
"# plt.savefig('data/CLs/plots/CLs-BR({:.1E}).png'.format(BR_steps[step]))\n",
" \n",
"# l.append(len(np.where(np.array(CLs_values[step]) < np.mean(CLs_values[0]))[0]))\n",
" \n",
"# for step in range(2*ste):\n",
"# if step%2 == 0:\n",
"# floaty = True\n",
"# else:\n",
"# floaty = False\n",
"# for key in pull_dic.keys():\n",
"# if not os.path.exists('data/CLs/plots/{}'.format(key)):\n",
"# os.mkdir('data/CLs/plots/{}'.format(key))\n",
"# plt.clf()\n",
"# plt.title('Pull {} - Ctt value {:.2f} - floating {}'.format(key, Ctt_steps[int(step/2)], floaty))\n",
"# plt.hist(pull_dic[key][step], bins = 50, range = (-5,5))\n",
"# plt.xlabel('Pull')\n",
"# plt.savefig('data/CLs/plots/{}/{:.2f}Ctt{}s{}f.png'.format(key, Ctt_steps[int(step/2)], step, floaty))\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"scrolled": false
},
"outputs": [],
"source": [
"for s in range(len(l)):\n",
" print('BR: {:.4f}'.format(BR_steps[s]))\n",
" print(2*l[s]/len(CLs_values[s]))\n",
" print()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"for step in range(2*ste):\n",
" for key in pull_dic.keys():\n",
" print(pull_dic[key][step])"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# for param in total_f_fit.get_dependents():\n",
"# if param.floating:\n",
"# print(params[param]['value'])"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"print(display_time(int(time.time()-start)))"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# variab['mi'] =! mi"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.7.3"
}
},
"nbformat": 4,
"nbformat_minor": 2
}
