Using Python for Gate Analysis¶
Python in a high level (in terms of abstraction) versatile programming language well adapted for scientific use
Both Imperative and Oriented Object approach (at the same time)
Interpeted and dynamic typing (implicite type) and easy object manipulation and indexing
At the price of speed and memory managment (compared to C)
Various scientific libaries for data analysis and ploting (numpy, scipy, matplotlib)
Can read many file format like text and root (with dedicated library)
Free and open source
Written in C and able to read any C Class
Reading Root files with Python¶
3 Approaches :
The Root approach PyROOT: directly using Root in Python
The inbetween approach : rootpy : Redefine some ROOT classes in a more Pythonic way. Seems not be compatible with Python 3. Abbandoned ?
The Python approach root_numpy: Transfoming Root tree to numpy structured array
The Root Flavor¶
ROOT comes with its own Python implementation : PyROOT
Pro :
- No additional libraries needed to acces ROOT files
- Possibility to use both ROOT and Python libraries for data manipulation and plotting
Cons :
- Still using ROOT ;)
import ROOT as rt
tfil = rt.TFile('Simu_TReCam_contact_puit1.root')
tree = tfil.Get('Singles')
h1 = rt.TH1D('energy', 'energy', 200, 0, 0.15)
for i in range(tree.GetEntries()):
tree.GetEntry(i)
h1.Fill(tree.energy)
c1 = rt.TCanvas( 'c1', '', 200, 10, 700, 500 )
h1.Draw()
c1.Draw()
tfil.Close()
- Works fine but it's just using ROOT with Python syntax, so what's the point ?!
- The data can be converted to an python numpy ndarray to benefit of numpy
%matplotlib inline
import numpy as np # Data manipulaion
import matplotlib.pyplot as plt # Graphic library
import ROOT as rt
tfil = rt.TFile('Simu_TReCam_contact_puit1.root')
tree = tfil.Get('Singles')
ene = np.zeros(tree.GetEntries()) # Initialize a 1 dim array with 0
for i in range(tree.GetEntries()):
tree.GetEntry(i)
ene[i] = tree.energy
plt.hist(ene, bins=200, range=(0, 0.15), histtype='step') # Both fill and plot the histogram
plt.show()
- This way benefits from both ROOT and numpy/matplotlib advantages, but still use ROOT
The Numpy Flavor¶
- The root_numpy library directly read root files and convert ROOT Trees to numpy structured array (more advanced ndarray)
- Pro:
- Very easy to use for end users and benefit from all numpy functiunalities
- No ROOT
- Cons:
- Need third party library -> compatibility with ROOT version
- Not memory friendly: the whole tree is loaded at once by default (option to limit this)
- Pro:
import numpy as np
import matplotlib.pyplot as plt
import root_numpy as rnp
singles = rnp.root2array('Simu_TReCam_contact_puit1.root', 'Singles')
# One function to load a whole ROOT Tree in an array
print(singles)
print(singles.dtype)
- Acces to "leaves" is done using their names
ene1 = singles['energy']
posx = singles['globalPosX']
print(ene)
print(posx)
plt.hist(ene1, bins=100, range=(0, 0.15), histtype='step')
plt.show()
- Numpy array allow easy math operations
noise = np.random.normal(0, 0.003, len(ene1))
ene2 = ene1 + noise # Add Gaussian random fluctuation
- And easy indexing
ene3 = ene2[(ene2>0.11) & (ene2<0.13)] # Energy Cut
plt.hist(ene2, bins=100, range=(0, 0.15), histtype='step')
plt.hist(ene3, bins=100, range=(0, 0.15), histtype='step')
plt.show()
Numpy also possess its own format to save arrays either in text or binary format
Adding direct save of Gate simulation in numpy format would solve the issue of root_numpy compatibility with ROOT and get completely rid of ROOT ;)
I'm going to have a look at it to add it to Gate and also add a couple of exemple for data anlaysis
- Python provide powerful tool for data analysis that can easily be addapted and used for Gate.
- Easier to use and to learn for end users or lazy physicists (like me)
- Lack the power and speed of pure C and ROOT libraries for high amount of Data