Parallelization using joblib¶

In [1]:

import time
import gmspy as gm
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from joblib import Parallel, delayed

import time import gmspy as gm import matplotlib.pyplot as plt import numpy as np import pandas as pd from joblib import Parallel, delayed

You can check the configuration of this computer in advance

In [2]:

import psutil
print("the number of physical cores:", psutil.cpu_count(logical=False))
print("the number of logical cores:", psutil.cpu_count())
print("CPU frequency:", psutil.cpu_freq())

import psutil print("the number of physical cores:", psutil.cpu_count(logical=False)) print("the number of logical cores:", psutil.cpu_count()) print("CPU frequency:", psutil.cpu_freq())

the number of physical cores: 14
the number of logical cores: 20
CPU frequency: scpufreq(current=2300.0, min=0.0, max=2300.0)

Below is the ground motion data built into gmspy, you can also document the PEER database through loadPEER, or you can read the data from text using numpy.loadtxt.

In [3]:

ts, acc = gm.load_gm_examples("Kobe")
dt = ts[1] - ts[0]
# You can also use other acc data

ts, acc = gm.load_gm_examples("Kobe") dt = ts[1] - ts[0] # You can also use other acc data

Suppose we have 1000 ground motions:

In [4]:

all_accs = [acc] * 1000

all_accs = [acc] * 1000

A parallel function needs to be defined, here is the IMs of the ground motions

In [5]:

def get_ims(acc, dt):
    """Compute intensity measures for the given acceleration data."""
    GM = gm.SeismoGM(dt=dt, acc=acc, unit="g")
    ims = GM.get_ims(display_results=False)
    pga = ims["PGA"]
    pgv = ims["PGV"]
    pgd = ims["PGD"]
    ia = ims["Ia"]  # Arias Intensity 
    # elastic spectral acceleration for T=1.0s, index 0 is the spectral acceleration
    sat1 = GM.get_elas_spec(Ts=1.0, damp_ratio=0.05)[0]
    # average spectral acceleration, index 0 is the spectral acceleration
    Tavg = np.arange(0.05, 4.05, 0.05)
    avgsa = GM.get_avgsavd(Tavg=Tavg)[0]
    return pga, pgv, pgd, ia, sat1, avgsa

def get_ims(acc, dt): """Compute intensity measures for the given acceleration data.""" GM = gm.SeismoGM(dt=dt, acc=acc, unit="g") ims = GM.get_ims(display_results=False) pga = ims["PGA"] pgv = ims["PGV"] pgd = ims["PGD"] ia = ims["Ia"] # Arias Intensity # elastic spectral acceleration for T=1.0s, index 0 is the spectral acceleration sat1 = GM.get_elas_spec(Ts=1.0, damp_ratio=0.05)[0] # average spectral acceleration, index 0 is the spectral acceleration Tavg = np.arange(0.05, 4.05, 0.05) avgsa = GM.get_avgsavd(Tavg=Tavg)[0] return pga, pgv, pgd, ia, sat1, avgsa

In [6]:

start_time = time.time()

start_time = time.time()

For more details, please refer to joblib.Parallel.

n_jobs=-1 means using all possible CPU cores.

In [7]:

output = Parallel(n_jobs=-1)(delayed(get_ims)(acc, dt=dt) for acc in all_accs)

output = Parallel(n_jobs=-1)(delayed(get_ims)(acc, dt=dt) for acc in all_accs)

In [8]:

end_time = time.time()
print(f"Parallel computation took {end_time - start_time:.2f} seconds")

end_time = time.time() print(f"Parallel computation took {end_time - start_time:.2f} seconds")

Parallel computation took 18.98 seconds

Convert to pandas.DataFrame

In [9]:

IMS = pd.DataFrame(output, columns=["PGA", "PGV", "PGD", "IA", "SAT1", "AVGSA"])

IMS.head()

IMS = pd.DataFrame(output, columns=["PGA", "PGV", "PGD", "IA", "SAT1", "AVGSA"]) IMS.head()

Out[9]:

	PGA	PGV	PGD	IA	SAT1	AVGSA
0	0.3447	27.677934	9.693236	1.68744	0.351312	0.155794
1	0.3447	27.677934	9.693236	1.68744	0.351312	0.155794
2	0.3447	27.677934	9.693236	1.68744	0.351312	0.155794
3	0.3447	27.677934	9.693236	1.68744	0.351312	0.155794
4	0.3447	27.677934	9.693236	1.68744	0.351312	0.155794