def main():
# Get DataFrame of All Experiments
df = getAllExperiments()
# Get Only Runs from Experiment 15
df_filtered = df[df['Experiment'] == 15]
# Get Specific Run - No Rabbit Workload and No Stencil and 8 ppn on 2 nodes
df_filtered = df_filtered[df_filtered['cores'] == 8]
df_filtered = df_filtered[df_filtered['processors'] == 16] # 2 Nodes
df_filtered = df_filtered[df_filtered['rabbit_workload'] == 0]
df_filtered = df_filtered[df_filtered['stencil_size'] == 0]
print(df_filtered)
# Get Data from Specific Run
data = getData(df_filtered)
print(data.head())
# Get Data Only from Rank 0
data_rank0 = data[data['rank'] == 0]
# Find Shape, Location, and Scale of Max Data
shape, loc, scale = gevfit.fit(data_rank0['workload_max_usec'])
print("Shape: ", shape, "\tLocation: ", loc, "\tScale: ", scale)
# Get Overall Runtime
runtime0 = data_rank0['interval_max_usec'].sum()
dist = stats.genextreme(shape, loc, scale)
print("Runtime: ", runtime0, "Microseconds at Initial ", data['comm_size'].iloc[0], " Ranks")
# Projected Runtime at k = 8 -> 128 ranks (16 nodes)
projected = emma(dist, 8) * data['iterations'].iloc[0]
print("K = 8: Projected Runtime: ", projected, "\tProjected Efficiency: ", runtime0 / projected)
# Get Projected Scale of When Efficiency Reaches 50%
eh = ehalf(runtime0, data['comm_size'].iloc[0], dist, data['iterations'].iloc[0])
print("Expect 50% Efficiency at: ", eh, " Ranks")
def gev_project2(maxes, k):
project_samples = [ max(resample(maxes, n_samples=k, replace=False)) for _ in range(k*len(maxes))]
return gevfit.fit(project_samples)
def gev_project(params, k, samples=1000):
shape, loc, scale = params
project_samples = [ max(genextreme.rvs(shape, loc, scale, k)) for x in range(k*samples)]
return gevfit.fit(project_samples)
#s = np.random.normal(loc=5.0, scale=1.0, size=(_PROJ_NNODES,_NRANKS,_NITER))
# add per node disturbances
# compute stats for all thr workload
sall=np.reshape(s, (-1,_NITER))
# get maxima
all_max=np.amax(sall, axis=0)
all_workload=np.sum(all_max)
print("all workload {}".format(all_workload))
all_reshape=np.reshape(all_max,(-1,1))
print(all_reshape.shape)
all_gev = gevfit.fit(all_reshape)
print("all gev {}".format(all_gev))
all_pwm = pwm_fit(all_reshape)
print("all pwm {}".format(all_pwm))
all_mom = mom_fit(all_reshape)
print("all mom {}".format(all_mom))
s1=s[0:_NNODES,0:_NRANKS,0:_NITER]
sone=np.reshape(s1, (-1,_NITER))
one_workload=np.sum(np.amax(sone, axis=0))
print("one workload no B {}".format(one_workload))
sone=np.reshape(s1, (-1,1))
def gev_resample_project(samples, nsamples, k):
shape, loc, scale = gevfit.fit(samples)
return [max(genextreme.rvs(shape, loc, scale, k)) for x in range(nsamples)]