e7cabf65ac
- Much cleaner and more concise source - Better exceptions handling - Checks local minima to see if retuning is needed. Resolved conflicts: bench/blas.cpp include/isaac/backend/templates/mproduct.h include/isaac/driver/buffer.h lib/array.cpp lib/backend/templates/mproduct.cpp lib/driver/buffer.cpp python/setup.py tune/pysrc/autotune.py tune/pysrc/dataset.py tune/pysrc/misc_tools.py
49 lines
1.5 KiB
Python
49 lines
1.5 KiB
Python
from sklearn import ensemble
|
|
import numpy as np
|
|
|
|
def gmean(a, axis=0, dtype=None):
|
|
if not isinstance(a, np.ndarray): # if not an ndarray object attempt to convert it
|
|
log_a = np.log(np.array(a, dtype=dtype))
|
|
elif dtype: # Must change the default dtype allowing array type
|
|
if isinstance(a,np.ma.MaskedArray):
|
|
log_a = np.log(np.ma.asarray(a, dtype=dtype))
|
|
else:
|
|
log_a = np.log(np.asarray(a, dtype=dtype))
|
|
else:
|
|
log_a = np.log(a)
|
|
return np.exp(log_a.mean(axis=axis))
|
|
|
|
def nrmse(y_ground, y):
|
|
N = y.size
|
|
rmsd = np.sqrt(np.sum((y_ground - y)**2)/N)
|
|
if len(y_ground) > 1:
|
|
return rmsd/(np.max(y_ground) - np.min(y_ground))
|
|
else:
|
|
return rmsd
|
|
|
|
def train(X, Y, profiles):
|
|
X = np.array(X)
|
|
Y = np.array(Y)
|
|
M = X.shape[0]
|
|
|
|
p = np.random.permutation(X.shape[0])
|
|
X = X[p,:]
|
|
Y = Y[p,:]
|
|
|
|
#Train the model
|
|
cut = int(0.9*M)
|
|
XTr, YTr = X[:cut,:], Y[:cut,:]
|
|
XCv, YCv = X[cut:,:], Y[cut:,:]
|
|
|
|
nrmses = {}
|
|
for N in range(1,min(M+1,20)):
|
|
for depth in range(1,min(M+1,20)):
|
|
clf = ensemble.RandomForestRegressor(N, max_depth=depth).fit(XTr, YTr)
|
|
t = np.argmax(clf.predict(XCv), axis = 1)
|
|
y = np.array([YCv[i,t[i]] for i in range(t.size)])
|
|
ground = np.max(YCv[:,:], axis=1)
|
|
nrmses[clf] = nrmse(ground, y)
|
|
|
|
clf = min(nrmses, key=nrmses.get)
|
|
return clf, nrmses[clf]
|