Skip to content

GitLab

Commit 1097bff3 authored by Marco Selig's avatar Marco Selig
Browse files

major bugfix in nifty_explicit; several minor improvements.

parent d4d18b1c
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 370 additions and 270 deletions
nifty_core.py
View file @ 1097bff3
This diff is collapsed.
nifty_explicit.py
View file @ 1097bff3
......@@ -227,8 +227,8 @@ class explicit_operator(operator):
raise TypeError(about._errors.cstring("ERROR: insufficient input."))
else:
raise ValueError(about._errors.cstring("ERROR: dimension mismatch ( "+str(np.size(matrix,axis=None))+" <> "+str(self.domain.dim(split=False))+" )."))
if(val.size>1048576):
about.infos.cprint("INFO: matrix size > 2 ** 20.")
# if(val.size>1048576):
# about.infos.cprint("INFO: matrix size > 2 ** 20.")
self.target = target
## check datatype
......@@ -417,8 +417,8 @@ class explicit_operator(operator):
if(self.target!=self.domain):
sym = False
uni = False
if(val.size>1048576):
about.infos.cprint("INFO: matrix size > 2 ** 20.")
# if(val.size>1048576):
# about.infos.cprint("INFO: matrix size > 2 ** 20.")
else:
raise ValueError(about._errors.cstring("ERROR: dimension mismatch ( "+str(np.size(matrix,axis=None))+" <> "+str(self.nrow())+" x "+str(self.ncol())+" )."))
......@@ -579,6 +579,42 @@ class explicit_operator(operator):
##+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
def _briefing(self,x,domain,inverse): ## > prepares x for `multiply`
## inspect x
if(not isinstance(x,field)):
x_ = field(domain,val=x,target=None)
else:
## check x.domain
if(domain==x.domain):
x_ = x
## transform
else:
x_ = x.transform(target=domain,overwrite=False)
## weight if ...
if(not self.imp)and(not domain.discrete):
x_ = x_.weight(power=1,overwrite=False)
return x_
def _debriefing(self,x,x_,target,inverse): ## > evaluates x and x_ after `multiply`
if(x_ is None):
return None
else:
## inspect x_
if(not isinstance(x_,field)):
x_ = field(target,val=x_,target=None)
elif(x_.domain!=target):
raise ValueError(about._errors.cstring("ERROR: invalid output domain."))
## inspect x
if(isinstance(x,field)):
## repair ...
if(self.domain==self.target!=x.domain):
x_ = x_.transform(target=x.domain,overwrite=False) ## ... domain
if(x_.domain==x.domain)and(x_.target!=x.target):
x_.set_target(newtarget=x.target) ## ... codomain
return x_
##+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
def inverse_times(self,x,**kwargs):
"""
Applies the inverse operator to a given object.
......@@ -1090,7 +1126,7 @@ class explicit_operator(operator):
The transposed matrix.
"""
return explicit_operator(self.domain,self.val.T,bare=True,sym=self.sym,uni=self.uni,target=self.target)
return explicit_operator(self.target,self.val.T,bare=True,sym=self.sym,uni=self.uni,target=self.domain)
def conjugate(self):
"""
......@@ -1214,7 +1250,7 @@ class explicit_operator(operator):
uni = None
X = X*np.ones(self.domain.dim(split=False),dtype=np.int,order='C')
X = self.domain.calc_weight(X,power=-1).astype(max(min(X.dtype,self.domain.datatype),min(X.dtype,self.target.datatype)))
matrix = self.val+np.diag(X,k=0)
matrix = self.val+np.diag(X.flatten(order='C'),k=0)
elif(np.size(X)==np.size(self.val)):
sym = None
uni = None
......@@ -1254,7 +1290,7 @@ class explicit_operator(operator):
raise ValueError(about._errors.cstring("ERROR: identity ill-defined for "+str(self.nrow())+" x "+str(self.ncol())+" matrices."))
self.uni = None
X = X*np.ones(self.domain.dim(split=False),dtype=np.int,order='C')
X = np.diag(self.domain.calc_weight(X,power=-1).astype(max(min(X.dtype,self.domain.datatype),min(X.dtype,self.target.datatype))),k=0)
X = np.diag(self.domain.calc_weight(X,power=-1).astype(max(min(X.dtype,self.domain.datatype),min(X.dtype,self.target.datatype))).flatten(order='C'),k=0)
elif(np.size(X)==np.size(self.val)):
self.sym = None
self.uni = None
......@@ -1304,7 +1340,7 @@ class explicit_operator(operator):
uni = None
X = X*np.ones(self.domain.dim(split=False),dtype=np.int,order='C')
X = self.domain.calc_weight(X,power=-1).astype(max(min(X.dtype,self.domain.datatype),min(X.dtype,self.target.datatype)))
matrix = self.val-np.diag(X,k=0)
matrix = self.val-np.diag(X.flatten(order='C'),k=0)
elif(np.size(X)==np.size(self.val)):
sym = None
uni = None
......@@ -1344,7 +1380,7 @@ class explicit_operator(operator):
uni = None
X = X*np.ones(self.domain.dim(split=False),dtype=np.int,order='C')
X = self.domain.calc_weight(X,power=-1).astype(max(min(X.dtype,self.domain.datatype),min(X.dtype,self.target.datatype)))
matrix = np.diag(X,k=0)-self.val
matrix = np.diag(X.flatten(order='C'),k=0)-self.val
elif(np.size(X)==np.size(self.val)):
sym = None
uni = None
......@@ -1382,7 +1418,7 @@ class explicit_operator(operator):
raise ValueError(about._errors.cstring("ERROR: identity ill-defined for "+str(self.nrow())+" x "+str(self.ncol())+" matrices."))
self.uni = None
X = X*np.ones(self.domain.dim(split=False),dtype=np.int,order='C')
X = np.diag(self.domain.calc_weight(X,power=-1).astype(max(min(X.dtype,self.domain.datatype),min(X.dtype,self.target.datatype))),k=0)
X = np.diag(self.domain.calc_weight(X,power=-1).astype(max(min(X.dtype,self.domain.datatype),min(X.dtype,self.target.datatype))).flatten(order='C'),k=0)
elif(np.size(X)==np.size(self.val)):
self.sym = None
self.uni = None
......@@ -1439,7 +1475,7 @@ class explicit_operator(operator):
sym = None
uni = None
X = X*np.ones(self.domain.dim(split=False),dtype=np.int,order='C')
X = np.diag(self.domain.calc_weight(X,power=-1).astype(max(min(X.dtype,self.domain.datatype),min(X.dtype,self.target.datatype))),k=0)
X = np.diag(self.domain.calc_weight(X,power=-1).astype(max(min(X.dtype,self.domain.datatype),min(X.dtype,self.target.datatype))).flatten(order='C'),k=0)
elif(np.size(X)==self.val.shape[1]**2):
newdomain = self.domain
sym = None
......@@ -1472,7 +1508,7 @@ class explicit_operator(operator):
sym = None
uni = None
X = X*np.ones(self.domain.dim(split=False),dtype=np.int,order='C')
X = np.diag(self.domain.calc_weight(X,power=-1).astype(max(min(X.dtype,self.domain.datatype),min(X.dtype,self.target.datatype))),k=0)
X = np.diag(self.domain.calc_weight(X,power=-1).astype(max(min(X.dtype,self.domain.datatype),min(X.dtype,self.target.datatype))).flatten(order='C'),k=0)
elif(np.size(X)==self.val.shape[0]**2):
newtarget = self.target
sym = None
......@@ -1505,7 +1541,7 @@ class explicit_operator(operator):
self.sym = None
self.uni = None
X = X*np.ones(self.domain.dim(split=False),dtype=np.int,order='C')
X = np.diag(self.domain.calc_weight(X,power=-1).astype(max(min(X.dtype,self.domain.datatype),min(X.dtype,self.target.datatype))),k=0)
X = np.diag(self.domain.calc_weight(X,power=-1).astype(max(min(X.dtype,self.domain.datatype),min(X.dtype,self.target.datatype))).flatten(order='C'),k=0)
elif(np.size(X)==self.val.shape[1]**2):
self.sym = None
self.uni = None
......@@ -2088,30 +2124,25 @@ class explicit_probing(probing):
return self.probing(zipped[0],probe)
def _serial_probing(self,zipped): ## > performs the probing operation serially
try:
result = self._single_probing(zipped)
except:
## kill pool
os.kill()
else:
if(result is not None):
result = np.array(result).flatten(order='C')
rindex = zipped[0]*self.codomain.dim(split=False)
if(isinstance(_share.mat,tuple)):
_share.mat[0].acquire(block=True,timeout=None)
_share.mat[0][rindex:rindex+self.codomain.dim(split=False)] = np.real(result)
_share.mat[0].release()
_share.mat[1].acquire(block=True,timeout=None)
_share.mat[1][rindex:rindex+self.codomain.dim(split=False)] = np.imag(result)
_share.mat[1].release()
else:
_share.mat.acquire(block=True,timeout=None)
_share.mat[rindex:rindex+self.codomain.dim(split=False)] = result
_share.mat.release()
_share.num.acquire(block=True,timeout=None)
_share.num.value += 1
_share.num.release()
self._progress(_share.num.value)
result = self._single_probing(zipped)
if(result is not None):
result = np.array(result).flatten(order='C')
rindex = zipped[0]*self.codomain.dim(split=False)
if(isinstance(_share.mat,tuple)):
_share.mat[0].acquire(block=True,timeout=None)
_share.mat[0][rindex:rindex+self.codomain.dim(split=False)] = np.real(result)
_share.mat[0].release()
_share.mat[1].acquire(block=True,timeout=None)
_share.mat[1][rindex:rindex+self.codomain.dim(split=False)] = np.imag(result)
_share.mat[1].release()
else:
_share.mat.acquire(block=True,timeout=None)
_share.mat[rindex:rindex+self.codomain.dim(split=False)] = result
_share.mat.release()
_share.num.acquire(block=True,timeout=None)
_share.num.value += 1
_share.num.release()
self._progress(_share.num.value)
def _parallel_probing(self): ## > performs the probing operations in parallel
## define weighted canonical base
......@@ -2134,11 +2165,13 @@ class explicit_probing(probing):
about.infos.cflush(" done.")
pool.close()
pool.join()
except:
except BaseException as exception:
## terminate and join pool
about._errors.cprint("\nERROR: terminating pool.")
pool.terminate()
pool.join()
raise Exception(about._errors.cstring("ERROR: unknown. NOTE: pool terminated.")) ## traceback by looping
## re-raise exception
raise exception ## traceback by looping
## evaluate
if(issubclass(self.codomain.datatype,np.complexfloating)):
_mat = (np.array(_mat[0][:])+np.array(_mat[1][:])*1j) ## comlpex array
......
nifty_power.py
View file @ 1097bff3
......@@ -210,16 +210,16 @@ def smooth_power(spec,domain=None,kindex=None,mode="2s",exclude=1,sigma=-1,**kwa
## check power spectrum
if(isinstance(spec,field)):
spec = spec.val.astype(kindex.dtype)
spec = spec.val
elif(callable(spec)):
try:
spec = np.array(spec(kindex),dtype=kindex.dtype)
spec = np.array(spec(kindex),dtype=None)
except:
TypeError(about._errors.cstring("ERROR: invalid power spectra function.")) ## exception in ``spec(kindex)``
elif(np.isscalar(spec)):
spec = np.array([spec],dtype=kindex.dtype)
spec = np.array([spec],dtype=None)
else:
spec = np.array(spec,dtype=kindex.dtype)
spec = np.array(spec,dtype=None)
## drop imaginary part
spec = np.real(spec)
## check finiteness and positivity (excluding null)
......@@ -663,13 +663,13 @@ def interpolate_power(spec,mode="linear",domain=None,kindex=None,newkindex=None,
spec = spec.val.astype(kindex.dtype)
elif(callable(spec)):
try:
spec = np.array(spec(kindex),dtype=kindex.dtype)
spec = np.array(spec(kindex),dtype=None)
except:
TypeError(about._errors.cstring("ERROR: invalid power spectra function.")) ## exception in ``spec(kindex)``
elif(np.isscalar(spec)):
spec = np.array([spec],dtype=kindex.dtype)
spec = np.array([spec],dtype=None)
else:
spec = np.array(spec,dtype=kindex.dtype)
spec = np.array(spec,dtype=None)
## drop imaginary part
spec = np.real(spec)
## check finiteness and positivity (excluding null)
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment