def test_invalid_range():
sample = torch.Tensor([[1,2,0,7],[2,1,4,6],[3,5,1,0]])
try:
H,axes = histogramdd_pytorch(sample,bins = [3,4,8], range=[(1,2),None,(5,4)])
ok = False
except ValueError:
ok = True
assert ok, "Accepted invalid input: invalid range"
def test_invalid_bins():
sample = torch.Tensor([[1,2,0,7],[2,1,4,6],[3,5,1,0]])
try:
H,axes = histogramdd_pytorch(sample,bins = [3,4,0])
ok = False
except ValueError:
ok = True
assert ok, "Accepted invalid input: bins must be positive"
def test_invalid_bins_custombinning_empty_axis():
try:
from torch import searchsorted
except ImportError:
pytest.skip("Current Pytorch version doesn't supoport searchsorted")
sample = torch.Tensor([[1,2,0,7],[2,1,4,6],[3,5,1,0]])
try:
H,axes = histogramdd_pytorch(sample,bins = [torch.tensor([1,2,3]),torch.tensor([2,3,4]),torch.tensor([1])])
ok = False
except ValueError:
ok = True
assert ok, "Accepted invalid input: each axis must have at least 2 edges for custom binning"
def test_histogram_uniform(seed,N,D,bins,device):
if device == "cuda:0" and not torch.cuda.is_available():
pytest.skip("CUDA is not available with pytorch")
torch.random.manual_seed(seed)
sample = torch.rand(D,N,device=device)
H,axes = histogramdd_pytorch(sample,bins)
Hnp,npaxes = np.histogramdd(sample.T.cpu().numpy(),bins)
assert len(axes) == D, "Number of axes invalid"
assert torch.sum(H) <= N, "Sum of histogram too high"
assert np.max(np.abs(H.cpu().numpy() - Hnp)) <= 1, "Histogram inconsistent with numpy histogram"
for i in range(len(axes)):
assert len(axes[i]) == bins[i]+1, "Invalid number of edges"
def makeHistogram(data, hisString, **kwargs):
r"""
Make N dimensional numpy histograms for given histogram string
:param data: input panda
:param hisString: histogram setup string
:param kwargs options
:return:
dictionary with histogram information
* H - ndarrray
* axis - axis description
* varNames - variable Names
* name - histogram name
Example usage:
>>> makeHistogram(MCdata,"TRD:pmeas:particle:#TRD>0>>hisTRDPP(50,0.5,3,20,0.3,5,5,0,5)",3)
"""
options = {}
options.update(histoNDOptions)
options.update(kwargs)
verbose = options['verbose']
if verbose & 0x1:
logging.info("makeHistogram :%s", hisString)
varList = hisString.split(":")
description = varList[-1].replace('#', '')
description = description.split(">>")
selection = description[0]
histoInfo = description[1]
del (varList[-1])
histoInfo = histoInfo.replace('(', ',').replace(')', ',').split(",")
bins = []
hRange = []
for idx, a in enumerate(varList):
bins.append(int(histoInfo[idx * 3 + 1]))
hRange.append(
(float(histoInfo[idx * 3 + 2]), float(histoInfo[idx * 3 + 3])))
if verbose & 0x2:
logging.info("Histogram Name :%s", histoInfo[0])
logging.info("Variable list :%s", varList)
logging.info("Histogram bins :%s", bins)
logging.info("Histogram range :%s", hRange)
if options['use_pytorch']:
H, axis = histogramdd_pytorch(torch.from_numpy(
data.query(selection)[varList].values).T,
bins=bins,
range=hRange)
H = H.numpy()
axis = [i.numpy() for i in axis]
else:
H, axis = np.histogramdd(data.query(selection)[varList].values,
bins=bins,
range=hRange)
return {"H": H, "axes": axis, "name": histoInfo[0], "varNames": varList}
def test_histogram_uniform_randombins(seed,N,D,bins,device):
if device == "cuda:0" and not torch.cuda.is_available():
pytest.skip("CUDA is not available with pytorch")
try:
from torch import searchsorted
except ImportError:
pytest.skip("Current Pytorch version doesn't supoport searchsorted")
torch.random.manual_seed(seed)
sample = torch.rand(D,N,device=device)
nbins = bins
bins = [torch.sort(torch.rand(i+1,device=device))[0] for i in nbins] #+1 because there's 1 more fence post than the length of the fence
H,axes = histogramdd_pytorch(sample,bins)
Hnp,npaxes = np.histogramdd(sample.T.cpu().numpy(),[i.cpu().numpy() for i in bins])
assert len(axes) == D, "Number of axes invalid"
assert torch.sum(H) <= N, "Sum of histogram too high"
assert np.max(np.abs(H.cpu().numpy() - Hnp)) <= 1, "Histogram inconsistent with numpy histogram"
for i in range(len(axes)):
assert len(axes[i]) == nbins[i]+1, "Invalid number of edges"
def test_histogram_singular():
H,axes = histogramdd_pytorch(torch.zeros(5,100000),9)
assert H[4,4,4,4,4] == 100000, "Test for all zeros failed"
assert len(axes) == 5, "Number of axes invalid"
def test_histogram_empty():
sample = torch.empty(5,0)
H,axes = histogramdd_pytorch(sample,bins = [2,3,4,5,6], range=[(1,2),None,None,None,None])
assert torch.sum(H) == 0, "Sum of empty histogram is not zero"
assert len(axes) == 5, "Number of axes invalid"