def nanmean(a, axis=None, keepdims=None, *args, **kwargs):
n = sp.nansum(a, axis=axis, keepdims=keepdims)
d = sp.nansum(sp.nan_to_num(a) != 0, axis=axis,
keepdims=keepdims).astype(n.dtype)
n = sp(data=n.data, coords=n.coords, fill_value=np.nan, shape=n.shape)
d = sp(data=d.data, coords=d.coords, fill_value=np.nan, shape=d.shape)
out = n / d
return sp(data=out.data, coords=out.coords, fill_value=0, shape=out.shape)
def array(a, *args, **kwargs):
if kwargs.get("fill_value", None) is not None:
fill_value = kwargs.pop("fill_value")
else:
fill_value = sp.nan
if type(a) == sp:
return sp(a, *args, **kwargs, fill_value=fill_value)
else:
return sp(np.array(a, *args, **kwargs), fill_value=fill_value)
def nanmean(a, axis=None, keepdims=None, *args, **kwargs):
n = sp.nansum(a, axis=axis, keepdims=keepdims)
d = sp.nansum(sp.nan_to_num(a)!=0, axis=axis, keepdims=keepdims).astype(n.dtype)
n.fill_value=np.nan
d.fill_value=np.nan
n = sp(n)
d = sp(d)
out = n / d
out.fill_value = 0
return sp(out)
def nan_to_num(a):
if type(a) in [int, float, np.int64, np.float64]:
return np.nan_to_num(a)
if hasattr(a, "fill_value"):
a = a.copy()
a.data[np.isnan(a.data)] = 0.0
return sp(coords=a.coords, data=a.data, fill_value=0.0, shape=a.shape)
def cumfunc(a, axis, func):
a = copy.deepcopy(a)
ax = np.arange(a.ndim)
axis = ax[axis]
x = pd.DataFrame(a.coords.T)
x.columns = ['0', '1', '2', '3']
cols = [item for item in x.columns if item != str(axis)]
x['y']=a.data
x = pd.pivot_table(
x , columns=cols,
index=str(axis), values='y')
missing = pd.Int64Index(
set(np.arange(a.shape[axis])) - set(x.index), name=str(axis))
if len(missing) > 0:
x = x.append(pd.DataFrame(
np.repeat((x.iloc[0:1]*0).values, len(missing), axis=0),
index=missing, columns=x.columns))
x = x.unstack().reset_index().fillna(0)
x.columns = [item for item in x.columns[:-1]] + ['y']
x = x.set_index(list(x.columns[:-1])).groupby(level=[0,1,2])
if func == 'cumsum':
x = x.cumsum().reset_index()
if func == 'cumprod':
x = x.cumprod().reset_index()
x = x[x['y']!=0]
a.coords = x[['0', '1', '2', '3']].values.T
a.data = x['y'].values
return sp(a)
def nanmedian(a, axis=None, keepdims=None, *args, **kwargs):
new_a = np.nanmedian(a.todense(),
axis=axis,
keepdims=keepdims,
*args,
**kwargs)
return sp(np.nan_to_num(new_a), fill_value=np.nan)
def nanquantile(a, q, axis=None, keepdims=None, *args, **kwargs):
new_a = np.nanquantile(a.todense(),
q=q,
axis=axis,
keepdims=keepdims,
*args,
**kwargs)
return sp(np.nan_to_num(new_a), fill_value=np.nan)
def nan_to_num(a):
if type(a) in [int, float, np.int64, np.float64]:
return np.nan_to_num(a)
if hasattr(a, "fill_value"):
a = a.copy()
a.data[np.isnan(a.data)] = 0.0
if a.fill_value != 0.0:
a.fill_value = 0.0
return sp(a)
def swapaxes(a, axis1, axis2):
ax = np.arange(a.ndim)
axis1 = ax[axis1]
axis2 = ax[axis2]
fv = a.fill_value
l = []
for item in range(a.ndim):
if item == axis1:
l.append(axis2)
elif item == axis2:
l.append(axis1)
else:
l.append(item)
coords = a.coords[l,:]
return sp(coords, a.data, shape=tuple([a.shape[item] for item in l]),
prune=True, fill_value=fv)
def cumsum(a, axis):
a = copy.deepcopy(a)
x = pd.DataFrame(a.coords.T)
x.columns = ['0', '1', '2', '3']
cols = [item for item in x.columns if item != str(axis)]
x['y'] = a.data
x = pd.pivot_table(x, columns=cols, index=str(axis), values='y')
missing = list(set(np.arange(a.shape[axis])) - set(x.index))
if len(missing) > 0:
x = x.append(
pd.DataFrame(np.repeat((x.iloc[0:1] * 0).values,
len(missing),
axis=0),
index=missing,
columns=x.columns))
x = x.unstack().reset_index().fillna(0)
x.columns = [0, 1, 2, 3, 'y']
x = x.set_index([0, 1, 2, 3]).groupby(level=2).cumsum().reset_index()
x = x[x['y'] > 0]
a.coords = x[[0, 1, 2, 3]].values.T
a.data = x['y'].values
return sp(a)
def ones(*args, **kwargs):
return sp(np.ones(*args, **kwargs), fill_value=sp.nan)
def apply_along_axis(func1d, axis, arr, *args, **kwargs):
fv = arr.fill_value
arr = np.apply_along_axis(func1d, axis, arr.todense(), *args, **kwargs)
return sp(arr, fill_value=fv)
def around(a, *args, **kwargs):
fv = a.fill_value
a = np.around(a.todense(), *args, **kwargs)
return sp(a, fill_value=fv)
def unique(a, *args, **kwargs):
fv = a.fill_value
a = np.unique(a.todense(), *args, **kwargs)
return sp(a, fill_value=fv)
def array(*args, **kwargs):
return sp(np.array(*args, **kwargs))
def ones(*args, **kwargs):
return sp(np.ones(*args, **kwargs))
def nansum(a, axis=None, keepdims=None, *args, **kwargs):
return sp(data=a.data, coords=a.coords, fill_value=0.0,
shape=a.shape).sum(axis=axis, keepdims=keepdims, *args, **kwargs)
