def set_backend(self, backend, inplace=False, deep=False):
""" Converts triangle array_backend.
Parameters
----------
backend : str
Currently supported options are 'numpy', 'sparse', and 'cupy'
inplace : bool
Whether to mutate the existing Triangle instance or return a new
one.
Returns
-------
Triangle with updated array_backend
"""
if hasattr(self, "array_backend"):
old_backend = self.array_backend
else:
if hasattr(self, "ldf_"):
old_backend = self.ldf_.array_backend
else:
raise ValueError("Unable to determine array backend.")
if inplace:
if backend in ["numpy", "sparse", "cupy"]:
lookup = {
"numpy": {
"sparse": lambda x: x.todense(),
"cupy": lambda x: cp.asnumpy(x),
},
"cupy": {
"numpy": lambda x: cp.array(x),
"sparse": lambda x: cp.array(x.todense()),
},
"sparse": {
"numpy": lambda x: sp.array(x),
"cupy": lambda x: sp.array(cp.asnumpy(x)),
},
}
if hasattr(self, "values"):
self.values = lookup[backend].get(old_backend, lambda x: x)(
self.values
)
if deep:
for k, v in vars(self).items():
if isinstance(v, Common):
v.set_backend(backend, inplace=True, deep=True)
if hasattr(self, "array_backend"):
self.array_backend = backend
else:
raise AttributeError(backend, "backend is not supported.")
return self
else:
obj = self.copy()
return obj.set_backend(backend=backend, inplace=True, deep=deep)
def set_backend(self, backend, inplace=False):
''' Converts triangle array_backend.
Parameters
----------
backend : str
Currently supported options are 'numpy', 'sparse', and 'cupy'
inplace : bool
Whether to mutate the existing Triangle instance or return a new
one.
Returns
-------
Triangle with updated array_backend
'''
if hasattr(self, 'array_backend'):
old_backend = self.array_backend
else:
if hasattr(self, 'ldf_'):
old_backend = self.ldf_.array_backend
else:
raise ValueError('Unable to determine array backend.')
if inplace:
if backend in ['numpy', 'sparse', 'cupy']:
lookup = {
'numpy': {
'sparse': lambda x: x.todense(),
'cupy': lambda x: cp.asnumpy(x)
},
'cupy': {
'numpy': lambda x: cp.array(x),
'sparse': lambda x: cp.array(x.todense())
},
'sparse': {
'numpy': lambda x: sp.array(x),
'cupy': lambda x: sp.array(cp.asnumpy(x))
}
}
if hasattr(self, 'values'):
self.values = lookup[backend].get(old_backend,
lambda x: x)(self.values)
for k, v in vars(self).items():
if isinstance(v, Common):
v.set_backend(backend, inplace=True)
if hasattr(self, 'array_backend'):
self.array_backend = backend
else:
raise AttributeError(backend, 'backend is not supported.')
return self
else:
obj = copy.deepcopy(self)
return obj.set_backend(backend=backend, inplace=True)
def _slice_valuation(self, key):
''' private method for handling of valuation slicing '''
obj = copy.deepcopy(self)
obj.valuation_date = min(obj.valuation[key].max(), obj.valuation_date)
key = key.reshape(self.shape[-2:], order='f')
nan_tri = np.ones(self.shape[-2:])
nan_tri = key * nan_tri
nan_tri[nan_tri == 0] = np.nan
o, d = nan_tri.shape
o_idx = np.arange(o)[list(np.sum(np.isnan(nan_tri), 1) != d)]
d_idx = np.arange(d)[list(np.sum(np.isnan(nan_tri), 0) != o)]
obj.odims = obj.odims[np.sum(np.isnan(nan_tri), 1) != d]
if len(obj.ddims) > 1:
obj.ddims = obj.ddims[np.sum(np.isnan(nan_tri), 0) != o]
xp = cp.get_array_module(obj.values)
if xp == cp:
nan_tri = cp.array(nan_tri)
if xp == sp:
nan_tri = sp(nan_tri)
obj.values = (obj.values * nan_tri)
if np.all(o_idx == np.array(range(o_idx[0], o_idx[-1] + 1))):
o_idx = slice(o_idx[0], o_idx[-1] + 1)
if np.all(d_idx == np.array(range(d_idx[0], d_idx[-1] + 1))):
d_idx = slice(d_idx[0], d_idx[-1] + 1)
if type(o_idx) is slice or type(d_idx) is slice:
# If contiguous slices, this is faster
obj.values = obj.values[..., o_idx, d_idx]
else:
obj.values = xp.take(xp.take(obj.values, o_idx, -2), d_idx, -1)
return obj
def _slice_development(self, key):
''' private method for handling of development slicing '''
obj = copy.deepcopy(self)
obj.ddims = obj.ddims[key]
if cp.get_array_module(obj.values) == cp:
key = cp.array(key)
obj.values = obj.values[..., key]
return self._cleanup_slice(obj)
def read_json(json_str, array_backend=None):
def sparse_in(json_str, dtype, shape):
k, v, o, d = shape
x = json.loads(json_str)
y = np.array([tuple([int(idx) for idx in item[1:-1].split(',')])
for item in x.keys()])
new = coo_matrix(
(np.array(list(x.values())), (y[:, 0], y[:, 1])),
shape=(k*v*o, d), dtype=dtype).toarray().reshape(k,v,o,d)
new[new==0] = np.nan
return new
if array_backend is None:
from chainladder import ARRAY_BACKEND
array_backend = ARRAY_BACKEND
json_dict = json.loads(json_str)
if type(json_dict) is list:
import chainladder as cl
return Pipeline(steps=[
(item['name'],
cl.__dict__[item['__class__']]().set_params(**item['params']))
for item in json_dict])
elif 'kdims' in json_dict.keys():
tri = Triangle(array_backend=array_backend)
arrays = ['kdims', 'vdims', 'odims', 'ddims']
for array in arrays:
setattr(tri, array, np.array(
json_dict[array]['array'], dtype=json_dict[array]['dtype']))
shape = (len(tri.kdims), len(tri.vdims), len(tri.odims), len(tri.ddims))
properties = ['key_labels', 'origin_grain', 'development_grain',
'nan_override', 'is_cumulative']
for prop in properties:
setattr(tri, prop, json_dict[prop])
if json_dict.get('is_val_tri', False):
tri.ddims = pd.PeriodIndex(tri.ddims, freq=tri.development_grain).to_timestamp(how='e')
tri.valuation_date = pd.to_datetime(
json_dict['valuation_date'], format='%Y-%m-%d').to_period('M').to_timestamp(how='e')
tri._set_slicers()
tri.valuation = tri._valuation_triangle()
if json_dict['values'].get('sparse', None):
tri.values = sparse_in(json_dict['values']['array'],
json_dict['values']['dtype'], shape)
if tri.is_cumulative:
tri.is_cumulative = False
tri = tri.incr_to_cum()
else:
tri.values = np.array(json_dict['values']['array'],
dtype=json_dict['values']['dtype'])
if array_backend == 'cupy':
tri.values = cp.array(tri.values)
return tri
else:
import chainladder as cl
return cl.__dict__[
json_dict['__class__']]().set_params(**json_dict['params'])
def _slice_valuation(self, key):
''' private method for handling of valuation slicing '''
obj = copy.deepcopy(self)
obj.valuation_date = min(obj.valuation[key].max(), obj.valuation_date)
key = key.reshape(self.shape[-2:], order='f')
nan_tri = np.ones(self.shape[-2:])
nan_tri = key * nan_tri
nan_tri[nan_tri == 0] = np.nan
o, d = nan_tri.shape
o_idx = np.arange(o)[list(np.sum(np.isnan(nan_tri), 1) != d)]
d_idx = np.arange(d)[list(np.sum(np.isnan(nan_tri), 0) != o)]
obj.odims = obj.odims[np.sum(np.isnan(nan_tri), 1) != d]
if len(obj.ddims) > 1:
obj.ddims = obj.ddims[np.sum(np.isnan(nan_tri), 0) != o]
xp = cp.get_array_module(obj.values)
if xp == cp:
nan_tri = cp.array(nan_tri)
obj.values = (obj.values * nan_tri)
obj.values = xp.take(xp.take(obj.values, o_idx, -2), d_idx, -1)
return self._cleanup_slice(obj)
def _validate_arithmetic(self, other):
''' Common functionality BEFORE arithmetic operations '''
obj = copy.deepcopy(self)
xp = cp.get_array_module(obj.values)
other = other if type(other) in [int, float] else copy.deepcopy(other)
ddims = None
odims = None
if type(other) not in [int, float, np.float64, np.int64]:
if len(self.vdims) != len(other.vdims):
raise ValueError('Triangles must have the same number of ' +
'columns')
if len(self.kdims) != len(other.kdims):
raise ValueError('Triangles must have the same number of ' +
'index')
if len(self.vdims) == 1:
other.vdims = np.array([None])
# If broadcasting doesn't work, then try union of origin/developments
# before failure
a, b = self.shape[-2:], other.shape[-2:]
if not (a[0] == 1 or b[0] == 1 or a[0] == b[0]) or \
not (a[1] == 1 or b[1] == 1 or a[1] == b[1]):
ddims = pd.concat((pd.Series(self.ddims, index=self.ddims),
pd.Series(other.ddims, index=other.ddims)), axis=1)
odims = pd.concat((pd.Series(self.odims, index=self.odims),
pd.Series(other.odims, index=other.odims)), axis=1)
other_arr = xp.zeros(
(other.shape[0], other.shape[1], len(odims), len(ddims)))
other_arr[:] = xp.nan
o_arr1 = odims[1].isna().values
o_arr0 = odims[0].isna().values
d_arr1 = ddims[1].isna().values
d_arr0 = ddims[0].isna().values
if xp == cp:
o_arr1 = cp.array(o_arr1)
o_arr0 = cp.array(o_arr0)
d_arr1 = cp.array(d_arr1)
d_arr0 = cp.array(d_arr0)
ol = int(xp.where(~o_arr1 == 1)[0].min())
oh = int(xp.where(~o_arr1 == 1)[0].max()+1)
if np.any(self.ddims != other.ddims):
dl = int(xp.where(~d_arr1 == 1)[0].min())
dh = int(xp.where(~d_arr1 == 1)[0].max()+1)
other_arr[:, :, ol:oh, dl:dh] = other.values
else:
other_arr[:, :, ol:oh, :] = other.values
obj_arr = xp.zeros(
(self.shape[0], self.shape[1], len(odims), len(ddims)))
obj_arr[:] = xp.nan
ol = int(xp.where(~o_arr0 == 1)[0].min())
oh = int(xp.where(~o_arr0 == 1)[0].max()+1)
if np.any(self.ddims != other.ddims):
dl = int(xp.where(~d_arr0 == 1)[0].min())
dh = int(xp.where(~d_arr0 == 1)[0].max()+1)
obj_arr[:, :, ol:oh, dl:dh] = self.values
else:
obj_arr[:, :, ol:oh, :] = self.values
odims = np.array(odims.index)
ddims = np.array(ddims.index)
obj.ddims = ddims
obj.odims = odims
obj.values = obj_arr
other.values = other_arr
obj._set_slicers()
obj.valuation = obj._valuation_triangle()
if hasattr(obj, '_nan_triangle_'):
# Force update on _nan_triangle at next access.
del obj._nan_triangle_
other = other.values
return obj, other
def read_json(json_str, array_backend=None):
def sparse_in(json_str, dtype, shape):
k, v, o, d = shape
x = json.loads(json_str)
y = np.array([
tuple([int(idx) for idx in item[1:-1].split(",")])
for item in x.keys()
])
new = (coo_matrix(
(np.array(list(x.values())), (y[:, 0], y[:, 1])),
shape=(k * v * o, d),
dtype=dtype,
).toarray().reshape(k, v, o, d))
new[new == 0] = np.nan
return new
if array_backend is None:
from chainladder import ARRAY_BACKEND
array_backend = ARRAY_BACKEND
json_dict = json.loads(json_str)
if type(json_dict) is list:
import chainladder as cl
return Pipeline(steps=[(
item["name"],
cl.__dict__[item["__class__"]]().set_params(**item["params"]),
) for item in json_dict])
elif "kdims" in json_dict.keys():
tri = Triangle()
tri.array_backend = array_backend
arrays = ["kdims", "vdims", "odims", "ddims"]
for array in arrays:
setattr(
tri,
array,
np.array(json_dict[array]["array"],
dtype=json_dict[array]["dtype"]),
)
shape = (len(tri.kdims), len(tri.vdims), len(tri.odims),
len(tri.ddims))
properties = [
"key_labels",
"origin_grain",
"development_grain",
"is_cumulative",
]
for prop in properties:
setattr(tri, prop, json_dict[prop])
if json_dict.get("is_val_tri", False):
tri.ddims = pd.PeriodIndex(
tri.ddims, freq=tri.development_grain).to_timestamp(how="e")
tri.valuation_date = (pd.to_datetime(
json_dict["valuation_date"],
format="%Y-%m-%d").to_period("M").to_timestamp(how="e"))
if json_dict["values"].get("sparse", None):
tri.values = sparse_in(json_dict["values"]["array"],
json_dict["values"]["dtype"], shape)
else:
tri.values = np.array(json_dict["values"]["array"],
dtype=json_dict["values"]["dtype"])
if array_backend == "cupy":
tri.values = cp.array(tri.values)
if tri.is_cumulative:
tri.is_cumulative = False
tri = tri.incr_to_cum()
if "sub_tris" in json_dict.keys():
for k, v in json_dict["sub_tris"].items():
setattr(tri, k, read_json(v, array_backend))
if "dfs" in json_dict.keys():
for k, v in json_dict["dfs"].items():
df = pd.read_json(v)
if len(df.columns) == 1:
df = df.iloc[:, 0]
setattr(tri, k, df)
tri._set_slicers()
return tri
else:
import chainladder as cl
return cl.__dict__[json_dict["__class__"]]().set_params(
**json_dict["params"])
