def sdc_nansum_overload(self):
"""
Intel Scalable Dataframe Compiler Developer Guide
*************************************************
Parallel replacement of numpy.nansum.
.. only:: developer
Test: python -m sdc.runtests sdc.tests.test_sdc_numpy -k nansum
"""
dtype = self.dtype
isnan = get_isnan(dtype)
if not isinstance(self, types.Array):
return None
if isinstance(dtype, types.Number):
def sdc_nansum_number_impl(self):
length = len(self)
result = 0
for i in prange(length):
if not numpy.isnan(self[i]):
result += self[i]
return result
return sdc_nansum_number_impl
if isinstance(dtype, (types.Boolean, bool)):
return gen_sum_bool_impl()
def ov_impl(a):
if not isinstance(a, types.Array):
return
if isinstance(a.dtype, (types.Float, types.Complex)):
isnan = get_isnan(a.dtype)
initial_result = {
min: numpy.inf,
max: -numpy.inf,
}[reduce_op]
def impl(a):
result = initial_result
nan_count = 0
length = len(a)
for i in prange(length):
v = a[i]
if not isnan(v):
result = reduce_op(result, v)
else:
nan_count += 1
if nan_count == length:
return numpy.nan
return result
return impl
else:
def impl(a):
result = a[0]
for i in prange(len(a) - 1):
result = reduce_op(result, a[i + 1])
return result
return impl
def np_nanmean(a):
if not isinstance(a, types.Array):
return
isnan = get_isnan(a.dtype)
def nanmean_impl(a):
c = 0.0
count = 0
for i in prange(len(a)):
v = a[i]
if not isnan(v):
c += v
count += 1
# np.divide() doesn't raise ZeroDivisionError
return np.divide(c, count)
return nanmean_impl
def np_nancumsum(arr, like_pandas=False):
if not isinstance(arr, types.Array):
return
if isinstance(arr.dtype, (types.Boolean, types.Integer)):
# dtype cannot possibly contain NaN
return lambda arr, like_pandas=False: cumsum(arr)
else:
retty = arr.dtype
is_nan = get_isnan(retty)
zero = retty(0)
def nancumsum_impl(arr, like_pandas=False):
chunks = parallel_chunks(len(arr))
partial_sum = numpy.zeros(len(chunks), dtype=retty)
result = numpy.empty_like(arr)
# below line is only needed since Literal[bool] var cannot be converted to bool
# in a prange due to a bug related to absence of BooleanLiterals in Numba
_like_pandas = True if like_pandas else False
for i in prange(len(chunks)):
chunk = chunks[i]
partial = zero
for j in range(chunk.start, chunk.stop):
if _like_pandas:
result[j] = partial + arr[j]
if ~is_nan(arr[j]):
partial = result[j]
else:
if ~is_nan(arr[j]):
partial += arr[j]
result[j] = partial
partial_sum[i] = partial
for i in prange(len(chunks)):
prefix = sum(partial_sum[0:i])
chunk = chunks[i]
for j in range(chunk.start, chunk.stop):
result[j] += prefix
return result
return nancumsum_impl
def np_nancumsum(arr, like_pandas=False):
if not isinstance(arr, types.Array):
return
if isinstance(arr.dtype, (types.Boolean, types.Integer)):
# dtype cannot possibly contain NaN
return lambda arr, like_pandas=False: cumsum(arr)
else:
retty = arr.dtype
is_nan = get_isnan(retty)
zero = retty(0)
def nancumsum_impl(arr, like_pandas=False):
chunks = parallel_chunks(len(arr))
partial_sum = numpy.zeros(len(chunks), dtype=retty)
result = numpy.empty_like(arr)
for i in prange(len(chunks)):
chunk = chunks[i]
partial = zero
for j in range(chunk.start, chunk.stop):
if like_pandas:
result[j] = partial + arr[j]
if ~is_nan(arr[j]):
partial = result[j]
else:
if ~is_nan(arr[j]):
partial += arr[j]
result[j] = partial
partial_sum[i] = partial
for i in prange(len(chunks)):
prefix = sum(partial_sum[0:i])
chunk = chunks[i]
for j in range(chunk.start, chunk.stop):
result[j] += prefix
return result
return nancumsum_impl
def np_nanvar(a):
if not isinstance(a, types.Array):
return
isnan = get_isnan(a.dtype)
def nanvar_impl(a):
# Compute the mean
m = nanmean(a)
# Compute the sum of square diffs
ssd = 0.0
count = 0
for i in prange(len(a)):
v = a[i]
if not isnan(v):
val = (v.item() - m)
ssd += np.real(val * np.conj(val))
count += 1
# np.divide() doesn't raise ZeroDivisionError
return np.divide(ssd, count)
return nanvar_impl
def np_nanprod(a):
"""
Reimplemented with parfor from numba.np.arraymath.
"""
if not isinstance(a, types.Array):
return
if isinstance(a.dtype, types.Integer):
retty = types.intp
else:
retty = a.dtype
one = retty(1)
isnan = get_isnan(a.dtype)
def nanprod_impl(a):
c = one
for i in prange(len(a)):
v = a[i]
if not isnan(v):
c *= v
return c
return nanprod_impl
def sdc_isnan_overload(self):
"""
Intel Scalable Dataframe Compiler Developer Guide
*************************************************
Parallel replacement of numpy.isnan.
.. only:: developer
Test: python -m sdc.runtests sdc.tests.test_sdc_numpy -k isnan
"""
if not isinstance(self, types.Array):
return None
ty_checker = TypeChecker("numpy-like 'isnan'")
dtype = self.dtype
isnan = get_isnan(dtype)
if isinstance(dtype, (types.Integer, types.Boolean, bool)):
def sdc_isnan_int_impl(self):
length = len(self)
res = numpy.zeros(shape=length, dtype=numpy.bool_)
return res
return sdc_isnan_int_impl
if isinstance(dtype, types.Float):
def sdc_isnan_float_impl(self):
length = len(self)
res = numpy.empty(shape=length, dtype=numpy.bool_)
for i in prange(length):
res[i] = isnan(self[i])
return res
return sdc_isnan_float_impl
ty_checker.raise_exc(dtype, 'int or float', 'self.dtype')
def dropna_overload(arr, idx, name):
dtype = arr.dtype
dtype_idx = idx.dtype
isnan = get_isnan(dtype)
def dropna_impl(arr, idx, name):
chunks = parallel_chunks(len(arr))
arr_len = numpy.empty(len(chunks), dtype=numpy.int64)
length = 0
for i in prange(len(chunks)):
chunk = chunks[i]
res = 0
for j in range(chunk.start, chunk.stop):
if not isnan(arr[j]):
res += 1
length += res
arr_len[i] = res
result_data = numpy.empty(shape=length, dtype=dtype)
result_index = numpy.empty(shape=length, dtype=dtype_idx)
for i in prange(len(chunks)):
chunk = chunks[i]
new_start = int(sum(arr_len[0:i]))
new_stop = new_start + arr_len[i]
current_pos = new_start
for j in range(chunk.start, chunk.stop):
if not isnan(arr[j]):
result_data[current_pos] = arr[j]
result_index[current_pos] = idx[j]
current_pos += 1
return pandas.Series(result_data, result_index, name)
return dropna_impl
def sdc_fillna_overload(self, inplace=False, value=None):
"""
Intel Scalable Dataframe Compiler Developer Guide
*************************************************
Parallel replacement of fillna.
.. only:: developer
Test: python -m sdc.runtests sdc.tests.test_sdc_numpy -k fillna
"""
if not isinstance(self, (types.Array, StringArrayType)):
return None
if not isinstance(inplace,
(types.Literal, types.Omitted) or inplace is False):
return None
dtype = self.dtype
isnan = get_isnan(dtype)
if (isinstance(inplace, types.Literal) and inplace.literal_value is True):
def sdc_fillna_inplace_noop(self, inplace=False, value=None):
return None
if isinstance(value, (types.NoneType, types.Omitted)) or value is None:
return sdc_fillna_inplace_noop
if isinstance(dtype, (types.Integer, types.Boolean)):
return sdc_fillna_inplace_noop
if isinstance(dtype, types.Float):
def sdc_fillna_inplace_float_impl(self, inplace=False, value=None):
_value = np.nan if value is None else value
length = len(self)
for i in prange(length):
if isnan(self[i]):
self[i] = _value
return None
return sdc_fillna_inplace_float_impl
if isinstance(dtype, types.UnicodeType):
# TO-DO: not supported, since no generic setitem for StringArray
return None
else:
def sdc_fillna_noop(self, inplace=False, value=None):
return copy(self)
if isinstance(value, (types.NoneType, types.Omitted)) or value is None:
return sdc_fillna_noop
if isinstance(dtype, (types.Integer, types.Boolean)):
return sdc_fillna_noop
if isinstance(dtype, types.Float):
def sdc_fillna_impl(self, inplace=False, value=None):
_value = np.nan if value is None else value
length = len(self)
filled_data = numpy.empty(length, dtype=dtype)
for i in prange(length):
if isnan(self[i]):
filled_data[i] = _value
else:
filled_data[i] = self[i]
return filled_data
return sdc_fillna_impl
if isinstance(self.dtype, types.UnicodeType):
def sdc_fillna_str_impl(self, inplace=False, value=None):
n = len(self)
num_chars = 0
# get total chars in new array
for i in prange(n):
s = self[i]
if sdc.hiframes.api.isna(self, i):
num_chars += get_utf8_size(value)
else:
num_chars += get_utf8_size(s)
filled_data = pre_alloc_string_array(n, num_chars)
# StringArray doesn't support parallel setitem, thus no prange here
for i in numpy.arange(n):
if sdc.hiframes.api.isna(self, i):
filled_data[i] = value
else:
filled_data[i] = self[i]
return filled_data
return sdc_fillna_str_impl
def arg_impl(self):
"""
Intel Scalable Dataframe Compiler Developer Guide
*************************************************
Parallel replacement of numpy.argmin/numpy.argmax.
.. only:: developer
Test: python -m sdc.runtests sdc.tests.test_sdc_numpy -k argmin
Test: python -m sdc.runtests sdc.tests.test_sdc_numpy -k argmax
"""
ty_checker = TypeChecker("numpy-like 'argmin'/'argmax'")
dtype = self.dtype
isnan = get_isnan(dtype)
max_int64 = max_dtype_int_val(numpy_support.from_dtype(numpy.int64))
if isinstance(dtype, types.Integer):
initial_result = {
min: max_dtype_int_val(dtype),
max: min_dtype_int_val(dtype),
}[reduce_op]
if isinstance(dtype, types.Float):
initial_result = {
min: max_dtype_float_val(dtype),
max: min_dtype_float_val(dtype),
}[reduce_op]
if not isinstance(self, types.Array):
return None
if isinstance(dtype, types.Number):
def sdc_argmin_impl(self):
chunks = parallel_chunks(len(self))
arr_res = numpy.empty(shape=len(chunks), dtype=dtype)
arr_pos = numpy.empty(shape=len(chunks), dtype=numpy.int64)
for i in prange(len(chunks)):
chunk = chunks[i]
res = initial_result
pos = max_int64
for j in range(chunk.start, chunk.stop):
if not isnan(self[j]):
if reduce_op(res, self[j]) != self[j]:
continue
if res == self[j]:
pos = min(pos, j)
else:
pos = j
res = self[j]
else:
if numpy.isnan(res):
pos = min(pos, j)
else:
pos = j
res = self[j]
arr_res[i] = res
arr_pos[i] = pos
general_res = initial_result
general_pos = max_int64
for i in range(len(chunks)):
if not isnan(arr_res[i]):
if reduce_op(general_res, arr_res[i]) != arr_res[i]:
continue
if general_res == arr_res[i]:
general_pos = min(general_pos, arr_pos[i])
else:
general_pos = arr_pos[i]
general_res = arr_res[i]
else:
if numpy.isnan(general_res):
general_pos = min(general_pos, arr_pos[i])
else:
general_pos = arr_pos[i]
general_res = arr_res[i]
return general_pos
return sdc_argmin_impl
ty_checker.raise_exc(dtype, 'number', 'self.dtype')
def sdc_fillna_overload(self, inplace=False, value=None):
"""
Intel Scalable Dataframe Compiler Developer Guide
*************************************************
Parallel replacement of fillna.
.. only:: developer
Test: python -m sdc.runtests sdc.tests.test_sdc_numpy -k fillna
"""
if not isinstance(self, (types.Array, StringArrayType)):
return None
dtype = self.dtype
isnan = get_isnan(dtype)
if ((isinstance(inplace, types.Literal) and inplace.literal_value == True)
or # noqa
(isinstance(inplace, bool) and inplace == True) # noqa
):
if isinstance(dtype, (types.Integer, types.Boolean)):
def sdc_fillna_inplace_int_impl(self, inplace=False, value=None):
return None
return sdc_fillna_inplace_int_impl
def sdc_fillna_inplace_float_impl(self, inplace=False, value=None):
length = len(self)
for i in prange(length):
if isnan(self[i]):
self[i] = value
return None
return sdc_fillna_inplace_float_impl
else:
if isinstance(self.dtype, types.UnicodeType):
def sdc_fillna_str_impl(self, inplace=False, value=None):
n = len(self)
num_chars = 0
# get total chars in new array
for i in prange(n):
s = self[i]
if sdc.hiframes.api.isna(self, i):
num_chars += len(value)
else:
num_chars += len(s)
filled_data = pre_alloc_string_array(n, num_chars)
for i in prange(n):
if sdc.hiframes.api.isna(self, i):
filled_data[i] = value
else:
filled_data[i] = self[i]
return filled_data
return sdc_fillna_str_impl
if isinstance(dtype, (types.Integer, types.Boolean)):
def sdc_fillna_int_impl(self, inplace=False, value=None):
return copy(self)
return sdc_fillna_int_impl
def sdc_fillna_impl(self, inplace=False, value=None):
length = len(self)
filled_data = numpy.empty(length, dtype=dtype)
for i in prange(length):
if isnan(self[i]):
filled_data[i] = value
else:
filled_data[i] = self[i]
return filled_data
return sdc_fillna_impl
