def inv_sinc(sinc_x, tol=1e-12):
# Invalid input
if sinc_x > 1.0:
raise ValueError("sinc_x > 1.0")
# Initial guess from reversion of Taylor series
# https://math.stackexchange.com/questions/3189307/inverse-of-frac-sinxx
x = t_pow = np.sqrt(6 * np.abs((1 - sinc_x)))
t_squared = t_pow * t_pow
for coeff in numba.literal_unroll(_SERIES_COEFFS):
t_pow *= t_squared
x += coeff * t_pow
# Use Newton Raphson to go the rest of the way
# https://www.wolframalpha.com/input/?i=simplify+%28sinc%5Bx%5D+-+c%29+%2F+D%5Bsinc%5Bx%5D%2Cx%5D
while True:
# evaluate delta between this iteration sinc(x) and original
sinx = np.sin(x)
𝞓sinc_x = (1.0 if x == 0.0 else sinx / x) - sinc_x
# Stop if converged
if np.abs(𝞓sinc_x) < tol:
break
# Next iteration
x -= (x * x * 𝞓sinc_x) / (x * np.cos(x) - sinx)
return x
def rebroadcast(x):
for axis, value in numba.literal_unroll(op_axis):
if value and x.shape[axis] != 1:
raise ValueError(
("Dimension in Rebroadcast's input was supposed to be 1")
)
return x
def bar(x):
acc = 0
for i in literal_unroll(a):
if i in {'1': x}:
acc += 1
else:
acc += np.sqrt(x[0, 0])
return np.sin(x), acc
def reshape(x, shape):
new_shape = create_zeros_tuple()
for i in numba.literal_unroll(range(ndim)):
new_shape = tuple_setitem(new_shape, i, shape[i])
return np.reshape(x, new_shape)
def bar(fcs):
x = 0
a = 10
i, j = fcs[0]
x += i(j(a))
for t in literal_unroll(fcs):
i, j = t
x += i(j(a))
return x
def pd_multi_index_names_impl(self):
levels_names = sdc_tuple_map(lambda x: x.name, self._levels)
# this exploits undesired side-effect of literal_unroll - type-unification
# of resulting list dtype that will be types.Optional(types.unicode_type)
# as using typed.List of Optional values currently fails to compile
res = []
for i in literal_unroll(levels_names):
res.append(i)
return res
def all_to_all(w_halosend, taille, mystruct, variables, scount, sdepl, rcount, rdepl):
w_halorecv = np.zeros(taille, dtype=mystruct)
for var in literal_unroll(variables):
s_msg = r_msg = 0
s_msg = [np.array(w_halosend[var]), (scount, sdepl), MPI.DOUBLE_PRECISION]
r_msg = [np.array(w_halorecv[var]), (rcount, rdepl), MPI.DOUBLE_PRECISION]
COMM.Alltoallv(s_msg, r_msg)
w_halorecv[var] = r_msg[0]
return w_halorecv
def _tuple_hash(tup):
tl = len(tup)
acc = _PyHASH_XXPRIME_5
for x in literal_unroll(tup):
lane = hash(x)
if lane == _Py_uhash_t(-1):
return -1
acc += lane * _PyHASH_XXPRIME_2
acc = _PyHASH_XXROTATE(acc)
acc *= _PyHASH_XXPRIME_1
acc += tl ^ (_PyHASH_XXPRIME_5 ^ _Py_uhash_t(3527539))
if acc == _Py_uhash_t(-1):
return process_return(1546275796)
return process_return(acc)
def set_field4(rec):
for f in literal_unroll(_FS):
rec[f] = 10
return rec
def set_field2(rec):
fs = ('e', 'f')
for f in literal_unroll(fs):
rec[f] = 10
return rec
def get_field4(rec):
out = 0
for f in literal_unroll(_FS):
out += rec[f]
return out
def impl(*args):
s = 0
for arg in literal_unroll(args):
s += len(arg)
return s
def pyfunc(rec):
x = np.zeros((n,))
for o in literal_unroll(keys):
x += rec[o]
return x
def impl(lst, x):
count = 0
for val in literal_unroll(lst):
if val == x:
count += 1
return count
def get_field2(rec):
fs = ("e", "f")
out = 0
for f in literal_unroll(fs):
out += rec[f]
return out
def impl(lst, item):
for val in literal_unroll(lst):
if val == item:
return True
return False
def get_field(rec):
out = 0
for f in literal_unroll(fs):
out += rec[f]
return out
def bar(fcs):
r = 0
for pair in literal_unroll(fcs):
f1, f2 = pair
r += f1() + f2(2)
return r
def set_field(rec):
for f in literal_unroll(fs):
rec[f] = 10
return rec
def run_etl():
# Read only these columns
keep_cols = ('YEAR', 'DATANUM', 'SERIAL', 'CBSERIAL', 'HHWT', 'GQ',
'PERNUM', 'SEX', 'AGE', 'INCTOT', 'EDUC', 'EDUCD',
'EDUC_HEAD', 'EDUC_POP', 'EDUC_MOM', 'EDUCD_MOM2',
'EDUCD_POP2', 'INCTOT_MOM', 'INCTOT_POP', 'INCTOT_MOM2',
'INCTOT_POP2', 'INCTOT_HEAD', 'SEX_HEAD', 'CPI99')
dtypes = {
'YEAR': np.float64,
'DATANUM': np.float64,
'SERIAL': np.float64,
'CBSERIAL': np.float64,
'HHWT': np.float64,
'GQ': np.float64,
'PERNUM': np.float64,
'SEX': np.float64,
'AGE': np.float64,
'INCTOT': np.float64,
'EDUC': np.float64,
'EDUCD': np.float64,
'EDUC_HEAD': np.float64,
'EDUC_POP': np.float64,
'EDUC_MOM': np.float64,
'EDUCD_MOM2': np.float64,
'EDUCD_POP2': np.float64,
'INCTOT_MOM': np.float64,
'INCTOT_POP': np.float64,
'INCTOT_MOM2': np.float64,
'INCTOT_POP2': np.float64,
'INCTOT_HEAD': np.float64,
'SEX_HEAD': np.float64,
'CPI99': np.float64
}
# https://rapidsai-data.s3.us-east-2.amazonaws.com/datasets/ipums_education2income_1970-2010.csv.gz
df = pd.read_csv('ipums_education2income_1970-2010.csv',
usecols=keep_cols,
dtype=dtypes)
mask = df['INCTOT'] != 9999999
df = df[mask]
df = df.reset_index(drop=True)
res = df['INCTOT'] * df['CPI99']
df = setitem_impl(df, 'INCTOT', res)
mask1 = df['EDUC'].notna()
df = df[mask1]
df = df.reset_index(drop=True)
mask2 = df['EDUCD'].notna()
df = df[mask2]
df = df.reset_index(drop=True)
for col in literal_unroll(keep_cols):
df[col].fillna(-1, inplace=True)
y = df['EDUC']
X = df.drop(columns='EDUC')
return X, y
def get_field2(rec):
fs = ('e', 'f')
out = 0
for f in literal_unroll(fs):
out += rec[f]
return out
def foo():
l = ['a', 1, 'a', 2, 'a', 3, 'b', 4, 'b', 5, 'c']
r = []
for x in literal_unroll(('a', 'd', 2, 6)):
r.append(x in l)
return r
def get_field_sum(rec):
out = 0
for f in literal_unroll(fields_gl):
out += rec[f]
return out