List of NumPy math supported functions for omnisci

[guilhermeleobas]

The following numpy functions are tested by creating UDF definitions taking with scalar inputs.

Trigonometric functions

• np.sin(x)
• np.cos(x)
• np.tan(x)
• np.arcsin(x)
• np.arccos(x)
• np.arctan(x)
• np.hypot(x,y) - see forbidden name section below
• np.arctan2(x,y)
• np.radians
• np.degrees
• np.deg2rad
• np.rad2deg

Hyperbolic functions

• np.sinh(x) - see forbidden name section below
• np.cosh(x) - see forbidden name section below
• np.tanh(x) - see forbidden name section below
• np.arcsinh(x)
• np.arccosh(x)
• np.arctanh(x)

Rounding

• np.around(a)
• np.round_(a) - see forbidden name section below
• np.rint(x) - see forbidden name section below
• np.fix(x) - not supported by Numba
• np.floor(x)
• np.ceil(x)
• np.trunc(x) - see forbidden name section below
• np.rint(x)
• np.spacing(x)
• np.nextafter(x, y)

Sums, products, differences

Needs a better list/array support.

• np.prod(x)
• np.sum(x)

Needs support for returning an omnisci array

• np.nanprod(x)
• np.nansum(x)
• np.cumprod(x)
• np.cumsum(x)
• np.nancumprod(x)
• np.nancumsum(x)
• np.diff(x)
• np.ediff1d(x)
• np.gradient(x)
• np.cross(x)
• np.trapz(x)

Exponents and logarithms

• np.exp(x)
• np.expm1(x) - see forbidden name section below
• np.exp2(x) - see forbidden name section below
• np.log(x)
• np.log10(x)
• np.log2(x) - see forbidden name section below
• np.log1p(x) - see forbidden name section below
• np.logaddexp(x, y)
• np.logaddexp2(x, y)
• np.frexp - not supported by Numba, returns (<float>, <int>)
• np.ldexp(x)

Rational routines

• np.lcm(x1, x2) - see forbidden name section below
• np.gcd(x1, x2) - see forbidden name section below
• np.left_shift(x1, x2)
• np.right_shift(x1, x2)

Arithmetic operations

• np.absolute(x)
• np.conjugate(x)
• np.fabs(x)
• np.fmin(x, y)
• np.fmax(x, y)
• np.minimum(x, y)
• np.maximum(x, y)
• np.negative(x)
• np.positive(x) - not supported by Numba
• np.sign(x)
• np.reciprocal(x)
• np.add(x1, x2)
• np.subtract(x1, x2)
• np.multiply(x1, x2)
• np.divide(x1, x2)
• np.true_divide(x1, x2)
• np.floor_divide(x1, x2)
• np.power(x1, x2)
• np.float_power(x1, x2) - not supported by Numba
• np.square(x)
• np.sqrt(x)
• np.cbrt(x) - not supported by Numba, np.cbrt cannot support in nopython mode  numba/numba#5385
• np.remainder(x1, x2)
• np.fmod(x1, x2)
• np.mod(x1, x2)
• np.modf(x)
• np.divmod(x1, x2) - not supported by Numba, returns tuple
• np.copysign(x, y)
• np.heaviside(x, y) - not supported by Numba, np.heaviside in no_python mode numba/numba#4706

Logical functions

• np.isfinite(x)
• np.isinf(x)
• np.isnan(x)
• np.isnat(x)
• np.signbit(x)
• np.less(x, y)
• np.less_equal(x, y)
• np.greater(x, y)
• np.greater_equal(x, y)
• np.equal(x, y)
• np.not_equal(x, y)
• np.logical_not(x)
• np.logical_or(x, y)
• np.logical_xor(x, y)
• np.logical_and(x, y)

Forbidden UDF names (OmnisciDB 5.1 or earlier)

The following names cannot be used as the names of UDF function definitions:

• fmod
• sinh
• cosh
• tanh
• rint
• trunc
• expm1
• exp2
• log2
• log1p
• gcd
• lcm
• around
• hypot

https://github.com/omnisci/omniscidb-internal/pull/4133 removes the above restriction.

[guilhermeleobas]

We need to test if those functions works when applied to an array. For instance:

@omnisci(double[](int64))
def apply_sin(size):
    arr = Array(size, 'double')
    for i in range(size):
        arr[i] = nb_types.double(i)
    return np.sin(arr)

query = 'select apply_sin(5);`
result, _ = omni.sql_execute(query)
out = list(result)[0]
print(out) # should print [ 0.        ,  0.84147098,  0.90929743,  0.14112001, -0.7568025 ]

Edit: They do not work on arrays.

[guilhermeleobas]

Closing this issue in favor of #31

[guilhermeleobas]

I will re-open this issue and update it to contain a table of math functions we support on CPU and CUDA.

[guilhermeleobas]

Closing again in favor of #31