import math # ============================================================ # Helper function for floating point comparison # ============================================================ def isclose(a, b): return abs(a-b) < 0.000001 # ============================================================ # Constants: pi, e, inf, nan # ============================================================ assert isclose(math.pi, 3.141592653589793) assert isclose(math.e, 2.718281828459045) assert math.isinf(math.inf) assert math.isnan(math.nan) # ============================================================ # Rounding functions: ceil, floor, trunc, fabs # ============================================================ # ceil - round up to nearest integer assert math.ceil(1.2) == 2 assert math.ceil(-1.2) == -1 assert math.ceil(2.0) == 2 assert math.ceil(0) == 0 # floor - round down to nearest integer assert math.floor(1.2) == 1 assert math.floor(-1.2) == -2 assert math.floor(2.0) == 2 assert math.floor(0) == 0 # trunc - truncate towards zero assert math.trunc(1.7) == 1 assert math.trunc(-1.7) == -1 assert math.trunc(2.0) == 2 assert math.trunc(0) == 0 # fabs - absolute value (returns float) assert isclose(math.fabs(-1.2), 1.2) assert isclose(math.fabs(1.2), 1.2) assert isclose(math.fabs(0), 0.0) assert isclose(math.fabs(-0.0), 0.0) # ============================================================ # Classification functions: isnan, isinf, isfinite, isclose # ============================================================ # isnan - check if value is NaN assert math.isnan(float('nan')) assert math.isnan(math.nan) assert not math.isnan(1.0) assert not math.isnan(math.inf) # isinf - check if value is infinity assert math.isinf(float('inf')) assert math.isinf(float('-inf')) assert math.isinf(math.inf) assert not math.isinf(1.0) assert not math.isinf(math.nan) # isfinite - check if value is finite (not inf or nan) assert math.isfinite(1.0) assert math.isfinite(0.0) assert math.isfinite(-1.0) assert not math.isfinite(math.inf) assert not math.isfinite(-math.inf) assert not math.isfinite(math.nan) # isclose - check if two values are close assert math.isclose(1.0, 1.0) assert math.isclose(1.0, 1.0 + 1e-10) assert not math.isclose(1.0, 1.1) # ============================================================ # Power and logarithmic functions: exp, log, log2, log10, pow, sqrt # ============================================================ # exp - e raised to the power x assert isclose(math.exp(0), 1.0) assert isclose(math.exp(1), math.e) assert isclose(math.exp(2), math.e * math.e) # log - natural logarithm (base e) or logarithm with custom base assert isclose(math.log(1), 0.0) assert isclose(math.log(math.e), 1.0) assert isclose(math.log(10), 2.302585092994046) assert isclose(math.log(100, 10), 2.0) # log base 10 assert isclose(math.log(8, 2), 3.0) # log base 2 # log2 - logarithm base 2 assert isclose(math.log2(1), 0.0) assert isclose(math.log2(2), 1.0) assert isclose(math.log2(8), 3.0) assert isclose(math.log2(10), 3.321928094887362) # log10 - logarithm base 10 assert isclose(math.log10(1), 0.0) assert isclose(math.log10(10), 1.0) assert isclose(math.log10(100), 2.0) assert isclose(math.log10(1000), 3.0) # pow - x raised to the power y assert isclose(math.pow(2, 3), 8.0) assert isclose(math.pow(2, 0), 1.0) assert isclose(math.pow(2, -1), 0.5) assert isclose(math.pow(4, 0.5), 2.0) assert isclose(math.pow(0, 0), 1.0) # sqrt - square root assert isclose(math.sqrt(4), 2.0) assert isclose(math.sqrt(2), 1.4142135623730951) assert isclose(math.sqrt(0), 0.0) assert isclose(math.sqrt(1), 1.0) # ============================================================ # Trigonometric functions: sin, cos, tan, asin, acos, atan, atan2 # ============================================================ # sin - sine assert isclose(math.sin(0), 0.0) assert isclose(math.sin(math.pi / 2), 1.0) assert isclose(math.sin(math.pi), 0.0) assert isclose(math.sin(-math.pi / 2), -1.0) # cos - cosine assert isclose(math.cos(0), 1.0) assert isclose(math.cos(math.pi / 2), 0.0) assert isclose(math.cos(math.pi), -1.0) assert isclose(math.cos(-math.pi), -1.0) # tan - tangent assert isclose(math.tan(0), 0.0) assert isclose(math.tan(math.pi / 4), 1.0) assert isclose(math.tan(-math.pi / 4), -1.0) # asin - arc sine (inverse sine) assert isclose(math.asin(0), 0.0) assert isclose(math.asin(1), math.pi / 2) assert isclose(math.asin(-1), -math.pi / 2) assert isclose(math.asin(0.5), math.pi / 6) # acos - arc cosine (inverse cosine) assert isclose(math.acos(1), 0.0) assert isclose(math.acos(0), math.pi / 2) assert isclose(math.acos(-1), math.pi) assert isclose(math.acos(0.5), math.pi / 3) # atan - arc tangent (inverse tangent) assert isclose(math.atan(0), 0.0) assert isclose(math.atan(1), math.pi / 4) assert isclose(math.atan(-1), -math.pi / 4) # atan2 - arc tangent of y/x (handles quadrants correctly) assert isclose(math.atan2(0, 1), 0.0) assert isclose(math.atan2(1, 0), math.pi / 2) assert isclose(math.atan2(0, -1), math.pi) assert isclose(math.atan2(-1, 0), -math.pi / 2) assert isclose(math.atan2(1, 1), math.pi / 4) assert isclose(math.atan2(-1, -1), -3 * math.pi / 4) # ============================================================ # Angle conversion functions: degrees, radians # ============================================================ # degrees - convert radians to degrees assert isclose(math.degrees(0), 0.0) assert isclose(math.degrees(math.pi), 180.0) assert isclose(math.degrees(math.pi / 2), 90.0) assert isclose(math.degrees(2 * math.pi), 360.0) # radians - convert degrees to radians assert isclose(math.radians(0), 0.0) assert isclose(math.radians(180), math.pi) assert isclose(math.radians(90), math.pi / 2) assert isclose(math.radians(360), 2 * math.pi) # ============================================================ # Arithmetic functions: fmod, modf, copysign, fsum, gcd, factorial # ============================================================ # fmod - floating point modulo assert isclose(math.fmod(5.0, 3.0), 2.0) assert isclose(math.fmod(-5.0, 3.0), -2.0) assert isclose(math.fmod(5.0, -3.0), 2.0) assert isclose(math.fmod(-5.0, -3.0), -2.0) assert isclose(math.fmod(4.0, 3.0), 1.0) assert isclose(math.fmod(-4.0, 3.0), -1.0) assert isclose(math.fmod(-2.0, 3.0), -2.0) assert isclose(math.fmod(2.0, 3.0), 2.0) # modf - split into fractional and integer parts frac, integer = math.modf(1.5) assert isclose(frac, 0.5) assert isclose(integer, 1.0) frac, integer = math.modf(-1.5) assert isclose(frac, -0.5) assert isclose(integer, -1.0) frac, integer = math.modf(2.0) assert isclose(frac, 0.0) assert isclose(integer, 2.0) frac, integer = math.modf(0.0) assert isclose(frac, 0.0) assert isclose(integer, 0.0) # copysign - return x with the sign of y assert isclose(math.copysign(1.0, -1.0), -1.0) assert isclose(math.copysign(-1.0, 1.0), 1.0) assert isclose(math.copysign(1.0, 1.0), 1.0) assert isclose(math.copysign(-1.0, -1.0), -1.0) assert isclose(math.copysign(5.0, -0.0), -5.0) assert isclose(math.copysign(0.0, -1.0), 0.0) # sign of zero may vary # fsum - accurate floating point sum assert math.fsum([0.1] * 10) == 1.0 assert math.fsum([]) == 0.0 assert math.fsum([1.0, 2.0, 3.0]) == 6.0 assert math.fsum([0.1, 0.2, 0.3]) == 0.6 # gcd - greatest common divisor assert math.gcd(10, 5) == 5 assert math.gcd(10, 6) == 2 assert math.gcd(10, 7) == 1 assert math.gcd(10, 10) == 10 assert math.gcd(-10, 10) == 10 assert math.gcd(10, -10) == 10 assert math.gcd(-10, -10) == 10 assert math.gcd(0, 5) == 5 assert math.gcd(5, 0) == 5 assert math.gcd(0, 0) == 0 # factorial - n! assert math.factorial(0) == 1 assert math.factorial(1) == 1 assert math.factorial(2) == 2 assert math.factorial(3) == 6 assert math.factorial(4) == 24 assert math.factorial(5) == 120 assert math.factorial(10) == 3628800 # ============================================================ # Special value tests # ============================================================ # Test NaN generation from invalid operations a = -0.1 a = a ** a assert math.isnan(a) assert not math.isinf(a) # Test infinity assert math.isinf(float("inf")) assert math.isinf(float("-inf")) assert not math.isnan(float("inf"))