def test_hash_numpy_different_values(self):
hashed1 = get_hash(np.array([1, 2], dtype=np.int64))
hashed2 = get_hash(np.array([1, 3], dtype=np.int64))
assert_hash(hashed1)
assert_hash(hashed2)
assert hashed1 != hashed2
def test_hash_numpy(self, dtype):
hashed1 = get_hash(np.array([1, 2], dtype=dtype))
hashed2 = get_hash(np.array([1, 2], dtype=dtype))
assert_hash(hashed1)
assert_hash(hashed2)
assert hashed1 == hashed2
def test_hash_object_different(self):
hashed1 = get_hash(dict(a=1, b='x'))
hashed2 = get_hash(dict(a=2, b='x'))
assert_hash(hashed1)
assert_hash(hashed2)
assert hashed1 != hashed2
def test_hash_object(self):
hashed1 = get_hash(dict(a=1, b='x'))
hashed2 = get_hash(dict(a=1, b='x'))
assert_hash(hashed1)
assert_hash(hashed2)
assert hashed1 == hashed2
def test_hash_scalar_different(self, obj):
hashed1 = get_hash(obj)
hashed2 = get_hash('different')
assert_hash(hashed1)
assert_hash(hashed2)
assert hashed1 != hashed2
def test_hash_scalar(self, obj):
hashed1 = get_hash(obj)
hashed2 = get_hash(obj)
assert_hash(hashed1)
assert_hash(hashed2)
assert hashed1 == hashed2
def test_hash_pandas_different(self):
df1 = pd.DataFrame(dict(a=[1, 2], b=[3, 4]))
hashed1 = get_hash(df1)
df2 = pd.DataFrame(dict(a=[1, 2], b=[2, 4]))
hashed2 = get_hash(df2)
assert_hash(hashed1)
assert_hash(hashed2)
assert hashed1 != hashed2
def maybe_pure(self, func, inputs, result):
"""
Check whether the function is pure.
Actually, it only compares the hash of the function result based on
its input.
"""
# inputs is a tuple of args, kwargs
assert isinstance(inputs, tuple)
assert len(inputs) == 2
args, kwargs = inputs
input_hash = get_hash(*args, **kwargs)
output_hash = get_hash(result)
input_key = func.__name__ + '-' + input_hash
previous_hash = self._update_step_hash(input_key, output_hash)
maybe_pure = (output_hash == previous_hash)
if not maybe_pure:
msg = ('Experiment step result is changed with the same input: '
'(step: {}, args: {}, kwargs: {})')
logger.warning(msg.format(func.__name__, args, kwargs))
return maybe_pure