def test_tree_reduce_futures(n_parts, client):
a = [client.submit(s, i) for i in range(n_parts)]
b = tree_reduce(a)
c = client.compute(b, sync=True)
assert (sum(range(n_parts)) == c)
def test_tree_reduce_delayed(n_parts, client):
func = delayed(sum)
a = [delayed(i) for i in range(n_parts)]
b = tree_reduce(a, func=func)
c = client.compute(b, sync=True)
assert (sum(range(n_parts)) == c)
def test_tree_reduce_futures(n_parts, cluster):
client = Client(cluster)
try:
a = [client.submit(s, i) for i in range(n_parts)]
b = tree_reduce(a)
c = client.compute(b, sync=True)
assert (sum(range(n_parts)) == c)
finally:
client.close()
def score(self, X, y):
"""
Compute accuracy score
Parameters
----------
X : Dask.Array
Features to predict. Note- it is assumed that chunk sizes and
shape of X are known. This can be done for a fully delayed
Array by calling X.compute_chunks_sizes()
y : Dask.Array
Labels to use for computing accuracy. Note- it is assumed that
chunk sizes and shape of X are known. This can be done for a fully
delayed Array by calling X.compute_chunks_sizes()
Returns
-------
score : float the resulting accuracy score
"""
y_hat = self.predict(X)
@dask.delayed
def _count_accurate_predictions(y_hat, y):
y_hat = rmm_cupy_ary(cp.asarray, y_hat, dtype=y_hat.dtype)
y = rmm_cupy_ary(cp.asarray, y, dtype=y.dtype)
return y.shape[0] - cp.count_nonzero(y - y_hat)
delayed_parts = zip(y_hat.to_delayed(), y.to_delayed())
accuracy_parts = [
_count_accurate_predictions(*p) for p in delayed_parts
]
reduced = first(dask.compute(tree_reduce(accuracy_parts)))
return reduced / X.shape[0]