def test_local_shuffle(ray_start_regular_shared): # confirm that no data disappears, and they all stay within the same shard it = from_range(8, num_shards=2).local_shuffle(shuffle_buffer_size=2) assert repr(it) == ("ParallelIterator[from_range[8, shards=2]" + ".local_shuffle(shuffle_buffer_size=2, seed=None)]") shard_0 = it.get_shard(0) shard_1 = it.get_shard(1) assert set(shard_0) == {0, 1, 2, 3} assert set(shard_1) == {4, 5, 6, 7} # check that shuffling results in different orders it1 = from_range(100, num_shards=10).local_shuffle(shuffle_buffer_size=5) it2 = from_range(100, num_shards=10).local_shuffle(shuffle_buffer_size=5) assert list(it1.gather_sync()) != list(it2.gather_sync()) # buffer size of 1 should not result in any shuffling it3 = from_range(10, num_shards=1).local_shuffle(shuffle_buffer_size=1) assert list(it3.gather_sync()) == list(range(10)) # statistical test it4 = from_items([0, 1] * 10000, num_shards=1).local_shuffle(shuffle_buffer_size=100) result = "".join(it4.gather_sync().for_each(str)) freq_counter = Counter(zip(result[:-1], result[1:])) assert len(freq_counter) == 4 for key, value in freq_counter.items(): assert value / len(freq_counter) > 0.2
def test_repartition_consistent(ray_start_regular_shared): # repartition should be deterministic it1 = from_range(9, num_shards=1).repartition(2) it2 = from_range(9, num_shards=1).repartition(2) assert it1.num_shards() == 2 assert it2.num_shards() == 2 assert set(it1.get_shard(0)) == set(it2.get_shard(0)) assert set(it1.get_shard(1)) == set(it2.get_shard(1))
def test_repartition_consistent(ray_start_regular_shared): # repartition should be deterministic it1 = from_range(9, num_shards=1).repartition(2) it2 = from_range(9, num_shards=1).repartition(2) # union should work after repartition it3 = it1.union(it2) assert it1.num_shards() == 2 assert it2.num_shards() == 2 assert set(it1.get_shard(0)) == set(it2.get_shard(0)) assert set(it1.get_shard(1)) == set(it2.get_shard(1)) assert it3.num_shards() == 4 assert set(it3.gather_async()) == set(it1.gather_async()) | set( it2.gather_async())
def test_transform(ray_start_regular_shared): def f(it): for item in it: yield item * 2 def g(it): for item in it: if item >= 2: yield item it = from_range(4).transform(f) assert repr(it) == "ParallelIterator[from_range[4, shards=2].transform()]" assert list(it.gather_sync()) == [0, 4, 2, 6] it = from_range(4) assert list(it.gather_sync().transform(g)) == [2, 3]
def test_local_shuffle(ray_start_regular_shared): para_it = parallel_it.from_range(100).for_each(lambda x: [x]) # batch_size larger than 1 and shuffle_buffer_size larger than 1 ds = ml_data.from_parallel_iter(para_it, batch_size=10) ds1 = ds.local_shuffle(shuffle_buffer_size=5) ds2 = ds.local_shuffle(shuffle_buffer_size=5) l1 = list(ds1.gather_sync()) l2 = list(ds2.gather_sync()) assert not all(df1.equals(df2) for df1, df2 in zip(l1, l2)) # batch_size equals 1 and shuffle_buffer_size larger than 1 ds = ml_data.from_parallel_iter(para_it, batch_size=1) ds1 = ds.local_shuffle(shuffle_buffer_size=5) ds2 = ds.local_shuffle(shuffle_buffer_size=5) l1 = list(ds1.gather_sync()) l2 = list(ds2.gather_sync()) assert not all(df1.equals(df2) for df1, df2 in zip(l1, l2)) # batch_size equals 1 and shuffle_buffer_size equals 1 ds = ml_data.from_parallel_iter(para_it, batch_size=1) ds1 = ds.local_shuffle(shuffle_buffer_size=1) ds2 = ds.local_shuffle(shuffle_buffer_size=1) l1 = list(ds1.gather_sync()) l2 = list(ds2.gather_sync()) assert all(df1.equals(df2) for df1, df2 in zip(l1, l2))
def test_gather_async(ray_start_regular_shared): it = from_range(4) it = it.gather_async() assert ( repr(it) == "LocalIterator[ParallelIterator[from_range[4, shards=2]]" ".gather_async()]") assert sorted(it) == [0, 1, 2, 3]
def test_get_shard_optimized(ray_start_regular_shared): it = from_range(6, num_shards=3) shard1 = it.get_shard(shard_index=0, batch_ms=25, num_async=2) shard2 = it.get_shard(shard_index=1, batch_ms=15, num_async=3) shard3 = it.get_shard(shard_index=2, batch_ms=5, num_async=4) assert list(shard1) == [0, 1] assert list(shard2) == [2, 3] assert list(shard3) == [4, 5]
def test_duplicate(ray_start_regular_shared): it = from_range(5, num_shards=1) it1, it2 = it.gather_sync().duplicate(2) it1 = it1.batch(2) it3 = it1.union(it2, deterministic=False) results = it3.take(20) assert results == [0, [0, 1], 1, 2, [2, 3], 3, 4, [4]]
def test_repartition_more(ray_start_regular_shared): it = from_range(100, 2).repartition(3) assert it.num_shards() == 3 assert set(it.get_shard(0)) == set(range(0, 50, 3)) | set( (range(50, 100, 3))) assert set( it.get_shard(1)) == set(range(1, 50, 3)) | set(range(51, 100, 3)) assert set( it.get_shard(2)) == set(range(2, 50, 3)) | set(range(52, 100, 3))
def test_repartition_less(ray_start_regular_shared): it = from_range(9, num_shards=3) it1 = it.repartition(2) assert repr(it1) == ("ParallelIterator[from_range[9, " + "shards=3].repartition[num_partitions=2]]") assert it1.num_shards() == 2 shard_0_set = set(it1.get_shard(0)) shard_1_set = set(it1.get_shard(1)) assert shard_0_set == {0, 2, 3, 5, 6, 8} assert shard_1_set == {1, 4, 7}
def test_from_parallel_it(ray_start_regular_shared): para_it = parallel_it.from_range(4).for_each(lambda x: [x]) ds = ml_data.from_parallel_iter(para_it, batch_size=2) assert repr(ds) == ("MLDataset[from_range[4, shards=2]" ".for_each().batch(2).to_pandas()]") collected = list(ds.gather_sync()) assert len(collected) == 2 assert all(d.shape == (2, 1) for d in collected) expected = para_it.flatten().batch(2).gather_sync().flatten() flattened = ds.gather_sync().for_each(lambda x: x[0].to_list()).flatten() assert list(flattened) == list(expected)
def test_repartition_less(ray_start_regular_shared): it = from_range(9, num_shards=3) # chaining operations after a repartition should work it1 = it.repartition(2).for_each(lambda x: 2 * x) assert repr(it1) == ("ParallelIterator[from_range[9, " + "shards=3].repartition[num_partitions=2].for_each()]") assert it1.num_shards() == 2 shard_0_set = set(it1.get_shard(0)) shard_1_set = set(it1.get_shard(1)) assert shard_0_set == {0, 4, 6, 10, 12, 16} assert shard_1_set == {2, 8, 14}
def test_union(ray_start_regular_shared): para_it1 = parallel_it.from_range(4, 2, False).for_each(lambda x: [x]) ds1 = ml_data.from_parallel_iter(para_it1, True, 2, False) para_it2 = parallel_it.from_range(4, 2, True).for_each(lambda x: [x]) ds2 = ml_data.from_parallel_iter(para_it2, True, 2, True) with pytest.raises(TypeError) as ex: ds1.union(ds2) assert "two MLDataset which have different repeated type" in str(ex.value) # union two MLDataset with same batch size para_it2 = parallel_it.from_range(4, 2, False).for_each(lambda x: [x]) ds2 = ml_data.from_parallel_iter(para_it2, True, 2, False) ds = ds1.union(ds2) assert ds.batch_size == 2 # union two MLDataset with different batch size para_it2 = parallel_it.from_range(4, 2, False).for_each(lambda x: [x]) ds2 = ml_data.from_parallel_iter(para_it2, True, 1, False) ds = ds1.union(ds2) # batch_size 0 means batch_size unknown assert ds.batch_size == 0
def test_metrics(ray_start_regular_shared): workers = make_workers(1) workers.foreach_worker(lambda w: w.sample()) a = from_range(10, repeat=True).gather_sync() b = StandardMetricsReporting( a, workers, { "min_iter_time_s": 2.5, "metrics_smoothing_episodes": 10, "collect_metrics_timeout": 10, }) start = time.time() res1 = next(b) assert res1["episode_reward_mean"] > 0, res1 res2 = next(b) assert res2["episode_reward_mean"] > 0, res2 assert time.time() - start > 2.4 workers.stop()
def test_metrics(self): workers = make_workers(1) workers.foreach_worker(lambda w: w.sample()) a = from_range(10, repeat=True).gather_sync() b = StandardMetricsReporting( a, workers, { "min_time_s_per_reporting": 2.5, "timesteps_per_iteration": 0, "metrics_num_episodes_for_smoothing": 10, "metrics_episode_collection_timeout_s": 10, "keep_per_episode_custom_metrics": False, }, ) start = time.time() res1 = next(b) assert res1["episode_reward_mean"] > 0, res1 res2 = next(b) assert res2["episode_reward_mean"] > 0, res2 assert time.time() - start > 2.4 workers.stop()
def test_from_range(ray_start_regular_shared): it = from_range(4) assert repr(it) == "ParallelIterator[from_range[4, shards=2]]" assert list(it.gather_sync()) == [0, 2, 1, 3]
def test_batch(ray_start_regular_shared): it = from_range(4, 1).batch(2) assert repr(it) == "ParallelIterator[from_range[4, shards=1].batch(2)]" assert list(it.gather_sync()) == [[0, 1], [2, 3]]
def test_gather_async_optimized(ray_start_regular_shared): it = from_range(100) it = it.gather_async(batch_ms=100, num_async=4) assert sorted(it) == list(range(100))
def test_gather_async_queue(ray_start_regular_shared): it = from_range(100) it = it.gather_async(num_async=4) assert sorted(it) == list(range(100))
def test_chain(ray_start_regular_shared): it = from_range(4).for_each(lambda x: x * 2).for_each(lambda x: x * 2) assert repr( it ) == "ParallelIterator[from_range[4, shards=2].for_each().for_each()]" assert list(it.gather_sync()) == [0, 8, 4, 12]
def test_filter(ray_start_regular_shared): it = from_range(4).filter(lambda x: x < 3) assert repr(it) == "ParallelIterator[from_range[4, shards=2].filter()]" assert list(it.gather_sync()) == [0, 2, 1]
def test_union_local(ray_start_regular_shared): it1 = from_items(["a", "b", "c"], 1).gather_async() it2 = from_range(5, 2).for_each(str).gather_async() it = it1.union(it2) assert sorted(it) == ["0", "1", "2", "3", "4", "a", "b", "c"]
def test_combine(ray_start_regular_shared): it = from_range(4, 1).combine(lambda x: [x, x]) assert repr(it) == "ParallelIterator[from_range[4, shards=1].combine()]" assert list(it.gather_sync()) == [0, 0, 1, 1, 2, 2, 3, 3]