add GetDatasetSize to concat node

This commit is contained in:
mohammad 2021-01-21 16:25:48 -05:00
parent ddf84551ab
commit b5e91139e2
7 changed files with 149 additions and 29 deletions
mindspore
ccsrc/minddata/dataset
dataset/engine
tests/ut

View File

@ -402,7 +402,7 @@ std::shared_ptr<MnistDataset> Mnist(const std::string &dataset_dir, const std::s
// Function to overload "+" operator to concat two datasets
std::shared_ptr<ConcatDataset> operator+(const std::shared_ptr<Dataset> &datasets1,
const std::shared_ptr<Dataset> &datasets2) {
return std::make_shared<ConcatDataset>(std::vector({datasets2, datasets1}));
return std::make_shared<ConcatDataset>(std::vector({datasets1, datasets2}));
}
// Function to create a TextFileDataset.

View File

@ -73,6 +73,51 @@ Status ConcatNode::ValidateParams() {
return Status::OK();
}
// Get Dataset size
Status ConcatNode::GetDatasetSize(const std::shared_ptr<DatasetSizeGetter> &size_getter, bool estimate,
int64_t *dataset_size) {
if (dataset_size_ > 0) {
*dataset_size = dataset_size_;
return Status::OK();
}
// calculate the total size of all nodes
int64_t total_dataset_size = 0;
int64_t child_dataset_size = 0;
for (int idx = 0; idx < children_.size(); idx++) {
if (children_flag_and_nums_.empty() || children_flag_and_nums_[idx].second == 0) {
children_[idx]->GetDatasetSize(size_getter, false, &child_dataset_size);
total_dataset_size += child_dataset_size;
} else {
total_dataset_size += children_flag_and_nums_[idx].second;
}
}
// calculate the size of the shard
int64_t shard_dataset_size = 0;
if (sampler_ != nullptr) {
std::shared_ptr<DistributedSamplerRT> sampler_rt =
std::static_pointer_cast<DistributedSamplerRT>(sampler_->SamplerBuild());
sampler_rt->SetNumRowsInDataset(total_dataset_size);
sampler_rt->InitSampler();
// (total_size % num_shards != 0) & shard_id >= (remainder) ? CalculateNumSamples()-1 : CalculateNumSamples()
// example: 23 rows, 10 shards --> shard sizes = {3,3,3,2,2,2,2,2,2,2}
if ((sampler_rt->GetNumSamples() % sampler_rt->GetDeviceNum()) > 0 &&
sampler_rt->GetDeviceID() >= (sampler_rt->GetNumSamples() % sampler_rt->GetDeviceNum())) {
shard_dataset_size = sampler_rt->CalculateNumSamples(sampler_rt->GetNumSamples()) - 1;
} else {
shard_dataset_size = sampler_rt->CalculateNumSamples(sampler_rt->GetNumSamples());
}
} else {
shard_dataset_size = total_dataset_size;
}
*dataset_size = shard_dataset_size;
dataset_size_ = *dataset_size;
return Status::OK();
}
Status ConcatNode::Build(std::vector<std::shared_ptr<DatasetOp>> *const node_ops) {
if (children_flag_and_nums_.empty() || children_start_end_index_.empty()) {
node_ops->push_back(std::make_shared<ConcatOp>(connector_que_size_));

View File

@ -55,6 +55,15 @@ class ConcatNode : public DatasetNode {
/// \return Status Status::OK() if build successfully
Status Build(std::vector<std::shared_ptr<DatasetOp>> *const node_ops) override;
/// \brief Base-class override for GetDatasetSize
/// \param[in] size_getter Shared pointer to DatasetSizeGetter
/// \param[in] estimate This is only supported by some of the ops and it's used to speed up the process of getting
/// dataset size at the expense of accuracy.
/// \param[out] dataset_size the size of the dataset
/// \return Status of the function
Status GetDatasetSize(const std::shared_ptr<DatasetSizeGetter> &size_getter, bool estimate,
int64_t *dataset_size) override;
/// \brief Parameters validation
/// \return Status Status::OK() if all the parameters are valid
Status ValidateParams() override;

View File

@ -261,8 +261,8 @@ class Dataset : public std::enable_shared_from_this<Dataset> {
/// \param[in] datasets List of shared pointers to the dataset that should be concatenated together
/// \return Shared pointer to the current ConcatDataset
std::shared_ptr<ConcatDataset> Concat(const std::vector<std::shared_ptr<Dataset>> &datasets) {
std::vector<std::shared_ptr<Dataset>> all_datasets = datasets;
all_datasets.push_back(shared_from_this());
std::vector<std::shared_ptr<Dataset>> all_datasets{shared_from_this()};
all_datasets.insert(std::end(all_datasets), std::begin(datasets), std::end(datasets));
return std::make_shared<ConcatDataset>(all_datasets);
}

View File

@ -372,7 +372,7 @@ class Dataset:
Args:
condition_name (str): The condition name that is used to toggle sending next row.
num_batch (int): the number of batches without blocking at the start of each epoch.
callback (function): The callback funciton that will be invoked when sync_update is called.
callback (function): The callback function that will be invoked when sync_update is called.
Returns:
SyncWaitDataset, dataset added a blocking condition.
@ -398,7 +398,7 @@ class Dataset:
1. Make a shuffle buffer that contains the first buffer_size rows.
2. Randomly select an element from the shuffle buffer to be the next row
propogated to the child node.
propagated to the child node.
3. Get the next row (if any) from the parent node and put it in the shuffle buffer.
4. Repeat steps 2 and 3 until there are no more rows left in the shuffle buffer.
@ -1649,7 +1649,6 @@ class MappableDataset(SourceDataset):
def add_sampler(self, new_sampler):
# note: By adding a sampler, the sampled IDs will flow to new_sampler
# after first passing through the current samplers attached to this dataset.
if self.dataset_size is not None:
self.dataset_size = None
new_sampler.add_child(self.sampler)
self.sampler = new_sampler
@ -1676,7 +1675,6 @@ class MappableDataset(SourceDataset):
raise TypeError("Input sampler can not be None.")
if not isinstance(new_sampler, (samplers.BuiltinSampler, samplers.Sampler)):
raise TypeError("Input sampler is not an instance of a sampler.")
if self.dataset_size is not None:
self.dataset_size = None
self.sampler = self.sampler.child_sampler
@ -1718,7 +1716,7 @@ class MappableDataset(SourceDataset):
- The sum of split sizes < K, the difference will be added to the first split.
- The sum of split sizes > K, the difference will be removed from the first large
enough split such that it will have atleast 1 row after removing the difference.
enough split such that it will have at least 1 row after removing the difference.
randomize (bool, optional): Determines whether or not to split the data randomly (default=True).
If True, the data will be randomly split. Otherwise, each split will be created with
@ -2647,6 +2645,8 @@ class ConcatDataset(Dataset):
if sampler.get_num_samples() is not None:
raise ValueError("The parameter num_samples of DistributedSampler is not support to be set!")
self.dataset_size = None
self._sampler = _select_sampler(None, sampler, None, None, None)
cumulative_samples_nums = 0
for index, child in enumerate(self.children):

View File

@ -53,7 +53,7 @@ TEST_F(MindDataTestTreeModifying, AppendChild) {
std::shared_ptr<Dataset> ds2 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds6 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds3 = ds1->Take(10);
std::shared_ptr<Dataset> ds4 = ds2->Concat({ds3}); // ds2 is the second child and ds3 is the first child!!!
std::shared_ptr<Dataset> ds4 = ds3->Concat({ds2});
Status rc;
std::shared_ptr<DatasetNode> root = ds4->IRNode();
@ -110,7 +110,7 @@ TEST_F(MindDataTestTreeModifying, InsertChildAt01) {
std::shared_ptr<Dataset> ds3 = ds1->Take(10);
std::shared_ptr<Dataset> ds5 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds2 = ds5->Repeat(4);
std::shared_ptr<Dataset> ds4 = ds2->Concat({ds3}); // ds2 is the second child and ds3 is the first child!!!
std::shared_ptr<Dataset> ds4 = ds3->Concat({ds2});
Status rc;
std::shared_ptr<DatasetNode> root = ds4->IRNode();
auto ir_tree = std::make_shared<TreeAdapter>();
@ -173,7 +173,7 @@ TEST_F(MindDataTestTreeModifying, InsertChildAt04) {
std::shared_ptr<Dataset> ds3 = ds1->Take(10);
std::shared_ptr<Dataset> ds5 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds2 = ds5->Repeat(4);
std::shared_ptr<Dataset> ds4 = ds2->Concat({ds3}); // ds2 is the second child and ds3 is the first child!!!
std::shared_ptr<Dataset> ds4 = ds3->Concat({ds2});
Status rc;
std::shared_ptr<DatasetNode> root = ds4->IRNode();
auto ir_tree = std::make_shared<TreeAdapter>();
@ -253,7 +253,7 @@ TEST_F(MindDataTestTreeModifying, InsertAbove01) {
std::shared_ptr<Dataset> ds1 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds2 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds3 = ds1->Take(10);
std::shared_ptr<Dataset> ds4 = ds2->Concat({ds3}); // ds2 is the second child and ds3 is the first child!!!
std::shared_ptr<Dataset> ds4 = ds3->Concat({ds2});
Status rc;
std::shared_ptr<DatasetNode> root = ds4->IRNode();
@ -280,7 +280,7 @@ TEST_F(MindDataTestTreeModifying, InsertAbove02) {
std::shared_ptr<Dataset> ds1 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds2 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds3 = ds1->Take(10);
std::shared_ptr<Dataset> ds4 = ds2 + ds3; // ds2 is the second child and ds3 is the first child!!!
std::shared_ptr<Dataset> ds4 = ds3 + ds2;
Status rc;
std::shared_ptr<DatasetNode> root = ds4->IRNode();
@ -307,7 +307,7 @@ TEST_F(MindDataTestTreeModifying, InsertAbove03) {
std::shared_ptr<Dataset> ds1 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds2 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds3 = ds1->Take(10);
std::shared_ptr<Dataset> ds4 = ds2->Concat({ds3}); // ds2 is the second child and ds3 is the first child!!!
std::shared_ptr<Dataset> ds4 = ds3->Concat({ds2});
Status rc;
std::shared_ptr<DatasetNode> root = ds4->IRNode();
@ -372,9 +372,9 @@ TEST_F(MindDataTestTreeModifying, Drop01) {
std::shared_ptr<Dataset> ds9 = ds8->Skip(1);
std::shared_ptr<Dataset> ds3 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds2 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds4 = ds2->Concat({ds3}); // ds2 is the second child and ds3 is the first child!!!
std::shared_ptr<Dataset> ds4 = ds3->Concat({ds2});
std::shared_ptr<Dataset> ds6 = ds4->Take(13);
std::shared_ptr<Dataset> ds10 = ds6 + ds9;
std::shared_ptr<Dataset> ds10 = ds9 + ds6;
Status rc;
std::shared_ptr<DatasetNode> root = ds10->IRNode();
@ -437,9 +437,9 @@ TEST_F(MindDataTestTreeModifying, Drop03) {
std::shared_ptr<Dataset> ds9 = ds8->Skip(1);
std::shared_ptr<Dataset> ds3 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds2 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds4 = ds2->Concat({ds3}); // ds2 is the second child and ds3 is the first child!!!
std::shared_ptr<Dataset> ds4 = ds3->Concat({ds2});
std::shared_ptr<Dataset> ds6 = ds4->Take(13);
std::shared_ptr<Dataset> ds10 = ds6 + ds9;
std::shared_ptr<Dataset> ds10 = ds9 + ds6;
Status rc;
std::shared_ptr<DatasetNode> root = ds10->IRNode();
@ -487,11 +487,11 @@ TEST_F(MindDataTestTreeModifying, Drop04) {
std::shared_ptr<Dataset> ds9 = ds8->Skip(1);
std::shared_ptr<Dataset> ds3 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds2 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds4 = ds2->Concat({ds3}); // ds2 is the second child and ds3 is the first child!!!
std::shared_ptr<Dataset> ds4 = ds3->Concat({ds2});
std::shared_ptr<Dataset> ds5 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds1 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds6 = ds1->Concat({ds5, ds4}); // ds1 is put after (ds5, ds4)!!!
std::shared_ptr<Dataset> ds10 = ds6 + ds9;
std::shared_ptr<Dataset> ds6 = ds5->Concat({ds4, ds1});
std::shared_ptr<Dataset> ds10 = ds9 + ds6;
Status rc;
std::shared_ptr<DatasetNode> root = ds10->IRNode();
@ -548,8 +548,8 @@ TEST_F(MindDataTestTreeModifying, Drop05) {
std::shared_ptr<Dataset> ds4 = ds3->Skip(1);
std::shared_ptr<Dataset> ds5 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds1 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds6 = ds1->Concat({ds5, ds4}); // ds1 is put after (ds5, ds4)!!!
std::shared_ptr<Dataset> ds10 = ds6 + ds9;
std::shared_ptr<Dataset> ds6 = ds5->Concat({ds4, ds1});
std::shared_ptr<Dataset> ds10 = ds9 + ds6;
Status rc;
std::shared_ptr<DatasetNode> root = ds10->IRNode();
@ -603,11 +603,11 @@ TEST_F(MindDataTestTreeModifying, Drop06) {
std::shared_ptr<Dataset> ds9 = ds8->Skip(1);
std::shared_ptr<Dataset> ds3 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds2 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds4 = ds2->Concat({ds3}); // ds2 is the second child and ds3 is the first child!!!
std::shared_ptr<Dataset> ds4 = ds3->Concat({ds2});
std::shared_ptr<Dataset> ds5 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds1 = ImageFolder(folder_path, false, SequentialSampler(0, 11));
std::shared_ptr<Dataset> ds6 = ds1->Concat({ds5, ds4}); // ds1 is put after (ds5, ds4)!!!
std::shared_ptr<Dataset> ds10 = ds6 + ds9;
std::shared_ptr<Dataset> ds6 = ds5->Concat({ds4, ds1}); // ds1 is put after (ds5, ds4)!!!
std::shared_ptr<Dataset> ds10 = ds9 + ds6;
Status rc;
std::shared_ptr<DatasetNode> root = ds10->IRNode();

View File

@ -33,12 +33,19 @@ def generator_10():
for i in range(3, 10):
yield (np.array([i]),)
# In generator_20 dataset: Number of rows is 10; its values are 10, 11, 12 ... 19
def generator_20():
for i in range(10, 20):
yield (np.array([i]),)
# In generator_29 dataset: Number of rows is 9; its values are 20, 21, 22 ... 28
def generator_29():
for i in range(20, 29):
yield (np.array([i]),)
def test_concat_01():
"""
Test concat: test concat 2 datasets that have the same column name and data type
@ -316,7 +323,7 @@ def test_concat_13():
def test_concat_14():
"""
Test concat: create dataset with different dataset folder, and do diffrent operation then concat
Test concat: Testing concat on two different source datasets with different dataset operations.
"""
logger.info("test_concat_14")
DATA_DIR = "../data/dataset/testPK/data"
@ -365,6 +372,63 @@ def test_concat_15():
assert sum([1 for _ in data3]) == 47
def test_concat_16():
"""
Test concat: test get_dataset_size on nested concats
"""
logger.info("test_concat_16")
DATA_DIR = "../data/dataset/testPK/data"
DATA_DIR2 = ["../data/dataset/test_tf_file_3_images/train-0000-of-0001.data"]
data1 = ds.ImageFolderDataset(DATA_DIR)
data2 = ds.TFRecordDataset(DATA_DIR2, columns_list=["image"])
data3 = ds.GeneratorDataset(generator, ["col1"])
data4 = ds.GeneratorDataset(generator_10, ["col1"])
data5 = data1 + data2
data6 = data3 + data4
data7 = data5 + data6
ds.config.set_seed(1)
# 57 is the total size of all 4 leaf datasets
assert data7.get_dataset_size() == 57
def test_concat_17():
"""
Test concat: test get_dataset_size on nested concats (with sampler)
"""
logger.info("test_concat_17")
data1 = ds.GeneratorDataset(generator, ["col1"])
data2 = ds.GeneratorDataset(generator_10, ["col1"])
data3 = ds.GeneratorDataset(generator_20, ["col1"])
data4 = ds.GeneratorDataset(generator_29, ["col1"])
data5 = data1 + data2
data6 = data3 + data4
data7 = data5 + data6
ds.config.set_seed(1)
shard_num = 10
counter = 0
for i in range(shard_num):
distributed_sampler = ds.DistributedSampler(num_shards=shard_num, shard_id=i, shuffle=False, num_samples=None)
data7.use_sampler(distributed_sampler)
iter_counter = 0
for _ in data7.create_dict_iterator(num_epochs=1, output_numpy=True):
counter += 1
iter_counter += 1
assert data7.get_dataset_size() == iter_counter
# 29 is the total size of all 4 leaf datasets
assert counter == 29
if __name__ == "__main__":
test_concat_01()
test_concat_02()
@ -381,3 +445,5 @@ if __name__ == "__main__":
test_concat_13()
test_concat_14()
test_concat_15()
test_concat_16()
test_concat_17()