NVIDIA Clara Train 3.1
3.1

ai4med.libs.data package

class Balancer(items, categorizer: ai4med.libs.data.balancer.Categorizer, items_per_category='max')

Bases: object

balance()
class Categorizer

Bases: abc.ABC

abstract get_category_id(item)
abstract get_category_weight_factor(category_id)
class Cacher

Bases: object

contains(cache_id)
get(cache_id)
get_cache_ids()
initialize(cache_id)
put(cache_id, data)
replace(old_id, new_id, new_data)
saves_data()
class NullCacher

Bases: ai4med.libs.data.cacher.Cacher

get(cache_id)
put(cache_id, data)
replace(old_id, new_id, new_data)
saves_data()
class MultiLabelCategorizer

Bases: ai4med.libs.data.balancer.Categorizer

get_category_id(item)
get_category_weight_factor(category_id)
class SingleTaskCategorizer(weights, default_weight=1)

Bases: ai4med.libs.data.balancer.Categorizer

get_category_id(item)
get_category_weight_factor(category_id)
class ChainTransformer(build_ctx: ai4med.common.build_ctx.BuildContext)

Bases: object

get_stats()
transform(transforms, x)
class DatalistManager(list_gen: ai4med.libs.data.list_gen.ListGenerator, transforms, build_ctx: ai4med.common.build_ctx.BuildContext)

Bases: object

get_dataset_size()
get_list_generator()
initialize(state=- 1)
set_sharding(rank, num_shards, equal_shard_size=True, fixed_shard_data=False)
shutdown()
transform(transform_ctx: ai4med.common.transform_ctx.TransformContext)
class DatalistManagerWithCache(list_gen: ai4med.libs.data.list_gen.ListGenerator, transforms, build_ctx: ai4med.common.build_ctx.BuildContext, cache_obj_count, replace_percent=0.1, caches_data=True)

Bases: ai4med.libs.data.datalist_manager.DatalistManager

initialize(state=- 1)
set_sharding(rank, num_shards, equal_shard_size=True, fixed_shard_data=False)
shutdown()
transform(transform_ctx: ai4med.common.transform_ctx.TransformContext)
class DeterministicTransformer(transforms, chain_transformer)

Bases: ai4med.libs.data.smart_cache.CachePreparer

get_cache_id(data: dict)
prepare(data)
class Dataset(datalist_manager, output_shape_dict, output_type_dict)

Bases: object

get_dataset_size()
get_next_batch(session)
initialize(session, state=- 1)
static make_dataset(datalist_manager: ai4med.libs.data.datalist_manager.DatalistManager, output_shape_dict, output_type_dict, batch_size=1, num_workers=1, prefetch_size=0, shuffle_size=0, repeat_count=None)
set_sharding(rank, num_shards, equal_shard_size=True, fixed_shard_data=False)
shutdown()
class ListGenerator(items_list, output_types, extra_elements=None)

Bases: object

Filename generator.

This generator outputs a dict, which consists of items with any keys.

Parameters

items_list – this is a list of data items. Each item is a dict keyed by the names of data elements.

get_item_at_index(index)

Return an item at a specific index.

get_item_count()
get_items()
next()

Generator requires a callable object, which returns one data sample.

set_items(items_list)

Sets the item

Parameters

items_list – this is a list of data items. Each item is a dict keyed by the names of data elements.

sample_weights_by_categories(items_list)

“Calculates sample weights based on item count per class.

class CachePreparer

Bases: abc.ABC

get_cache_id(data)
abstract prepare(data)
class SmartCache(data_list, preparer, cache_count, replace_count, start_pos=0, cacher=None)

Bases: object

get_all_data_ids()
get_cached_data_ids()
get_data(x)
get_id_map()
manage_replacement()
shutdown()
start()
update_cache()
© Copyright 2020, NVIDIA. Last updated on Feb 2, 2023.