コード例 #1
0
def mt_export_as_deliverable_model(
    output_dir,
    tensorflow_saved_model=None,
    converter_for_request: Union[None, Callable] = None,
    converter_for_response: Union[None, Callable] = None,
    keras_saved_model=None,
    keras_h5_model=None,
    meta_content_id="algorithmId-corpusId-configId-runId",
    lookup_tables: Dict = None,
    padding_parameter=None,
    addition_model_dependency=None,
    custom_object_dependency=None,
):
    # check parameters
    assert any(
        [tensorflow_saved_model, keras_saved_model, keras_h5_model]
    ), "one and only one of [tensorflow_saved_model, keras_saved_model, keras_h5_model] must be set up"
    assert (
        sum(
            int(bool(i)) for i in
            [tensorflow_saved_model, keras_saved_model, keras_h5_model]) == 1
    ), "one and only one of [tensorflow_saved_model, keras_saved_model, keras_h5_model] must be set up"

    # default value
    addition_model_dependency = ([] if addition_model_dependency is None else
                                 addition_model_dependency)
    custom_object_dependency = ([] if custom_object_dependency is None else
                                custom_object_dependency)

    # setup main object
    deliverable_model_builder = DeliverableModelBuilder(output_dir)

    # metadata builder
    metadata_builder = MetadataBuilder()

    meta_content = MetaContent(meta_content_id)

    metadata_builder.set_meta_content(meta_content)

    metadata_builder.save()

    # processor builder

    vocabulary_lookup_table = lookup_tables['vocab_lookup']
    tag_lookup_table = lookup_tables['tag_lookup']
    label_lookup_table = lookup_tables['label_lookup']

    processor_builder = ProcessorBuilder()

    decode_processor = BILUOEncodeProcessor()
    decoder_processor_handle = processor_builder.add_processor(
        decode_processor)

    pad_processor = PadProcessor(padding_parameter=padding_parameter)
    pad_processor_handle = processor_builder.add_processor(pad_processor)

    vocab_lookup_processor = LookupProcessor(vocabulary_lookup_table)
    vocab_lookup_processor_handle = processor_builder.add_processor(
        vocab_lookup_processor)

    tag_lookup_processor = LookupProcessor(tag_lookup_table)
    tag_lookup_processor_handle = processor_builder.add_processor(
        tag_lookup_processor)

    label_lookup_processor = LookupProcessor(
        label_lookup_table, **{
            "post_input_key": 'cls',
            "post_output_key": 'cls'
        })
    label_lookup_processor_handle = processor_builder.add_processor(
        label_lookup_processor)

    # # pre process: encoder[memory text] > lookup[str -> num] > pad[to fixed length]
    processor_builder.add_preprocess(decoder_processor_handle)
    processor_builder.add_preprocess(vocab_lookup_processor_handle)
    processor_builder.add_preprocess(pad_processor_handle)

    # # post process: lookup[num -> str] > encoder
    processor_builder.add_postprocess(tag_lookup_processor_handle)
    processor_builder.add_postprocess(label_lookup_processor_handle)
    processor_builder.add_postprocess(decoder_processor_handle)

    processor_builder.save()

    # model builder
    model_builder = ModelBuilder()
    model_builder.append_dependency(addition_model_dependency)
    model_builder.set_custom_object_dependency(custom_object_dependency)

    if converter_for_request:
        model_builder.add_converter_for_request(converter_for_request)

    if converter_for_response:
        model_builder.add_converter_for_response(converter_for_response)

    if tensorflow_saved_model:
        model_builder.add_tensorflow_saved_model(tensorflow_saved_model)
    elif keras_saved_model:
        model_builder.add_keras_saved_model(keras_saved_model)
    else:
        model_builder.add_keras_h5_model(keras_h5_model)

    model_builder.save()

    # compose all the parts
    deliverable_model_builder.add_processor(processor_builder)
    deliverable_model_builder.add_metadata(metadata_builder)
    deliverable_model_builder.add_model(model_builder)

    metadata = deliverable_model_builder.save()

    return metadata
コード例 #2
0
def export_as_deliverable_model(
    output_dir,
    tensorflow_saved_model=None,
    converter_for_request: Union[None, Callable] = None,
    converter_for_response: Union[None, Callable] = None,
    keras_saved_model=None,
    keras_h5_model=None,
    meta_content_id="algorithmId-corpusId-configId-runId",
    vocabulary_lookup_table=None,
    tag_lookup_table=None,
    padding_parameter=None,
    addition_model_dependency=None,
    custom_object_dependency=None,
):
    # check parameters
    assert any(
        [tensorflow_saved_model, keras_saved_model, keras_h5_model]
    ), "one and only one of [tensorflow_saved_model, keras_saved_model, keras_h5_model] must be set up"
    assert (
        sum(
            int(bool(i)) for i in
            [tensorflow_saved_model, keras_saved_model, keras_h5_model]) == 1
    ), "one and only one of [tensorflow_saved_model, keras_saved_model, keras_h5_model] must be set up"

    # default value
    addition_model_dependency = ([] if addition_model_dependency is None else
                                 addition_model_dependency)
    custom_object_dependency = ([] if custom_object_dependency is None else
                                custom_object_dependency)

    # setup main object
    deliverable_model_builder = DeliverableModelBuilder(output_dir)

    # metadata builder
    metadata_builder = MetadataBuilder()
    meta_content = MetaContent(meta_content_id)
    metadata_builder.set_meta_content(meta_content)
    metadata_builder.save()

    # processor builder
    processor_builder = ProcessorBuilder()

    vocab_lookup_processor = LookupProcessor(vocabulary_lookup_table)
    vocab_lookup_processor_handle = processor_builder.add_processor(
        vocab_lookup_processor)

    tag_lookup_processor = LookupProcessor(tag_lookup_table)
    tag_lookup_processor_handle = processor_builder.add_processor(
        tag_lookup_processor)

    pad_processor = PadProcessor(padding_parameter=padding_parameter)
    pad_processor_handle = processor_builder.add_processor(pad_processor)

    decode_processor = BILUOEncodeProcessor()
    decoder_processor_handle = processor_builder.add_processor(
        decode_processor)

    ## pre process: encoder > vocab_lookup > pad
    processor_builder.add_preprocess(decoder_processor_handle)
    processor_builder.add_preprocess(vocab_lookup_processor_handle)
    processor_builder.add_preprocess(pad_processor_handle)

    ## post process: tag_lookup > encoder
    processor_builder.add_postprocess(tag_lookup_processor_handle)
    processor_builder.add_postprocess(decoder_processor_handle)

    processor_builder.save()

    # model builder
    model_builder = ModelBuilder()
    model_builder.append_dependency(addition_model_dependency)
    model_builder.set_custom_object_dependency(custom_object_dependency)
    model_builder.add_converter_for_request(converter_for_request)
    model_builder.add_converter_for_response(converter_for_response)
    model_builder.add_keras_saved_model(keras_saved_model)
    model_builder.save()

    # remote model builder
    remote_model_builder = RemoteModelBuilder("tf+grpc")
    remote_model_builder.add_converter_for_request(
        RemoteKerasConverterForRequest())
    remote_model_builder.add_converter_for_response(
        RemoteKerasConverterForResponse())
    remote_model_builder.save()

    # compose all the parts
    deliverable_model_builder.add_processor(processor_builder)
    deliverable_model_builder.add_metadata(metadata_builder)
    deliverable_model_builder.add_model(model_builder)
    deliverable_model_builder.add_remote_model(remote_model_builder)

    metadata = deliverable_model_builder.save()

    return metadata
コード例 #3
0
def export_as_deliverable_model(
    output_dir,
    tensorflow_saved_model=None,
    converter_for_request: Union[None, Callable] = None,
    converter_for_response: Union[None, Callable] = None,
    keras_saved_model=None,
    keras_h5_model=None,
    meta_content_id="algorithmId-corpusId-configId-runId",
    vocabulary_lookup_table=None,
    tag_lookup_table=None,
    padding_parameter=None,
    str_tokens_convert_parameter=None,
    addition_model_dependency=None,
    custom_object_dependency=None,
):
    # check parameters
    assert any(
        [tensorflow_saved_model, keras_saved_model, keras_h5_model]
    ), "one and only one of [tensorflow_saved_model, keras_saved_model, keras_h5_model] must be set up"
    assert (
        sum(
            int(bool(i))
            for i in [tensorflow_saved_model, keras_saved_model, keras_h5_model]
        )
        == 1
    ), "one and only one of [tensorflow_saved_model, keras_saved_model, keras_h5_model] must be set up"

    # default value
    addition_model_dependency = (
        [] if addition_model_dependency is None else addition_model_dependency
    )
    custom_object_dependency = (
        [] if custom_object_dependency is None else custom_object_dependency
    )

    # setup main object
    deliverable_model_builder = DeliverableModelBuilder(output_dir)

    # metadata builder
    metadata_builder = MetadataBuilder()

    meta_content = MetaContent(meta_content_id)

    metadata_builder.set_meta_content(meta_content)

    metadata_builder.save()

    # processor builder
    processor_builder = ProcessorBuilder()

    decode_processor = BILUOEncodeProcessor()
    decoder_processor_handle = processor_builder.add_processor(decode_processor)

    lookup_processor = LookupProcessor()

    pad_processor = PadProcessor(padding_parameter=padding_parameter)

    if vocabulary_lookup_table:
        lookup_processor.add_vocabulary_lookup_table(vocabulary_lookup_table)

    if tag_lookup_table:
        lookup_processor.add_tag_lookup_table(tag_lookup_table)

    if vocabulary_lookup_table or tag_lookup_table:
        lookup_processor_handle = processor_builder.add_processor(lookup_processor)
        pad_processor_handle = processor_builder.add_processor(pad_processor)

    # # pre process: encoder > [lookup] > [pad]
    processor_builder.add_preprocess(decoder_processor_handle)

    if vocabulary_lookup_table or tag_lookup_table:
        processor_builder.add_preprocess(lookup_processor_handle)

    if vocabulary_lookup_table or tag_lookup_table:
        processor_builder.add_preprocess(pad_processor_handle)

    # # post process: lookup > encoder
    processor_builder.add_postprocess(decoder_processor_handle)
    if vocabulary_lookup_table or tag_lookup_table:
        processor_builder.add_postprocess(lookup_processor_handle)

    # add str_tokens_convert_processor
    str_tokens_convert_processor = StrTokensConvertProcessor()
    if str_tokens_convert_processor:
        str_tokens_convert_processor_handle = processor_builder.add_processor(str_tokens_convert_processor)
        processor_builder.add_preprocess(str_tokens_convert_processor_handle)

    processor_builder.save()

    # model builder
    model_builder = ModelBuilder()
    model_builder.append_dependency(addition_model_dependency)
    model_builder.set_custom_object_dependency(custom_object_dependency)

    if converter_for_request:
        model_builder.add_converter_for_request(converter_for_request)

    if converter_for_response:
        model_builder.add_converter_for_response(converter_for_response)

    if tensorflow_saved_model:
        model_builder.add_tensorflow_saved_model(tensorflow_saved_model)
    elif keras_saved_model:
        model_builder.add_keras_saved_model(keras_saved_model)
    else:
        model_builder.add_keras_h5_model(keras_h5_model)

    model_builder.save()

    # compose all the parts
    deliverable_model_builder.add_processor(processor_builder)
    deliverable_model_builder.add_metadata(metadata_builder)
    deliverable_model_builder.add_model(model_builder)

    metadata = deliverable_model_builder.save()

    return metadata