'Image processing in Tensor flow TFX pipelines
I am trying to get a Tensorflow TFX pipeline up and running using the MNIST dataset.
# Imports
import pandas as pd
import numpy as np
from keras.datasets import mnist
import tensorflow as tf
from tfx import v1 as tfx
import os
from tfx.components import ImportExampleGen
from platform import python_version
python_version() #'3.8.8'
# Load the data - 60,000 training examples and 10,000 testing examples
(train_x, train_y), (test_x, test_y) = mnist.load_data()
Setup pipeline paths
_pipeline_root = './pipeline'
_data_root = './data'
if not os.path.isdir(_pipeline_root) and not os.path.isdir(_data_root):
!mkdir {_pipeline_root}
!mkdir {_data_root}
Write the data to TF.record format and save in eval and train dirs. NOTE that the MNIST data starts as a numpy array 28x28 and is converted to a bytestring to enable it to be encoded as part of the Tf.record.
def _bytes_feature(value):
"""Returns a bytes_list from a string / byte."""
if isinstance(value, type(tf.constant(0))): # if value ist tensor
value = value.numpy() # get value of tensor
return tf.train.Feature(bytes_list=tf.train.BytesList(value=[value]))
def _int64_feature(value):
"""Returns an int64_list from a bool / enum / int / uint."""
return tf.train.Feature(int64_list=tf.train.Int64List(value=[value]))
def serialize_array(array):
array = tf.io.serialize_tensor(array)
return array
def image_label_to_tf_train(image, label):
image_shape = np.shape(image)
#define the dictionary -- the structure -- of our single example
data = {
'height': _int64_feature(image_shape[0]),
'width': _int64_feature(image_shape[1]),
'raw_image' : _bytes_feature(serialize_array(image)),
'label' : _int64_feature(label)
}
#create an Example, wrapping the single features
return tf.train.Example(features=tf.train.Features(feature=data))
def write_images_to_tfr_short(images, labels, filename:str="images", folder = ""):
if not os.path.isdir(folder):
!mkdir {folder}
filename= folder + "/" + filename+".tfrecords"
writer = tf.io.TFRecordWriter(filename) #create a writer that'll store our data to disk
count = 0
for index in range(len(images)):
#get the data we want to write
current_image = images[index]
current_label = labels[index]
out = image_label_to_tf_train(image=current_image, label=current_label)
writer.write(out.SerializeToString())
count += 1
writer.close()
print(f"Wrote {count} elements to TFRecord")
return count
The next stage is to call the transform component which uses the preprocessing_fn. This function should process all the data so for example divide the image array by 255 is a standard feature process. But the image is still as a bytestring and I can't for the life of me figure out how to turn it back into an array. The below is what I have tried.
def preprocessing_fn(inputs):
"""tf.transform's callback function for preprocessing inputs.
Args:
inputs: map from feature keys to raw not-yet-transformed features.
Returns:
Map from string feature key to transformed feature operations.
"""
# Initialize outputs dictionary
outputs = {}
raw_image_dataset = inputs[_IMAGE_KEY]
img = tf.io.decode_raw(raw_image_dataset, tf.int64)
outputs[_IMAGE_KEY] = img
outputs[_LABEL_KEY] = tf.cast(inputs[_LABEL_KEY], tf.int64)
return outputs
I get the following error:
WARNING:root:This output type hint will be ignored and not used for type-checking purposes. Typically, output type hints for a PTransform are single (or nested) types wrapped by a PCollection, PDone, or None. Got: Tuple[Dict[str, Union[NoneType, _Dataset]], Union[Dict[str, Dict[str, PCollection]], NoneType], int] instead.
WARNING:root:This output type hint will be ignored and not used for type-checking purposes. Typically, output type hints for a PTransform are single (or nested) types wrapped by a PCollection, PDone, or None. Got: Tuple[Dict[str, Union[NoneType, _Dataset]], Union[Dict[str, Dict[str, PCollection]], NoneType], int] instead.
WARNING:root:Make sure that locally built Python SDK docker image has Python 3.8 interpreter.
INFO:tensorflow:Assets written to: ./pipeline/Transform/transform_graph/225/.temp_path/tftransform_tmp/26150ae80de847fab932efeb0f0c610f/assets
INFO:tensorflow:Assets written to: ./pipeline/Transform/transform_graph/225/.temp_path/tftransform_tmp/26150ae80de847fab932efeb0f0c610f/assets
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/common.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.common.DoFnRunner.process()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/common.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.common.PerWindowInvoker.invoke_process()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/common.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.common.PerWindowInvoker._invoke_process_per_window()
/opt/conda/lib/python3.8/site-packages/apache_beam/transforms/core.py in <lambda>(x, *args, **kwargs)
1636 if fn_takes_side_inputs(fn):
-> 1637 wrapper = lambda x, *args, **kwargs: [fn(x, *args, **kwargs)]
1638 else:
/opt/conda/lib/python3.8/site-packages/tensorflow_transform/beam/impl.py in _create_v2_saved_model(tensor_replacement_map, base_temp_dir, preprocessing_fn, input_signature, baseline_analyzers_fingerprint, output_keys_to_name_map)
662 saved_model_dir = beam_common.get_unique_temp_path(base_temp_dir)
--> 663 impl_helper.trace_and_write_v2_saved_model(saved_model_dir, preprocessing_fn,
664 input_signature, base_temp_dir,
/opt/conda/lib/python3.8/site-packages/tensorflow_transform/impl_helper.py in trace_and_write_v2_saved_model(saved_model_dir, preprocessing_fn, input_signature, base_temp_dir, baseline_analyzers_fingerprint, tensor_replacement_map, output_keys_to_name_map)
893 analyzer_nodes.TENSOR_REPLACEMENTS):
--> 894 metadata = _trace_and_get_metadata(concrete_transform_fn, structured_inputs,
895 preprocessing_fn, base_temp_dir,
/opt/conda/lib/python3.8/site-packages/tensorflow_transform/impl_helper.py in _trace_and_get_metadata(concrete_transform_fn, structured_inputs, preprocessing_fn, base_temp_dir, tensor_replacement_map)
805 return dataset_metadata.DatasetMetadata(
--> 806 schema=schema_inference.infer_feature_schema_v2(
807 concrete_transform_fn.structured_outputs,
/opt/conda/lib/python3.8/site-packages/tensorflow_transform/schema_inference.py in infer_feature_schema_v2(features, concrete_metadata_fn, evaluate_schema_overrides)
255 metadata)
--> 256 return _infer_feature_schema_common(
257 features,
/opt/conda/lib/python3.8/site-packages/tensorflow_transform/schema_inference.py in _infer_feature_schema_common(features, tensor_ranges, feature_annotations, global_annotations, is_evaluation_complete)
300 min=min_value, max=max_value, is_categorical=True)
--> 301 feature_spec = _feature_spec_from_batched_tensors(features,
302 is_evaluation_complete)
/opt/conda/lib/python3.8/site-packages/tensorflow_transform/schema_inference.py in _feature_spec_from_batched_tensors(tensors, is_evaluation_complete)
128 dim is None for dim in shape.as_list()[1:]):
--> 129 raise ValueError(
130 'Feature {} ({}) had invalid shape {} for FixedLenFeature: apart '
ValueError: Feature raw_image (Tensor("Identity_1:0", shape=(None, 1, None), dtype=int64)) had invalid shape (None, 1, None) for FixedLenFeature: apart from the batch dimension, all dimensions must have known size
During handling of the above exception, another exception occurred:
ValueError Traceback (most recent call last)
<ipython-input-37-7beafa4fe436> in <module>
3 schema=schema_gen.outputs['schema'],
4 module_file=os.path.abspath(_mnist_transform_module))
----> 5 context.run(transform, enable_cache=False)
/opt/conda/lib/python3.8/site-packages/tfx/orchestration/experimental/interactive/interactive_context.py in run_if_ipython(*args, **kwargs)
61 # __IPYTHON__ variable is set by IPython, see
62 # https://ipython.org/ipython-doc/rel-0.10.2/html/interactive/reference.html#embedding-ipython.
---> 63 return fn(*args, **kwargs)
64 else:
65 absl.logging.warning(
/opt/conda/lib/python3.8/site-packages/tfx/orchestration/experimental/interactive/interactive_context.py in run(self, component, enable_cache, beam_pipeline_args)
181 telemetry_utils.LABEL_TFX_RUNNER: runner_label,
182 }):
--> 183 execution_id = launcher.launch().execution_id
184
185 return execution_result.ExecutionResult(
/opt/conda/lib/python3.8/site-packages/tfx/orchestration/launcher/base_component_launcher.py in launch(self)
198 # be immutable in this context.
199 # output_dict can still be changed, specifically properties.
--> 200 self._run_executor(execution_decision.execution_id,
201 copy.deepcopy(execution_decision.input_dict),
202 execution_decision.output_dict,
/opt/conda/lib/python3.8/site-packages/tfx/orchestration/launcher/in_process_component_launcher.py in _run_executor(self, execution_id, input_dict, output_dict, exec_properties)
71 # be immutable in this context.
72 # output_dict can still be changed, specifically properties.
---> 73 executor.Do(
74 copy.deepcopy(input_dict), output_dict, copy.deepcopy(exec_properties))
/opt/conda/lib/python3.8/site-packages/tfx/components/transform/executor.py in Do(self, input_dict, output_dict, exec_properties)
581 # remove the `_pip_dependencies` attribute.
582 with udf_utils.TempPipInstallContext(self._pip_dependencies):
--> 583 TransformProcessor().Transform(label_inputs, label_outputs, status_file)
584 logging.debug('Cleaning up temp path %s on executor success', temp_path)
585 io_utils.delete_dir(temp_path)
/opt/conda/lib/python3.8/site-packages/tfx/components/transform/executor.py in Transform(***failed resolving arguments***)
1114 materialization_format = (
1115 transform_paths_file_formats[-1] if materialize_output_paths else None)
-> 1116 self._RunBeamImpl(analyze_data_list, transform_data_list, preprocessing_fn,
1117 stats_options_updater_fn, force_tf_compat_v1,
1118 input_dataset_metadata, transform_output_path,
/opt/conda/lib/python3.8/site-packages/tfx/components/transform/executor.py in _RunBeamImpl(self, analyze_data_list, transform_data_list, preprocessing_fn, stats_options_updater_fn, force_tf_compat_v1, input_dataset_metadata, transform_output_path, raw_examples_data_format, temp_path, input_cache_dir, output_cache_dir, disable_statistics, per_set_stats_output_paths, materialization_format, analyze_paths_count, stats_output_paths, make_beam_pipeline_fn)
1496 for dataset in transform_data_list:
1497 infix = 'TransformIndex{}'.format(dataset.index)
-> 1498 (dataset.transformed
1499 | 'EncodeAndSerialize[{}]'.format(infix) >> beam.ParDo(
1500 self._RecordBatchToExamplesFn(transformed_schema_proto))
/opt/conda/lib/python3.8/site-packages/apache_beam/pipeline.py in __exit__(self, exc_type, exc_val, exc_tb)
594 try:
595 if not exc_type:
--> 596 self.result = self.run()
597 self.result.wait_until_finish()
598 finally:
/opt/conda/lib/python3.8/site-packages/apache_beam/pipeline.py in run(self, test_runner_api)
571 finally:
572 shutil.rmtree(tmpdir)
--> 573 return self.runner.run_pipeline(self, self._options)
574 finally:
575 if not is_in_ipython():
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/direct/direct_runner.py in run_pipeline(self, pipeline, options)
129 runner = BundleBasedDirectRunner()
130
--> 131 return runner.run_pipeline(pipeline, options)
132
133
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py in run_pipeline(self, pipeline, options)
197 options.view_as(pipeline_options.ProfilingOptions))
198
--> 199 self._latest_run_result = self.run_via_runner_api(
200 pipeline.to_runner_api(default_environment=self._default_environment))
201 return self._latest_run_result
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py in run_via_runner_api(self, pipeline_proto)
208 # TODO(pabloem, BEAM-7514): Create a watermark manager (that has access to
209 # the teststream (if any), and all the stages).
--> 210 return self.run_stages(stage_context, stages)
211
212 @contextlib.contextmanager
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py in run_stages(self, stage_context, stages)
393 )
394
--> 395 stage_results = self._run_stage(
396 runner_execution_context, bundle_context_manager)
397
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py in _run_stage(self, runner_execution_context, bundle_context_manager)
658 while True:
659 last_result, deferred_inputs, fired_timers, watermark_updates = (
--> 660 self._run_bundle(
661 runner_execution_context,
662 bundle_context_manager,
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py in _run_bundle(self, runner_execution_context, bundle_context_manager, data_input, data_output, input_timers, expected_timer_output, bundle_manager)
781 expected_timer_output)
782
--> 783 result, splits = bundle_manager.process_bundle(
784 data_input, data_output, input_timers, expected_timer_output)
785 # Now we collect all the deferred inputs remaining from bundle execution.
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/portability/fn_api_runner/fn_runner.py in process_bundle(self, inputs, expected_outputs, fired_timers, expected_output_timers, dry_run)
1092 process_bundle_descriptor.id,
1093 cache_tokens=[next(self._cache_token_generator)]))
-> 1094 result_future = self._worker_handler.control_conn.push(process_bundle_req)
1095
1096 split_results = [] # type: List[beam_fn_api_pb2.ProcessBundleSplitResponse]
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/portability/fn_api_runner/worker_handlers.py in push(self, request)
376 self._uid_counter += 1
377 request.instruction_id = 'control_%s' % self._uid_counter
--> 378 response = self.worker.do_instruction(request)
379 return ControlFuture(request.instruction_id, response)
380
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/worker/sdk_worker.py in do_instruction(self, request)
578 if request_type:
579 # E.g. if register is set, this will call self.register(request.register))
--> 580 return getattr(self, request_type)(
581 getattr(request, request_type), request.instruction_id)
582 else:
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/worker/sdk_worker.py in process_bundle(self, request, instruction_id)
616 with self.maybe_profile(instruction_id):
617 delayed_applications, requests_finalization = (
--> 618 bundle_processor.process_bundle(instruction_id))
619 monitoring_infos = bundle_processor.monitoring_infos()
620 monitoring_infos.extend(self.state_cache_metrics_fn())
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/worker/bundle_processor.py in process_bundle(self, instruction_id)
993 element.timer_family_id, timer_data)
994 elif isinstance(element, beam_fn_api_pb2.Elements.Data):
--> 995 input_op_by_transform_id[element.transform_id].process_encoded(
996 element.data)
997
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/worker/bundle_processor.py in process_encoded(self, encoded_windowed_values)
219 decoded_value = self.windowed_coder_impl.decode_from_stream(
220 input_stream, True)
--> 221 self.output(decoded_value)
222
223 def monitoring_infos(self, transform_id, tag_to_pcollection_id):
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/worker/operations.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.worker.operations.Operation.output()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/worker/operations.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.worker.operations.Operation.output()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/worker/operations.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.worker.operations.SingletonConsumerSet.receive()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/worker/operations.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.worker.operations.DoOperation.process()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/worker/operations.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.worker.operations.DoOperation.process()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/common.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.common.DoFnRunner.process()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/common.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.common.DoFnRunner._reraise_augmented()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/common.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.common.DoFnRunner.process()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/common.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.common.SimpleInvoker.invoke_process()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/common.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.common._OutputProcessor.process_outputs()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/worker/operations.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.worker.operations.SingletonConsumerSet.receive()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/worker/operations.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.worker.operations.DoOperation.process()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/worker/operations.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.worker.operations.DoOperation.process()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/common.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.common.DoFnRunner.process()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/common.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.common.DoFnRunner._reraise_augmented()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/common.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.common.DoFnRunner.process()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/common.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.common.SimpleInvoker.invoke_process()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/common.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.common._OutputProcessor.process_outputs()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/worker/operations.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.worker.operations.SingletonConsumerSet.receive()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/worker/operations.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.worker.operations.DoOperation.process()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/worker/operations.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.worker.operations.DoOperation.process()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/common.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.common.DoFnRunner.process()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/common.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.common.DoFnRunner._reraise_augmented()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/common.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.common.DoFnRunner.process()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/common.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.common.PerWindowInvoker.invoke_process()
/opt/conda/lib/python3.8/site-packages/apache_beam/runners/common.cpython-38-x86_64-linux-gnu.so in apache_beam.runners.common.PerWindowInvoker._invoke_process_per_window()
/opt/conda/lib/python3.8/site-packages/apache_beam/transforms/core.py in <lambda>(x, *args, **kwargs)
1635 from apache_beam.transforms.util import fn_takes_side_inputs
1636 if fn_takes_side_inputs(fn):
-> 1637 wrapper = lambda x, *args, **kwargs: [fn(x, *args, **kwargs)]
1638 else:
1639 wrapper = lambda x: [fn(x)]
/opt/conda/lib/python3.8/site-packages/tensorflow_transform/beam/impl.py in _create_v2_saved_model(tensor_replacement_map, base_temp_dir, preprocessing_fn, input_signature, baseline_analyzers_fingerprint, output_keys_to_name_map)
661 """
662 saved_model_dir = beam_common.get_unique_temp_path(base_temp_dir)
--> 663 impl_helper.trace_and_write_v2_saved_model(saved_model_dir, preprocessing_fn,
664 input_signature, base_temp_dir,
665 baseline_analyzers_fingerprint,
/opt/conda/lib/python3.8/site-packages/tensorflow_transform/impl_helper.py in trace_and_write_v2_saved_model(saved_model_dir, preprocessing_fn, input_signature, base_temp_dir, baseline_analyzers_fingerprint, tensor_replacement_map, output_keys_to_name_map)
892 if not concrete_transform_fn.graph.get_collection(
893 analyzer_nodes.TENSOR_REPLACEMENTS):
--> 894 metadata = _trace_and_get_metadata(concrete_transform_fn, structured_inputs,
895 preprocessing_fn, base_temp_dir,
896 tensor_replacement_map)
/opt/conda/lib/python3.8/site-packages/tensorflow_transform/impl_helper.py in _trace_and_get_metadata(concrete_transform_fn, structured_inputs, preprocessing_fn, base_temp_dir, tensor_replacement_map)
804 evaluate_schema_overrides=True)
805 return dataset_metadata.DatasetMetadata(
--> 806 schema=schema_inference.infer_feature_schema_v2(
807 concrete_transform_fn.structured_outputs,
808 concrete_metadata_fn,
/opt/conda/lib/python3.8/site-packages/tensorflow_transform/schema_inference.py in infer_feature_schema_v2(features, concrete_metadata_fn, evaluate_schema_overrides)
254 tensor_annotations, global_annotations = _get_schema_annotations_v2(
255 metadata)
--> 256 return _infer_feature_schema_common(
257 features,
258 tensor_ranges,
/opt/conda/lib/python3.8/site-packages/tensorflow_transform/schema_inference.py in _infer_feature_schema_common(features, tensor_ranges, feature_annotations, global_annotations, is_evaluation_complete)
299 domains[name] = schema_pb2.IntDomain(
300 min=min_value, max=max_value, is_categorical=True)
--> 301 feature_spec = _feature_spec_from_batched_tensors(features,
302 is_evaluation_complete)
303
/opt/conda/lib/python3.8/site-packages/tensorflow_transform/schema_inference.py in _feature_spec_from_batched_tensors(tensors, is_evaluation_complete)
127 if is_evaluation_complete and any(
128 dim is None for dim in shape.as_list()[1:]):
--> 129 raise ValueError(
130 'Feature {} ({}) had invalid shape {} for FixedLenFeature: apart '
131 'from the batch dimension, all dimensions must have known size'
ValueError: Feature raw_image (Tensor("Identity_1:0", shape=(None, 1, None), dtype=int64)) had invalid shape (None, 1, None) for FixedLenFeature: apart from the batch dimension, all dimensions must have known size [while running 'Analyze/CreateSavedModel[tf_v2_only]/CreateSavedModel']
I know the label feature is working as I can call the below code and get a print as so....
transform = tfx.components.Transform(
examples=example_gen.outputs['examples'],
schema=schema_gen.outputs['schema'],
module_file=os.path.abspath(_mnist_transform_module))
context.run(transform, enable_cache=False)
# Get the URI of the output artifact representing the transformed examples
train_uri = os.path.join(transform.outputs['transformed_examples'].get()[0].uri, 'Split-train')
# Get the list of files in this directory (all compressed TFRecord files)
tfrecord_filenames = [os.path.join(train_uri, name)
for name in os.listdir(train_uri)]
# Create a `TFRecordDataset` to read these files
dataset = tf.data.TFRecordDataset(tfrecord_filenames, compression_type="GZIP")
# Decode the first record and print output
for tfrecord in dataset.take(1):
serialized_example = tfrecord.numpy()
example = tf.train.Example()
example.ParseFromString(serialized_example)
print(example)
IF I remove the lines:
img = tf.io.decode_raw(raw_image_dataset, tf.int64)
outputs[_IMAGE_KEY] = img
I get printed
features {
feature {
key: "label"
value {
int64_list {
value: 5
}
}
}
}
This shows what I am doing to the label feature is working but I really can't figure how to transform the image bytes. Part of the issue is I'm not completely sure what the format is as it's just a tensor which is pretty opaque. It seems given the label operation I'm operating on a column of data effectively but again, can't figure the correct operation or syntax
Solution 1:[1]
For any future viewers this works
raw_image_dataset = tf.map_fn(fn = lambda x : tf.io.parse_tensor(x[0], tf.uint8, name=None), elems = raw_image_dataset, fn_output_signature = tf.TensorSpec((28,28),dtype=tf.uint8, name=None), infer_shape = True)
raw_image_dataset = tf.cast(raw_image_dataset, tf.int64)
outputs[_IMAGE_KEY] = raw_image_dataset
Solution 2:[2]
So I think I solved this using
raw_image_dataset = inputs[_IMAGE_KEY]
raw_image_dataset = tf.map_fn(fn = lambda x : tf.io.decode_image(x[0]) , elems = raw_image_dataset, dtype=tf.uint8)
Theres something about the data going in as a batch so needing to map it and also using the right component of the resulting tensor "x[0]", I'm still not 100% sure on why this is the case but it seems to run.
Now I'm struggling with TFX as it won't let me output features that are different to the what went in...
Sources
This article follows the attribution requirements of Stack Overflow and is licensed under CC BY-SA 3.0.
Source: Stack Overflow
| Solution | Source |
|---|---|
| Solution 1 | tim_ai |
| Solution 2 | tim_ai |