I am working on pre-processed data that were already siloed into separated csv files to represent separated local data for federated learning.
To correct implement the federated learning with these multiple CSVs on TensorFlow Federated, I am just trying to reproduce the same approach with a toy example in the iris dataset. However, when trying to use the method tff.simulation.datasets.TestClientData, I am getting the error:
TypeError: can't pickle _thread.RLock objects
The current code is as follows, first, load the three iris dataset CSV files (50 samples on each) into a dictionary from the filenames iris1.csv, iris2.csv, and iris3.csv:
silos = {}
for silo in silos_files:
silo_name = silo.replace(".csv", "")
silos[silo_name] = pd.read_csv(silos_path + silo)
silos[silo_name]["variety"].replace({"Setosa" : 0, "Versicolor" : 1, "Virginica" : 2}, inplace=True)
Creating a new dict with tensors:
silos_tf = collections.OrderedDict()
for key, silo in silos.items():
silos_tf[key] = tf.data.Dataset.from_tensor_slices((silo.drop(columns=["variety"]).values, silo["variety"].values))
Finally, trying to converting the Tensorflow Dataset into a Tensorflow Federated Dataset:
tff_dataset = tff.simulation.datasets.TestClientData(
silos_tf
)
That raises the error:
TypeError Traceback (most recent call last)
<ipython-input-58-a4b5686509ce> in <module>()
1 tff_dataset = tff.simulation.datasets.TestClientData(
----> 2 silos_tf
3 )
/usr/local/lib/python3.7/dist-packages/tensorflow_federated/python/simulation/datasets/from_tensor_slices_client_data.py in __init__(self, tensor_slices_dict)
59 """
60 py_typecheck.check_type(tensor_slices_dict, dict)
---> 61 tensor_slices_dict = copy.deepcopy(tensor_slices_dict)
62 structures = list(tensor_slices_dict.values())
63 example_structure = structures[0]
...
/usr/lib/python3.7/copy.py in deepcopy(x, memo, _nil)
167 reductor = getattr(x, "__reduce_ex__", None)
168 if reductor:
--> 169 rv = reductor(4)
170 else:
171 reductor = getattr(x, "__reduce__", None)
TypeError: can't pickle _thread.RLock objects
I also tried to use Python dictionary instead of OrderedDict but the error is the same. For this experiment, I am using Google Colab with this notebook as reference running with TensorFlow 2.8.0 and TensorFlow Federated version 0.20.0. I also used these previous questions as references:
Is there a reasonable way to create tff clients datat sets?
'tensorflow_federated.python.simulation' has no attribute 'FromTensorSlicesClientData' when using tff-nightly
I am not sure if this is a good way that derives for a case beyond the toy example, please, if any suggestion on how to bring already siloed data for TFF tests, I am thankful.
I did some search of public code in github using class tff.simulation.datasets.TestClientData, then I found the following implementation (source here):
def to_ClientData(clientsData: np.ndarray, clientsDataLabels: np.ndarray,
ds_info, is_train=True) -> tff.simulation.datasets.TestClientData:
"""Transform dataset to be fed to fedjax
:param clientsData: dataset for each client
:param clientsDataLabels:
:param ds_info: dataset information
:param train: True if processing train split
:return: dataset for each client cast into TestClientData
"""
num_clients = ds_info['num_clients']
client_data = collections.OrderedDict()
for i in range(num_clients if is_train else 1):
client_data[str(i)] = collections.OrderedDict(
x=clientsData[i],
y=clientsDataLabels[i])
return tff.simulation.datasets.TestClientData(client_data)
I understood from this snippet that the tff.simulation.datasets.TestClientData class requires as argument an OrderedDict composed by numpy arrays instead of a dict of tensors (as my previous implementation), now I changed the code for the following:
silos_tf = collections.OrderedDict()
for key, silo in silos.items():
silos_tf[key] = collections.OrderedDict(
x=silo.drop(columns=["variety"]).values,
y=silo["variety"].values)
Followed by:
tff_dataset = tff.simulation.datasets.TestClientData(
silos_tf
)
That correctly runs as the following output:
>>> tff_dataset.client_ids
['iris3', 'iris1', 'iris2']
Related
When trying to run the ljspeech example, I get the following error, even when the model is moved to the only GPU in the system. I am using Cuda 11.7, Pytorch 1.13.1, and Fairseq 0.12.2.
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument index in method wrapper__index_select)
The code used:
from fairseq.checkpoint_utils import load_model_ensemble_and_task_from_hf_hub
from fairseq.models.text_to_speech.hub_interface import TTSHubInterface
import IPython.display as ipd
import torch
models, cfg, task = load_model_ensemble_and_task_from_hf_hub(
"facebook/fastspeech2-en-ljspeech",
arg_overrides={"vocoder": "hifigan", "fp16": False}
)
model = models[0].to(torch.device('cuda'))
models[0] = model
TTSHubInterface.update_cfg_with_data_cfg(cfg, task.data_cfg)
generator = task.build_generator(models, cfg)
text = "Hello, this is a test run."
sample = TTSHubInterface.get_model_input(task, text)
wav, rate = TTSHubInterface.get_prediction(task, model, generator, sample)
ipd.Audio(wav, rate=rate)
I'm using the following code out of the box from this url: https://lightning-transformers.readthedocs.io/en/latest/tasks/nlp/question_answering.html
import pytorch_lightning as pl
from transformers import AutoTokenizer
from lightning_transformers.task.nlp.question_answering import (
QuestionAnsweringTransformer,
SquadDataModule,
)
tokenizer = AutoTokenizer.from_pretrained(pretrained_model_name_or_path="bert-base-uncased")
model = QuestionAnsweringTransformer(pretrained_model_name_or_path="bert-base-uncased")
dm = SquadDataModule(
batch_size=1,
dataset_config_name="plain_text",
max_length=384,
version_2_with_negative=False,
null_score_diff_threshold=0.0,
doc_stride=128,
n_best_size=20,
max_answer_length=30,
tokenizer=tokenizer,
)
trainer = pl.Trainer(accelerator="auto", devices="auto", max_epochs=1)
trainer.fit(model, dm)
which throws this error
AssertionError Traceback (most recent call last)
<ipython-input-2-0b608c02a52e> in <module>
14 trainer = pl.Trainer(accelerator="auto", devices="auto", max_epochs=1)
15
---> 16 trainer.fit(model, dm)
16 frames
/usr/local/lib/python3.8/dist-packages/lightning_transformers/task/nlp/question_answering/datasets/squad/processing.py in postprocess_qa_predictions(examples, features, predictions, version_2_with_negative, n_best_size, max_answer_length, null_score_diff_threshold, output_dir, prefix)
245 all_start_logits, all_end_logits, example_ids = predictions
246
--> 247 assert len(predictions[0]) == len(features), f"Got {len(predictions[0])} predictions and {len(features)} features."
248
249 # Build a map example to its corresponding features.
AssertionError: Got 2 predictions and 10784 features.
I was simply trying to get a single example from the documentation to run within google colab before investigating further if this would meet my use case, but I see an error when I try to use the example as is, which is disheartening to consider investigating it. Nothing comes up when I google "AssertionError: Got 2 predictions and 10784 features."
In this Notebook, we use Explainable AI SDK from Google to load a model, right after saving it. This fails with a message that the model is missing.
But note
the info message saying that the model was saved
checking working/model shows that the model is there.
However, working/model/assets is empty.
Why do we get this error message? How can we avoid it?
model_path = "working/model"
model.save(model_path)
builder = SavedModelMetadataBuilder(model_path)
builder.set_numeric_metadata(
"numpy_inputs",
input_baselines=[X_train.median().tolist()], # attributions relative to the median of the target
index_feature_mapping=X_train.columns.tolist(), # the names of each feature
)
builder.save_metadata(model_path)
explainer = explainable_ai_sdk.load_model_from_local_path(
model_path=model_path,
config=configs.SampledShapleyConfig(path_count=20),
)
INFO:tensorflow:Assets written to: working/model/assets
---------------------------------------------------------------------------
NotImplementedError Traceback (most recent call last)
/tmp/ipykernel_26061/1928503840.py in <module>
18 explainer = explainable_ai_sdk.load_model_from_local_path(
19 model_path=model_path,
---> 20 config=configs.SampledShapleyConfig(path_count=20),
21 )
22
/opt/conda/lib/python3.7/site-packages/explainable_ai_sdk/model/model_factory.py in load_model_from_local_path(model_path, config)
128 """
129 if _LOCAL_MODEL_KEY not in _MODEL_REGISTRY:
--> 130 raise NotImplementedError('There are no implementations of local model.')
131 return _MODEL_REGISTRY[_LOCAL_MODEL_KEY](model_path, config)
132
NotImplementedError: There are no implementations of local model.
I am using kedro.extras.datasets.pandas.SQLTableDataSet and would like to use the chunk_size argument from pandas. However, when running the pipeline, the table gets treated as a generator instead of a pd.dataframe().
How would you use the chunk_size within the pipeline?
My catalog:
table_name:
type: pandas.SQLTableDataSet
credentials: redshift
table_name : rs_table_name
layer: output
save_args:
if_exists: append
schema: schema.name
chunk_size: 1000
Looking at the latest pandas doc, the actual kwarg to be used is chunksize, not chunk_size. Please see https://pandas.pydata.org/docs/reference/api/pandas.DataFrame.to_sql.html.
Since kedro only wraps your save_args and passes them to pd.DataFrame.to_sql these need to match:
def _save(self, data: pd.DataFrame) -> None:
try:
data.to_sql(**self._save_args)
except ImportError as import_error:
raise _get_missing_module_error(import_error) from import_error
except NoSuchModuleError as exc:
raise _get_sql_alchemy_missing_error() from exc
EDIT: Once you have this working in your pipeline, the docs show that pandas.DataFrame.read_sql with chunksize set will return type Iterator[DataFrame]. This means that in your node function, you should iterate over the input (and annotate accordingly, if appropriate) such as:
def my_node_func(input_dfs: Iterator[pd.DataFrame], *args):
for df in input_dfs:
...
This works for the latest version of pandas. I have noticed, however, that pandas is aligning the API so that read_csv with chunksize set returns a ContextManager from pandas>=1.2 so I would expect this change to occur in read_sql as well.
I'm having trouble saving the output given by the Google Vision API. I'm using Python and testing with a demo image. I get the following error:
TypeError: [mid:...] + is not JSON serializable
Code that I executed:
import io
import os
import json
# Imports the Google Cloud client library
from google.cloud import vision
from google.cloud.vision import types
# Instantiates a client
vision_client = vision.ImageAnnotatorClient()
# The name of the image file to annotate
file_name = os.path.join(
os.path.dirname(__file__),
'demo-image.jpg') # Your image path from current directory
# Loads the image into memory
with io.open(file_name, 'rb') as image_file:
content = image_file.read()
image = types.Image(content=content)
# Performs label detection on the image file
response = vision_client.label_detection(image=image)
labels = response.label_annotations
print('Labels:')
for label in labels:
print(label.description, label.score, label.mid)
with open('labels.json', 'w') as fp:
json.dump(labels, fp)
the output appears on the screen, however I do not know exactly how I can save it. Anyone have any suggestions?
FYI to anyone seeing this in the future, google-cloud-vision 2.0.0 has switched to using proto-plus which uses different serialization/deserialization code. A possible error you can get if upgrading to 2.0.0 without changing the code is:
object has no attribute 'DESCRIPTOR'
Using google-cloud-vision 2.0.0, protobuf 3.13.0, here is an example of how to serialize and de-serialize (example includes json and protobuf)
import io, json
from google.cloud import vision_v1
from google.cloud.vision_v1 import AnnotateImageResponse
with io.open('000048.jpg', 'rb') as image_file:
content = image_file.read()
image = vision_v1.Image(content=content)
client = vision_v1.ImageAnnotatorClient()
response = client.document_text_detection(image=image)
# serialize / deserialize proto (binary)
serialized_proto_plus = AnnotateImageResponse.serialize(response)
response = AnnotateImageResponse.deserialize(serialized_proto_plus)
print(response.full_text_annotation.text)
# serialize / deserialize json
response_json = AnnotateImageResponse.to_json(response)
response = json.loads(response_json)
print(response['fullTextAnnotation']['text'])
Note 1: proto-plus doesn't support converting to snake_case names, which is supported in protobuf with preserving_proto_field_name=True. So currently there is no way around the field names being converted from response['full_text_annotation'] to response['fullTextAnnotation']
There is an open closed feature request for this: googleapis/proto-plus-python#109
Note 2: The google vision api doesn't return an x coordinate if x=0. If x doesn't exist, the protobuf will default x=0. In python vision 1.0.0 using MessageToJson(), these x values weren't included in the json, but now with python vision 2.0.0 and .To_Json() these values are included as x:0
Maybe you were already able to find a solution to your issue (if that is the case, I invite you to share it as an answer to your own post too), but in any case, let me share some notes that may be useful for other users with a similar issue:
As you can check using the the type() function in Python, response is an object of google.cloud.vision_v1.types.AnnotateImageResponse type, while labels[i] is an object of google.cloud.vision_v1.types.EntityAnnotation type. None of them seem to have any out-of-the-box implementation to transform them to JSON, as you are trying to do, so I believe the easiest way to transform each of the EntityAnnotation in labels would be to turn them into Python dictionaries, then group them all into an array, and transform this into a JSON.
To do so, I have added some simple lines of code to your snippet:
[...]
label_dicts = [] # Array that will contain all the EntityAnnotation dictionaries
print('Labels:')
for label in labels:
# Write each label (EntityAnnotation) into a dictionary
dict = {'description': label.description, 'score': label.score, 'mid': label.mid}
# Populate the array
label_dicts.append(dict)
with open('labels.json', 'w') as fp:
json.dump(label_dicts, fp)
There is a library released by Google
from google.protobuf.json_format import MessageToJson
webdetect = vision_client.web_detection(blob_source)
jsonObj = MessageToJson(webdetect)
I was able to save the output with the following function:
# Save output as JSON
def store_json(json_input):
with open(json_file_name, 'a') as f:
f.write(json_input + '\n')
And as #dsesto mentioned, I had to define a dictionary. In this dictionary I have defined what types of information I would like to save in my output.
with open(photo_file, 'rb') as image:
image_content = base64.b64encode(image.read())
service_request = service.images().annotate(
body={
'requests': [{
'image': {
'content': image_content
},
'features': [{
'type': 'LABEL_DETECTION',
'maxResults': 20,
},
{
'type': 'TEXT_DETECTION',
'maxResults': 20,
},
{
'type': 'WEB_DETECTION',
'maxResults': 20,
}]
}]
})
The objects in the current Vision library lack serialization functions (although this is a good idea).
It is worth noting that they are about to release a substantially different library for Vision (it is on master of vision's repo now, although not released to PyPI yet) where this will be possible. Note that it is a backwards-incompatible upgrade, so there will be some (hopefully not too much) conversion effort.
That library returns plain protobuf objects, which can be serialized to JSON using:
from google.protobuf.json_format import MessageToJson
serialized = MessageToJson(original)
You can also use something like protobuf3-to-dict