search

stellargraph / stellargraph

StellarGraph - Machine Learning on Graphs
https://stellargraph.readthedocs.io/
Apache License 2.0
2.94k stars 430 forks source link

Full batch models fail to save with "If specifying TensorSpec names for nested structures, either zero or all names have to be specified" #1251

Open huonw opened 4 years ago

huonw commented 4 years ago

Description

As reported in https://community.stellargraph.io/t/valueerror-if-specifying-tensorspec-names-for-nested-structures-either-zero-or-all-names-have-to-be-specified/97, saving some models in sparse mode fails with ValueError('If specifying TensorSpec names for nested structures, either zero or all names have to be specified.',). This is due to a bug in Tensorflow https://github.com/tensorflow/tensorflow/issues/38465 related to sparse tensors being passed between Keras layers. This bug is fixed in TensorFlow nightly (tf-nightly package) and the 2.3.0-rc0 release candidate.

Affected models (with sparse=True in their generators):

Work-around

Once the 2.3 release is out, either of these should be able to be replaced with pip install tensorflow==2.3.0 (the nightly version may need an explicit pip uninstall tf-nightly tf-estimator-nightly tb-nightly).

Original bug:

Describe the bug

As reported in https://community.stellargraph.io/t/valueerror-if-specifying-tensorspec-names-for-nested-structures-either-zero-or-all-names-have-to-be-specified/97, saving GAT or any other full batch model seems to fail.

To Reproduce

import stellargraph as sg
import tensorflow as tf

def test(x):
    print(f"#### {type(x).__name__}")
    print("```")
    model = tf.keras.Model(*x.in_out_tensors())
    try:
        tf.keras.models.save_model(model, "/tmp")
    except Exception as e:
        print(f"tf.keras.models.save_model: {e!r}")

    try:
        tf.saved_model.save(model, "/tmp")
    except Exception as e:
        print(f"tf.saved_model.save: {e!r}")

    try:
        model.save("/tmp")
    except Exception as e:
        print(f"model.save: {e!r}")
    print("```")

G, _ = sg.datasets.Cora().load()

fb_gen = sg.mapper.FullBatchNodeGenerator(G)
test(sg.layer.GAT([4], fb_gen))

test(sg.layer.GCN([4], fb_gen))

test(sg.layer.PPNP([4], fb_gen, ["relu"]))
test(sg.layer.APPNP([4], fb_gen, ["relu"]))

gs_gen = sg.mapper.GraphSAGENodeGenerator(G, 5, [2])
test(sg.layer.GraphSAGE([4], gs_gen))

hs_gen = sg.mapper.HinSAGENodeGenerator(G, 5, [2], head_node_type="paper")
test(sg.layer.HinSAGE([4], hs_gen))

cg_gen = sg.mapper.ClusterNodeGenerator(G)
test(sg.layer.ClusterGCN([4], ["relu"], cg_gen))

av_gen = sg.mapper.Attri2VecNodeGenerator(G, 10)
test(sg.layer.Attri2Vec([4], av_gen))

rg_gen = sg.mapper.RelationalFullBatchNodeGenerator(G)
test(sg.layer.RGCN([4], rg_gen))

kg_gen = sg.mapper.KGTripleGenerator(G, 10)
test(sg.layer.ComplEx(kg_gen, 10))
test(sg.layer.DistMult(kg_gen, 10))

pg_gen = sg.mapper.PaddedGraphGenerator([G])
test(sg.layer.GCNSupervisedGraphClassification([4], ["relu"], pg_gen))

test(sg.layer.DeepGraphInfomax(sg.layer.GCN([4], fb_gen)))

test(sg.layer.DeepGraphInfomax(sg.layer.GraphSAGE([4], gs_gen)))

Observed behavior

The full batch methods (GAT, GCN, PPNP, APPNP and DeepGraphInfomax using any of these) fails with:

ValueError: If specifying TensorSpec names for nested structures, either zero or all names have to be specified.

(RGCN fails too, but that's a separate issue: #1252.)

Full stack trace ``` --------------------------------------------------------------------------- ValueError Traceback (most recent call last) in 27 28 fb_gen = sg.mapper.FullBatchNodeGenerator(G) ---> 29 test(sg.layer.GAT([4], fb_gen)) 30 31 test(sg.layer.GCN([4], fb_gen)) in test(x) 7 model = tf.keras.Model(*x.in_out_tensors()) 8 try: ----> 9 tf.keras.models.save_model(model, "/tmp") 10 except Exception as e: 11 print(f"tf.keras.models.save_model: {e!r}") ~/.pyenv/versions/sg/lib/python3.6/site-packages/tensorflow_core/python/keras/saving/save.py in save_model(model, filepath, overwrite, include_optimizer, save_format, signatures, options) 113 else: 114 saved_model_save.save(model, filepath, overwrite, include_optimizer, --> 115 signatures, options) 116 117 ~/.pyenv/versions/sg/lib/python3.6/site-packages/tensorflow_core/python/keras/saving/saved_model/save.py in save(model, filepath, overwrite, include_optimizer, signatures, options) 76 # we use the default replica context here. 77 with distribution_strategy_context._get_default_replica_context(): # pylint: disable=protected-access ---> 78 save_lib.save(model, filepath, signatures, options) 79 80 if not include_optimizer: ~/.pyenv/versions/sg/lib/python3.6/site-packages/tensorflow_core/python/saved_model/save.py in save(obj, export_dir, signatures, options) 897 # Note we run this twice since, while constructing the view the first time 898 # there can be side effects of creating variables. --> 899 _ = _SaveableView(checkpoint_graph_view) 900 saveable_view = _SaveableView(checkpoint_graph_view) 901 ~/.pyenv/versions/sg/lib/python3.6/site-packages/tensorflow_core/python/saved_model/save.py in __init__(self, checkpoint_view) 185 # variables on first run. 186 concrete_functions = ( --> 187 function._list_all_concrete_functions_for_serialization()) # pylint: disable=protected-access 188 else: 189 concrete_functions = [function] ~/.pyenv/versions/sg/lib/python3.6/site-packages/tensorflow_core/python/eager/def_function.py in _list_all_concrete_functions_for_serialization(self) 825 concrete_functions = [] 826 for args, kwargs in seen_signatures: --> 827 concrete_functions.append(self.get_concrete_function(*args, **kwargs)) 828 return concrete_functions 829 ~/.pyenv/versions/sg/lib/python3.6/site-packages/tensorflow_core/python/keras/saving/saved_model/save_impl.py in get_concrete_function(self, *args, **kwargs) 554 def get_concrete_function(self, *args, **kwargs): 555 if not self.call_collection.tracing: --> 556 self.call_collection.add_trace(*args, **kwargs) 557 return super(LayerCall, self).get_concrete_function(*args, **kwargs) 558 ~/.pyenv/versions/sg/lib/python3.6/site-packages/tensorflow_core/python/keras/saving/saved_model/save_impl.py in add_trace(self, *args, **kwargs) 426 fn.get_concrete_function(*args, **kwargs) 427 --> 428 trace_with_training(True) 429 trace_with_training(False) 430 else: ~/.pyenv/versions/sg/lib/python3.6/site-packages/tensorflow_core/python/keras/saving/saved_model/save_impl.py in trace_with_training(value, fn) 424 utils.set_training_arg(value, self._training_arg_index, args, kwargs) 425 with K.learning_phase_scope(value): --> 426 fn.get_concrete_function(*args, **kwargs) 427 428 trace_with_training(True) ~/.pyenv/versions/sg/lib/python3.6/site-packages/tensorflow_core/python/keras/saving/saved_model/save_impl.py in get_concrete_function(self, *args, **kwargs) 555 if not self.call_collection.tracing: 556 self.call_collection.add_trace(*args, **kwargs) --> 557 return super(LayerCall, self).get_concrete_function(*args, **kwargs) 558 559 ~/.pyenv/versions/sg/lib/python3.6/site-packages/tensorflow_core/python/eager/def_function.py in get_concrete_function(self, *args, **kwargs) 917 # In this case we have not created variables on the first call. So we can 918 # run the first trace but we should fail if variables are created. --> 919 concrete = self._stateful_fn.get_concrete_function(*args, **kwargs) 920 if self._created_variables: 921 raise ValueError("Creating variables on a non-first call to a function" ~/.pyenv/versions/sg/lib/python3.6/site-packages/tensorflow_core/python/eager/function.py in get_concrete_function(self, *args, **kwargs) 2432 args, kwargs = None, None 2433 with self._lock: -> 2434 graph_function, args, kwargs = self._maybe_define_function(args, kwargs) 2435 if self.input_signature: 2436 args = self.input_signature ~/.pyenv/versions/sg/lib/python3.6/site-packages/tensorflow_core/python/eager/function.py in _maybe_define_function(self, args, kwargs) 2701 2702 self._function_cache.missed.add(call_context_key) -> 2703 graph_function = self._create_graph_function(args, kwargs) 2704 self._function_cache.primary[cache_key] = graph_function 2705 return graph_function, args, kwargs ~/.pyenv/versions/sg/lib/python3.6/site-packages/tensorflow_core/python/eager/function.py in _create_graph_function(self, args, kwargs, override_flat_arg_shapes) 2591 arg_names=arg_names, 2592 override_flat_arg_shapes=override_flat_arg_shapes, -> 2593 capture_by_value=self._capture_by_value), 2594 self._function_attributes, 2595 # Tell the ConcreteFunction to clean up its graph once it goes out of ~/.pyenv/versions/sg/lib/python3.6/site-packages/tensorflow_core/python/framework/func_graph.py in func_graph_from_py_func(name, python_func, args, kwargs, signature, func_graph, autograph, autograph_options, add_control_dependencies, arg_names, op_return_value, collections, capture_by_value, override_flat_arg_shapes) 896 kwarg_shapes = None 897 func_args = _get_defun_inputs_from_args( --> 898 args, arg_names, flat_shapes=arg_shapes) 899 func_kwargs = _get_defun_inputs_from_kwargs( 900 kwargs, flat_shapes=kwarg_shapes) ~/.pyenv/versions/sg/lib/python3.6/site-packages/tensorflow_core/python/framework/func_graph.py in _get_defun_inputs_from_args(args, names, flat_shapes) 1125 """Maps Python function positional args to graph-construction inputs.""" 1126 return _get_defun_inputs( -> 1127 args, names, structure=args, flat_shapes=flat_shapes) 1128 1129 ~/.pyenv/versions/sg/lib/python3.6/site-packages/tensorflow_core/python/framework/func_graph.py in _get_defun_inputs(args, names, structure, flat_shapes) 1189 specified_names = [arg.name for arg in tensor_specs if arg.name] 1190 if specified_names and len(specified_names) < len(tensor_specs): -> 1191 raise ValueError("If specifying TensorSpec names for nested structures, " 1192 "either zero or all names have to be specified.") 1193 ValueError: If specifying TensorSpec names for nested structures, either zero or all names have to be specified. ```
Full output #### GAT ``` tf.keras.models.save_model: ValueError('If specifying TensorSpec names for nested structures, either zero or all names have to be specified.',) tf.saved_model.save: ValueError('If specifying TensorSpec names for nested structures, either zero or all names have to be specified.',) model.save: ValueError('If specifying TensorSpec names for nested structures, either zero or all names have to be specified.',) ``` #### GCN ``` tf.keras.models.save_model: ValueError('If specifying TensorSpec names for nested structures, either zero or all names have to be specified.',) tf.saved_model.save: ValueError('If specifying TensorSpec names for nested structures, either zero or all names have to be specified.',) model.save: ValueError('If specifying TensorSpec names for nested structures, either zero or all names have to be specified.',) ``` #### PPNP ``` tf.keras.models.save_model: ValueError('If specifying TensorSpec names for nested structures, either zero or all names have to be specified.',) tf.saved_model.save: ValueError('If specifying TensorSpec names for nested structures, either zero or all names have to be specified.',) model.save: ValueError('If specifying TensorSpec names for nested structures, either zero or all names have to be specified.',) ``` #### APPNP ``` tf.keras.models.save_model: ValueError('If specifying TensorSpec names for nested structures, either zero or all names have to be specified.',) tf.saved_model.save: ValueError('If specifying TensorSpec names for nested structures, either zero or all names have to be specified.',) model.save: ValueError('If specifying TensorSpec names for nested structures, either zero or all names have to be specified.',) ``` #### GraphSAGE ``` INFO:tensorflow:Assets written to: /tmp/assets INFO:tensorflow:Assets written to: /tmp/assets INFO:tensorflow:Assets written to: /tmp/assets ``` #### HinSAGE ``` INFO:tensorflow:Assets written to: /tmp/assets INFO:tensorflow:Assets written to: /tmp/assets INFO:tensorflow:Assets written to: /tmp/assets ``` Number of clusters 1 0 cluster has size 2708 #### ClusterGCN ``` INFO:tensorflow:Assets written to: /tmp/assets INFO:tensorflow:Assets written to: /tmp/assets INFO:tensorflow:Assets written to: /tmp/assets ``` #### Attri2Vec ``` INFO:tensorflow:Assets written to: /tmp/assets INFO:tensorflow:Assets written to: /tmp/assets INFO:tensorflow:Assets written to: /tmp/assets ``` #### RGCN ``` tf.keras.models.save_model: AttributeError("'NoneType' object has no attribute 'replace'",) tf.saved_model.save: AttributeError("'NoneType' object has no attribute 'replace'",) model.save: AttributeError("'NoneType' object has no attribute 'replace'",) ``` #### ComplEx ``` INFO:tensorflow:Assets written to: /tmp/assets INFO:tensorflow:Assets written to: /tmp/assets INFO:tensorflow:Assets written to: /tmp/assets ``` #### DistMult ``` INFO:tensorflow:Assets written to: /tmp/assets INFO:tensorflow:Assets written to: /tmp/assets INFO:tensorflow:Assets written to: /tmp/assets ``` #### GCNSupervisedGraphClassification ``` INFO:tensorflow:Assets written to: /tmp/assets INFO:tensorflow:Assets written to: /tmp/assets INFO:tensorflow:Assets written to: /tmp/assets ``` #### DeepGraphInfomax ``` tf.keras.models.save_model: ValueError('If specifying TensorSpec names for nested structures, either zero or all names have to be specified.',) tf.saved_model.save: ValueError('If specifying TensorSpec names for nested structures, either zero or all names have to be specified.',) model.save: ValueError('If specifying TensorSpec names for nested structures, either zero or all names have to be specified.',) ``` #### DeepGraphInfomax ``` INFO:tensorflow:Assets written to: /tmp/assets INFO:tensorflow:Assets written to: /tmp/assets INFO:tensorflow:Assets written to: /tmp/assets ```

Expected behavior

Every model should support saving.

Environment

Operating system: Darwin-18.6.0-x86_64-i386-64bit Python version:

3.6.9 (default, Jul 10 2019, 12:25:55) 
[GCC 4.2.1 Compatible Apple LLVM 10.0.1 (clang-1001.0.46.4)]

Package versions:

``` absl-py==0.8.0 ansiwrap==0.8.4 appdirs==1.4.3 appnope==0.1.0 astor==0.8.0 atomicwrites==1.3.0 attrs==19.3.0 backcall==0.1.0 black==19.10b0 bleach==3.1.0 boto==2.49.0 boto3==1.9.230 botocore==1.12.230 cachetools==4.0.0 certifi==2019.9.11 chardet==3.0.4 Click==7.0 coverage==4.5.4 coveralls==1.8.2 cycler==0.10.0 decorator==4.4.0 defusedxml==0.6.0 docopt==0.6.2 docutils==0.15.2 entrypoints==0.3 gast==0.2.2 gensim==3.8.0 google-auth==1.10.0 google-auth-oauthlib==0.4.1 google-pasta==0.1.7 gprof2dot==2019.11.30 grpcio==1.23.0 h5py==2.10.0 idna==2.8 importlib-metadata==0.23 ipykernel==5.1.3 ipython==7.9.0 ipython-genutils==0.2.0 ipywidgets==7.5.1 isodate==0.6.0 jedi==0.15.1 Jinja2==2.10.3 jmespath==0.9.4 joblib==0.13.2 jsonschema==3.2.0 jupyter==1.0.0 jupyter-client==5.3.4 jupyter-console==6.0.0 jupyter-core==4.6.1 Keras==2.2.5 Keras-Applications==1.0.8 Keras-Preprocessing==1.1.0 kiwisolver==1.1.0 llvmlite==0.30.0 Mako==1.1.0 Markdown==3.1.1 MarkupSafe==1.1.1 matplotlib==3.1.1 mistune==0.8.4 more-itertools==7.2.0 mplleaflet==0.0.5 mypy==0.750 mypy-extensions==0.4.3 nbclient==0.1.0 nbconvert==5.6.1 nbformat==4.4.0 networkx==2.3 notebook==6.0.2 numba==0.46.0 numpy==1.17.2 oauthlib==3.1.0 opt-einsum==3.1.0 packaging==19.2 pandas==0.25.1 pandocfilters==1.4.2 papermill==1.2.1 parso==0.5.1 pathspec==0.6.0 pdoc3==0.7.2 pexpect==4.7.0 pickleshare==0.7.5 pluggy==0.13.1 prometheus-client==0.7.1 prompt-toolkit==2.0.10 protobuf==3.9.1 ptyprocess==0.6.0 py==1.8.0 py-cpuinfo==5.0.0 py4j==0.10.7 pyasn1==0.4.8 pyasn1-modules==0.2.7 pydot==1.4.1 Pygments==2.4.2 Pympler==0.8 pyparsing==2.4.2 pyrsistent==0.15.6 pyspark==2.4.4 pyspark-stubs==2.4.0.post6 pytest==5.3.1 pytest-benchmark==3.2.2 pytest-cov==2.8.1 pytest-profiling==1.7.0 pytest-repeat==0.8.0 python-dateutil==2.8.0 pytz==2019.2 PyYAML==5.1.2 pyzmq==18.1.1 qtconsole==4.6.0 rdflib==4.2.2 regex==2019.12.9 requests==2.22.0 requests-oauthlib==1.3.0 rsa==4.0 s3transfer==0.2.1 scikit-learn==0.21.3 scipy==1.4.1 seaborn==0.10.0 Send2Trash==1.5.0 six==1.12.0 smart-open==1.8.4 -e git+git@github.com:stellargraph/stellargraph.git@c337c18fa391169234ccc0f46b699493d96748a7#egg=stellargraph tenacity==6.0.0 tensorboard==2.1.1 tensorflow==2.1.0 tensorflow-estimator==2.1.0 termcolor==1.1.0 terminado==0.8.3 testpath==0.4.4 textwrap3==0.9.2 toml==0.10.0 tornado==6.0.3 tqdm==4.42.1 traitlets==4.3.3 treon==0.1.3 typed-ast==1.4.0 typing-extensions==3.7.4.1 urllib3==1.25.3 wcwidth==0.1.7 webencodings==0.5.1 Werkzeug==0.15.6 widgetsnbextension==3.5.1 wrapt==1.11.2 zipp==0.6.0 ```

Additional context

N/A

huonw commented 4 years ago

https://community.stellargraph.io/t/valueerror-if-specifying-tensorspec-names-for-nested-structures-either-zero-or-all-names-have-to-be-specified/97/3?u=huon notes that tf.compat.v1.keras.experimental.export_saved_model and tf.compat.v1.keras.experimental.load_from_saved_model work.

jakubvedral commented 4 years ago

Hi, For me in TensorFlow 2.2 event the workaround with tf.compat.v1.keras.experimental.export_saved_model and tf.compat.v1.keras.experimental.load_from_saved_model does not work :(

Still the same error as you described above.

huonw commented 4 years ago

I'm investigating this now.

Main finding so far: this seems to be specifically associated with sparse GraphConvolution, so a work-around for sufficiently small graphs is to pass sparse=True to the original FullBatchNodeGenerator.

huonw commented 4 years ago

This seems to be some problem with layers that take sparse tensors as input, even if they're not being used, e.g.:

Reduced form (via creduce and manual editing):

import tensorflow as tf

class SqueezedSparseConversion(tf.keras.layers.Layer):
    def call(self, inputs):
        return tf.SparseTensor([(0, 1)], [0.1], (3, 3))

class GraphConvolution(tf.keras.layers.Layer):
    def call(self, inputs):
        return inputs[0]

x_t = tf.keras.Input(0)
sp = SqueezedSparseConversion()(x_t)
out = GraphConvolution()([x_t, sp])

m = tf.keras.Model([x_t], out)
m.summary()
m.save("")
Full output ``` Model: "model" __________________________________________________________________________________________________ Layer (type) Output Shape Param # Connected to ================================================================================================== input_1 (InputLayer) [(None, 0)] 0 __________________________________________________________________________________________________ squeezed_sparse_conversion (Squ (3, 3) 0 input_1[0][0] __________________________________________________________________________________________________ graph_convolution (GraphConvolu (None, 0) 0 input_1[0][0] squeezed_sparse_conversion[0][0] ================================================================================================== Total params: 0 Trainable params: 0 Non-trainable params: 0 __________________________________________________________________________________________________ 2020-06-10 15:18:41.594860: I tensorflow/core/platform/cpu_feature_guard.cc:143] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA 2020-06-10 15:18:41.608329: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x7fbb46965a70 initialized for platform Host (this does not guarantee that XLA will be used). Devices: 2020-06-10 15:18:41.608343: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (0): Host, Default Version 2020-06-10 15:18:41.673648: W tensorflow/python/util/util.cc:329] Sets are not currently considered sequences, but this may change in the future, so consider avoiding using them. Traceback (most recent call last): File "issue-1251.py", line 17, in m.save("") File ".../site-packages/tensorflow/python/keras/engine/network.py", line 1052, in save signatures, options) File ".../site-packages/tensorflow/python/keras/saving/save.py", line 138, in save_model signatures, options) File ".../site-packages/tensorflow/python/keras/saving/saved_model/save.py", line 78, in save save_lib.save(model, filepath, signatures, options) File ".../site-packages/tensorflow/python/saved_model/save.py", line 951, in save obj, export_dir, signatures, options, meta_graph_def) File ".../site-packages/tensorflow/python/saved_model/save.py", line 1022, in _build_meta_graph _ = _SaveableView(checkpoint_graph_view) File ".../site-packages/tensorflow/python/saved_model/save.py", line 204, in __init__ function._list_all_concrete_functions_for_serialization()) # pylint: disable=protected-access File ".../site-packages/tensorflow/python/eager/def_function.py", line 841, in _list_all_concrete_functions_for_serialization concrete_functions.append(self.get_concrete_function(*args, **kwargs)) File ".../site-packages/tensorflow/python/keras/saving/saved_model/save_impl.py", line 546, in get_concrete_function self.call_collection.add_trace(*args, **kwargs) File ".../site-packages/tensorflow/python/keras/saving/saved_model/save_impl.py", line 421, in add_trace fn.get_concrete_function(*args, **kwargs) File ".../site-packages/tensorflow/python/keras/saving/saved_model/save_impl.py", line 547, in get_concrete_function return super(LayerCall, self).get_concrete_function(*args, **kwargs) File ".../site-packages/tensorflow/python/eager/def_function.py", line 959, in get_concrete_function concrete = self._get_concrete_function_garbage_collected(*args, **kwargs) File ".../site-packages/tensorflow/python/eager/def_function.py", line 877, in _get_concrete_function_garbage_collected *args, **kwargs) File ".../site-packages/tensorflow/python/eager/function.py", line 2496, in _get_concrete_function_garbage_collected graph_function, args, kwargs = self._maybe_define_function(args, kwargs) File ".../site-packages/tensorflow/python/eager/function.py", line 2777, in _maybe_define_function graph_function = self._create_graph_function(args, kwargs) File ".../site-packages/tensorflow/python/eager/function.py", line 2667, in _create_graph_function capture_by_value=self._capture_by_value), File ".../site-packages/tensorflow/python/framework/func_graph.py", line 898, in func_graph_from_py_func args, arg_names, flat_shapes=arg_shapes) File ".../site-packages/tensorflow/python/framework/func_graph.py", line 1132, in _get_defun_inputs_from_args args, names, structure=args, flat_shapes=flat_shapes) File ".../site-packages/tensorflow/python/framework/func_graph.py", line 1196, in _get_defun_inputs raise ValueError("If specifying TensorSpec names for nested structures, " ValueError: If specifying TensorSpec names for nested structures, either zero or all names have to be specified. ```
Package versions ``` absl-py==0.7.1 alabaster==0.7.12 appnope==0.1.0 astor==0.8.0 astroid==2.3.3 astunparse==1.6.3 attrs==19.3.0 avro==1.8.2 avro-python3==1.8.2 awscli==1.16.198 Babel==2.8.0 backcall==0.1.0 bleach==3.1.0 boto==2.49.0 boto3==1.9.208 botocore==1.12.208 cachetools==4.1.0 certifi==2019.6.16 chardet==3.0.4 colorama==0.3.9 cycler==0.10.0 dataclasses==0.6 decorator==4.4.0 defusedxml==0.6.0 Deprecated==1.2.7 dgl==0.3 docutils==0.14 entrypoints==0.3 gast==0.3.3 gensim==3.8.0 google-auth==1.14.3 google-auth-oauthlib==0.4.1 google-pasta==0.2.0 grpcio==1.25.0 gviz-api==1.9.0 h5py==2.10.0 idna==2.8 imagesize==1.2.0 importlib-metadata==1.3.0 ipykernel==5.1.3 ipython==7.11.1 ipython-genutils==0.2.0 isort==4.3.21 jedi==0.15.2 Jinja2==2.10.3 jmespath==0.9.4 joblib==0.13.2 json5==0.8.5 jsonschema==3.2.0 jupyter-client==5.3.4 jupyter-core==4.6.1 jupyterlab==1.2.4 jupyterlab-server==1.0.6 Keras==2.2.4 Keras-Applications==1.0.8 Keras-Preprocessing==1.1.0 kiwisolver==1.1.0 lazy-object-proxy==1.4.3 Mako==1.0.14 Markdown==3.1.1 MarkupSafe==1.1.1 matplotlib==3.1.1 mccabe==0.6.1 mistune==0.8.4 more-itertools==8.0.2 mpmath==1.1.0 nbconvert==5.6.1 nbformat==4.4.0 nbsphinx==0.6.1 networkx==2.3 notebook==6.0.2 numexpr==2.7.1 numpy==1.18.4 oauthlib==3.1.0 opt-einsum==3.2.1 packaging==20.3 pandas==1.0.3 pandocfilters==1.4.2 parso==0.5.2 patsy==0.5.1 pdoc3==0.6.3 pexpect==4.7.0 pickleshare==0.7.5 prometheus-client==0.7.1 prompt-toolkit==3.0.2 protobuf==3.9.0 ptyprocess==0.6.0 py-spy==0.3.2 py4j==0.10.7 pyasn1==0.4.8 pyasn1-modules==0.2.8 pydantic==0.32.1 PyGithub==1.45 Pygments==2.5.2 PyJWT==1.7.1 pylint==2.4.4 Pympler==0.8 pyparsing==2.4.2 pyrsistent==0.15.6 pyspark==2.4.3 python-dateutil==2.8.0 python-igraph==0.8.0 pytz==2019.1 PyYAML==5.1 pyzmq==18.1.1 requests==2.22.0 requests-oauthlib==1.3.0 rsa==3.4.2 s3transfer==0.2.1 scikit-learn==0.21.3 scipy==1.4.1 Send2Trash==1.5.0 sgplatform==0.5.2a0 six==1.12.0 smart-open==1.8.4 snowballstemmer==2.0.0 Sphinx==3.0.3 sphinxcontrib-applehelp==1.0.2 sphinxcontrib-devhelp==1.0.2 sphinxcontrib-htmlhelp==1.0.3 sphinxcontrib-jsmath==1.0.1 sphinxcontrib-qthelp==1.0.3 sphinxcontrib-serializinghtml==1.1.4 statsmodels==0.10.2 tabulate==0.8.3 tensorboard==2.2.1 tensorboard-plugin-profile==2.2.0 tensorboard-plugin-wit==1.6.0.post3 tensorflow==2.2.0 tensorflow-estimator==2.2.0 termcolor==1.1.0 terminado==0.8.3 testpath==0.4.4 texttable==1.6.2 torch==1.1.0.post2 tornado==6.0.3 traitlets==4.3.3 typed-ast==1.4.0 urllib3==1.25.3 virtualenv==16.7.9 wcwidth==0.1.8 webencodings==0.5.1 Werkzeug==0.15.5 wrapt==1.11.2 xarray==0.15.1 zipp==0.6.0 ```
huonw commented 4 years ago

I filed https://github.com/tensorflow/tensorflow/issues/40373. I think we can work around this by not doing the SqueezedSparseConversion as a layer but instead doing it within each GraphConvolution. This does require passing around the indices, data and shape of the sparse matrix independently. This may have some downsides:

huonw commented 4 years ago

I filed tensorflow/tensorflow#40373.

That issue was closed as a duplicate of https://github.com/tensorflow/tensorflow/issues/38465.

It seems that this is now fixed in tf-nightly (tested in https://github.com/stellargraph/stellargraph/pull/1689). That is, one can work around this problem by installing a version of https://pypi.org/project/tf-nightly/ .

As such, I think I'll put this on the back burner for now because we have other high priority tasks; but, please let me know if it's particularly important to you (and upgrading to tf-nightly temporarily isn't feasible).

huonw commented 4 years ago

TensorFlow released 2.3.0-rc0 recently, which includes the fix for this issue. https://github.com/stellargraph/stellargraph/pull/1742 validates that (the equivalent of) pip install tensorflow==2.3.0-rc0 works.