❓ Questions and Help

This repo presents couple of nice examples for the GNN.

I am particularly interested about:

• Semantic Parsing
• NMT

Do you have the model architecture described somewhere as part of the tutorial or documentation? Alternately do you have a canonical architecture described somewhere for these Graph2Seq based models? Is the model same as the Graph2Seq: A Generalized Seq2Seq Model for Graph Inputs?

In the examples mentioned in the demo, we saw that graph construction was done while training the specific use case like classification. Is there a way, we can construct graph on whole of our contextual data and then use that graph for our downstream applications like classification, knowledge graph completion etc.?

How to train the Graph2Seq model in a multiple-GPU environment? As an example, there is an NMT example here: https://github.com/graph4ai/graph4nlp/tree/master/examples/pytorch/nmt

The model is built here: https://github.com/graph4ai/graph4nlp/blob/master/examples/pytorch/nmt/build_model.py

Could this be extended to be trained in a multiple GPU environment?

❓ Questions and Help

I want to use the graph2seq model that would encode an input source code AST as a graph and would decode to a target sequence.

So far I am building the graph model using the GraphData API to build source code AST as a graph. However, how would I feed the sequence data to the decoder?

I don’t find one complete example showing the whole flow. Can anyone point me in the right direction?

Hello everyone, I want to ask something that is related to https://github.com/graph4ai/graph4nlp/issues/316#issue-918894490. I cannot fully understand how I can pass the data to the model for inference. In particular, for the evaluation in tutorial Graph2Tree about jobs data, I see that the model retrieves the data from the test data-loader, which is generate from JobsDatasetForTree, which gets data from text file and preprocess them. This is different from my goals: infact, after training, I save the model and I load it, after that I want to do inference on unseen data; so, I want to pass to model a list of strings, preprocess these strings (according to the method that I used in the training for the graph construction, i.e. Dependency or Constituency or something else) and do inference on this data. Can you explain me how i can do that? Thanks

🐛 Bug

To Reproduce

Steps to reproduce the behavior:

1. Run python examples/pytorch/name_entity_recognition/main.py --graph_type dependency_graph --gpu 0 --init_hidden_size 400 --hidden_size 128 --lr 0.01 --batch_size 100 --gnn_type graphsage --direction_option undirected

2. Getting TypeError: Can't instantiate abstract class ConllDataset with abstract methods download

Expected behavior

Environment

• Graph4NLP Version (e.g., 0.4.1):
• Backend Library & Version (e.g., PyTorch 1.6.0): 1.8.1
• OS (e.g., Linux): Windows 10
• How you installed Graph4NLP (pip, source): source
• Build command you used (if compiling from source): python setup.py install
• Python version: 3.6.5
• CUDA/cuDNN version (if applicable): CPU
• GPU models and configuration (e.g. 2080Ti):
• Any other relevant information:

Additional context

❓ Questions and Help

How to add an edge with attribute value?

    def add_edge(self, src: int, tgt: int):
        """
        Add one edge to the graph.

        Parameters
        ----------
        src : int
            Source node index
        tgt : int
            Target node index

Currently, I can only add an edge between two nodes using their node index. But how to set the edge_type? I cant find any API like add_edge(self, src: int, tgt: int, edge_type: str).

Can anyone please provide a pointer or code snippet?

❓ Questions and Help

Hello, I changed the pretrained model to 'bert' but I got the following error:

File "/home/hadi/miniconda3/envs/ongpu/lib/python3.9/site-packages/transformers-4.12.5-py3.9.egg/transformers/file_utils.py", line 1818, in get_list_of_files
   model_info = HfApi(endpoint=HUGGINGFACE_CO_RESOLVE_ENDPOINT).model_info(
 File "/home/hadi/miniconda3/envs/ongpu/lib/python3.9/site-packages/huggingface_hub-0.1.2-py3.9.egg/huggingface_hub/hf_api.py", line 585, in model_info
   r.raise_for_status()
 File "/home/hadi/miniconda3/envs/ongpu/lib/python3.9/site-packages/requests/models.py", line 953, in raise_for_status
   raise HTTPError(http_error_msg, response=self)
requests.exceptions.HTTPError: 404 Client Error: Not Found for url: https://huggingface.co/api/models/None

It seems to be due to huggingface servers, but how can I fix it? If I want to download pretrained model manually, where should I put the file?

❓ Questions and Help

Hi, in the summarization example, I want to try other graph neural networks (replacing GCN with GraphSAGE, GAT, ...) How can I do it with minimal change to the original code?

❓ Questions and Help

Came across this fantastic library just a week back. This library is looking great 🥇

I will be mostly using the example for machine translation. Will also use the **text summarization at a later point.

I am currently using opennmt for machine translation and use the copy_mechanism as my source and target has same vocabulary. Now I will be moving to your library for this NMT task so that my input could leverage the graph structure.

At the same time, I need to use the copy_mechanism for OOV issue. My questions:

1. It appears to me, I just need to set the use_copy as true for the decoder_args and that should do for the NMT example. Am I missing something?

decoder_args:
  rnn_decoder_share:
    rnn_type: "lstm"
    input_size: 300
    hidden_size: 512
    rnn_emb_input_size: 300
    use_copy: true
    use_coverage: true
    graph_pooling_strategy: null
    attention_type: "sep_diff_encoder_type"
    fuse_strategy: "concatenate"
    dropout: 0.3

2. I am using the Graph2Seq model as per the NMT example. In the NMT example, you have used GCN. For the copy mechanism to be more effective does it has a relation with what GNN I use?

You have four GNN models - GCN, GAT, GGNN, GraphSage. Do you think copy mechanism would be more effective for a specific GNN? Or it does not matter?

3. Would enabling the copy mechanism lower the accuracy/BLEU score of the translation task?

❓ Questions and Help

I am running the NMT example on the same dataset with GNN variants:

• GCN
• GGNN
• GraphSage

While the execution runs with GCN, I get Out-of-Memory (OOM) for GGNN and GraphSage. Can anyone help me with this query?

Description

Checklist

Please feel free to remove inapplicable items for your PR.

• [ ] The PR title starts with [$CATEGORY] (such as [Doc], [Feature]])
• [ ] Changes are complete (i.e. I finished coding on this PR)
• [ ] All changes have test coverage
• [ ] Code is well-documented
• [ ] To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change
• [ ] Related issue is referred in this PR
• [ ] If the PR is for a new model/paper, I've updated the example index here.

Changes

Description

Add RGCN for text classification.

Checklist

Please feel free to remove inapplicable items for your PR.

• [x] The PR title starts with [$CATEGORY] (such as [Doc], [Feature]])
• [x] Changes are complete (i.e. I finished coding on this PR)
• [x] All changes have test coverage
• [x] Code is well-documented
• [x] To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change

❓ Questions and Help

I am trying to come up with a simple example of how to use the NodeEmbeddingBasedRefinedGraphConstruction class, but I am getting the following error. could you please let me know how to congif the class to make it work?

from graph4nlp.pytorch.modules.graph_construction.node_embedding_based_refined_graph_construction import NodeEmbeddingBasedRefinedGraphConstruction
raw_data = "James went to the corner-shop. And bought some eggs."
gl = NodeEmbeddingBasedRefinedGraphConstruction(alpha_fusion= 0.2, input_size=(len(raw_data)))
graphdata = gl.init_topology(raw_data, lower_case=True)
graph = gl.dynamic_topology(graphdata)

But here is the error that I get.

    152 def __repr__(self):
--> 153     return self._graph._get_batch_node_features()

File ~/.conda/envs/graph4nlp/lib/python3.10/site-packages/graph4nlp-0.5.5-py3.10.egg/graph4nlp/pytorch/data/data.py:930, in GraphData._get_batch_node_features(self, item)
    915 """
    916 Get the batched view of node feature tensors, i.e., tensors in (B, N, D) view
    917 
   (...)
    927     batch-view tensors, or just the specified tensor.
    928 """
    929 if not self._is_batch:
--> 930     raise Exception("Calling batch_node_features() method on a non-batch graph.")
    931 if item is None:
    932     batch_node_features = dict()

Exception: Calling batch_node_features() method on a non-batch graph.

Any working simple example is highly appreciated.

❓ Questions and Help

I am trying to run the following example from the documentation (https://graph4ai.github.io/graph4nlp/guide/construction/iegraphconstruction.html), but I am getting an attribute error. I was wondering what is the correct version of running "IEBasedGraphConstruction" to construct a graph?

raw_data = ('James is on shop. He buys eggs.')

nlp_parser = StanfordCoreNLP('http://localhost', port=9000, timeout=300000)

props_coref = {
                'annotators': 'tokenize, ssplit, pos, lemma, ner, parse, coref',
                "tokenize.options":
                    "splitHyphenated=true,normalizeParentheses=true,normalizeOtherBrackets=true",
                "tokenize.whitespace": False,
                'ssplit.isOneSentence': False,
                'outputFormat': 'json'
            }

props_openie = {
    'annotators': 'tokenize, ssplit, pos, ner, parse, openie',
    "tokenize.options":
        "splitHyphenated=true,normalizeParentheses=true,normalizeOtherBrackets=true",
    "tokenize.whitespace": False,
    'ssplit.isOneSentence': False,
    'outputFormat': 'json',
    "openie.triple.strict": "true"
}

processor_args = [props_coref, props_openie]

graphdata = IEBasedGraphConstruction.topology(raw_data, nlp_parser,
                                              processor_args=processor_args,
                                              merge_strategy=None,
                                              edge_strategy=None)
print(graphdata)

Error:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
Input In [8], in <cell line: 26>()
     14 props_openie = {
     15     'annotators': 'tokenize, ssplit, pos, ner, parse, openie',
     16     "tokenize.options":
   (...)
     21     "openie.triple.strict": "true"
     22 }
     24 processor_args = [props_coref, props_openie]
---> 26 graphdata = ConstituencyBasedGraphConstruction.topology(raw_data, nlp_parser,
     27                                               processor_args=processor_args,
     28                                               merge_strategy=None,
     29                                               edge_strategy=None)
     30 print(graphdata)

AttributeError: type object 'ConstituencyBasedGraphConstruction' has no attribute 'topology'

Thank you

Description

Checklist

Please feel free to remove inapplicable items for your PR.

• [x] The PR title starts with [$CATEGORY] (such as [Doc], [Feature]])
• [x] Changes are complete (i.e. I finished coding on this PR)
• [x] All changes have test coverage
• [x] Code is well-documented
• [x] To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change
• [x] Related issue is referred in this PR
• [ ] If the PR is for a new model/paper, I've updated the example index here.

Changes

Description

Checklist

Please feel free to remove inapplicable items for your PR.

• [ ] The PR title starts with [$CATEGORY] (such as [Doc], [Feature]])
• [ ] Changes are complete (i.e. I finished coding on this PR)
• [ ] All changes have test coverage
• [ ] Code is well-documented
• [ ] To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change
• [ ] Related issue is referred in this PR
• [ ] If the PR is for a new model/paper, I've updated the example index here.

Changes