sriniiyer / concode Goto Github PK

I've found that many of the code snippets in ConCode reference methods that aren't defined in the memberFunctions field. An example is line 13 of test_shuffled.json, printed below. The code references a function called search(), but it's not defined in the metadata. I'm assuming that what happened here is that the file this was taken from statically imported the (static) search() method from some other class, but I just wanted to check that this is indeed why this happens and that you're aware of this feature of the dataset.

{
  "memberVariables": {
    "middle": "TSTNode",
    "data": "char",
    "left": "TSTNode",
    "root": "TSTNode",
    "right": "TSTNode",
    "al": "ArrayList<String>",
    "is_End_Of_String": "boolean"
  },
  "code": [
    "boolean",
    "function",
    "(",
    "String",
    "word",
    ")",
    "{",
    "return",
    "search",
    "(",
    "root",
    ",",
    "word",
    ".",
    "toCharArray",
    "(",
    ")",
    ",",
    "0",
    ")",
    ";",
    "}"
  ],
  "memberFunctions": {
    "traverse": [
      [
        "void",
        "TSTNode key",
        "String str"
      ]
    ],
    "makeEmpty": [
      [
        "void"
      ]
    ],
    "isEmpty": [
      [
        "boolean"
      ]
    ],
    "insert": [
      [
        "void",
        "String word"
      ],
      [
        "TSTNode",
        "TSTNode key",
        "char[] word",
        "int pos"
      ]
    ],
    "toString": [
      [
        "String"
      ]
    ],
    "main": [
      [
        "void",
        "String[] args"
      ]
    ],
    "delete": [
      [
        "void",
        "String word"
      ],
      [
        "void",
        "TSTNode key",
        "char[] word",
        "int pos"
      ]
    ]
  },
  "repo": "repo_24058027",
  "className": "TernarySearchTree",
  "renamed": [
    "boolean",
    "function",
    "(",
    "String",
    "arg0",
    ")",
    "{",
    "return",
    "search",
    "(",
    "root",
    ",",
    "arg0",
    ".",
    "toCharArray",
    "(",
    ")",
    ",",
    "0",
    ")",
    ";",
    "}"
  ],
  "nl": "function to search for a word",
  "nlToks": [
    "function",
    "to",
    "search",
    "for",
    "a",
    "word"
  ]
}

Hi Srini,
Do you have more than 100k instances? I found, in your google drive, the data is already truncated to 100k. Looks like CONCODE is a well-established dataset for code gen task. I am wondering if I can explore more CONCODE data. Btw, how the research going?
Thanks,

Hello,

I want to bypass all training process, to test a zero-shot to CONCODE model in some personal dataset.
Where Can I find CONCODE pretrained model ?

Thank you in advance.

Best Regards.

Hello,

While trying to execute the prediction task, I have the following Error :
RuntimeError: Only Tensors created explicitly by the user (graph leaves) support the deepcopy protocol at the moment

The error is caused by this line of code : "predictions.append(model.predict(batch, opt, None))" (line : 57 / translate.py)

Did you get the same error in some steps?
How could I fix this issue?

Thanks in advance,
Best Regards.

Hi,
Thanks for sharing your code. Could you please associate a license to it? Without any license, nobody has the right to reuse your code, even for other research.
If you don't know which license to chose, I would suggest MIT which gives the most liberty to people or GPL which ensures the copies of your code will stay free.
Best regards,

Hey @sriniiyer,
I am trying to augment the dataset with custom code snippets.
how can I convert code snippets which contain context and docstring as NL queries into concode dataset format.

Hello, I am very interested in your work in this article. I would like to know what hardware configuration is required to train this model?

hello,the link about Downloading data from Google drive is unusable.Can you provide the latest data set download link?Thanks.

Hi，Is anyone still maintaining this project? The data link in the manual is invalid.
I am quite interest in this project.I have read the related paper and studied the code for a few days. If anyone have the dataset please drop me a message