Skip to content

Question about text-code evaluation #183

Description

@Zhao0278

I trained the CodeBert model using CodeSearchNet on nl-code-search-webquery and fine-tuned it on Cosqa. But I have no access to the test_webquery.json file in the command from https://github.com/microsoft/CodeXGLUE/tree/main/Text-Code/NL-code-search-WebQuery#readme:

python code/run_classifier.py \
			--model_type roberta \
			--do_predict \
			--test_file test_webquery.json \
			--max_seq_length 200 \
			--per_gpu_eval_batch_size 2 \
			--data_dir ./data \
			--output_dir ./model_cosqa_continue_training/checkpoint-best-aver/ \
			--encoder_name_or_path microsoft/codebert-base \
			--pred_model_dir ./model_cosqa_continue_training/checkpoint-last/ \
			--prediction_file ./evaluator/webquery_predictions.txt 
			

I tried dev_webquery. json (from CodeSearchNet) and cosqa_dev. json. The validation results on cosqa_dev. json are similar to those in readme, but the accuracy of dev_webquery. json is very low. Additionally, the accuracy results in readme are inconsistent with those in the paper. Is the evaluation result in the paper obtained on CodeSearchNet or cosqa?
image
image

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions