removed original data

README.md

@@ -8,14 +8,6 @@ configs:
     path: "data/dev_instances.json"
   - split: test
     path: "data/test_instances.json"
-# - config_name: original
-#   data_files:
-#   - split: train
-#     path: "original_data/qasper-train-dev-v0.3/qasper-train-v0.3.json"
-#   - split: dev
-#     path: "original_data/qasper-train-dev-v0.3/qasper-dev-v0.3.json"
-#   - split: test
-#     path: "original_data/qasper-test-and-evaluator-v0.3/qasper-test-v0.3.json"
 ---
 
 # Preprocessed QASPER dataset

original_data/qasper-test-and-evaluator-v0.3/README-test.md

@@ -1,26 +0,0 @@
-Dataset of Information Seeking Questions and Answers Anchored in Research Papers: Test Set and Evaluator
---------------------------------------------------------------------------------------------------------                                      
-
-## Version: 0.3
-
-The tarball you found this file in should contain the test split of the Qasper dataset version 0.3 and the official evaluator script.
-
-Please make sure you access the test file only to evaluate your finalized model.
-
-## Images of tables and figures
-
-You can download them here: https://qasper-dataset.s3.us-west-2.amazonaws.com/test_figures_and_tables.tgz
-
-## Evaluation
-
-You can evaluate your model using the stand alone evaluator as follows:
-
-```
-python qasper_evaluator.py --predictions predictions.jsonl --gold qasper-test-v0.3.json [--text_evidence_only]
-```
-
-Run the following to understand the arguments
-
-```
-python qasper_evaluator.py -h
-```

original_data/qasper-test-and-evaluator-v0.3/qasper-test-v0.3.json

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:6e29ad410e6e39aa1936017fb965b30a20eb2e7751997f55b97c9d281aa884e5
-size 18078957

original_data/qasper-test-and-evaluator-v0.3/qasper_evaluator.py

@@ -1,167 +0,0 @@
-"""
-Official script for evaluating models built for the Qasper dataset. The script
-outputs Answer F1 and Evidence F1 reported in the paper.
-"""
-from collections import Counter
-import argparse
-import string
-import re
-import json
-
-
-def normalize_answer(s):
-    """
-    Taken from the official evaluation script for v1.1 of the SQuAD dataset.
-    Lower text and remove punctuation, articles and extra whitespace.
-    """
-
-    def remove_articles(text):
-        return re.sub(r"\b(a|an|the)\b", " ", text)
-
-    def white_space_fix(text):
-        return " ".join(text.split())
-
-    def remove_punc(text):
-        exclude = set(string.punctuation)
-        return "".join(ch for ch in text if ch not in exclude)
-
-    def lower(text):
-        return text.lower()
-
-    return white_space_fix(remove_articles(remove_punc(lower(s))))
-
-
-def token_f1_score(prediction, ground_truth):
-    """
-    Taken from the official evaluation script for v1.1 of the SQuAD dataset.
-    """
-    prediction_tokens = normalize_answer(prediction).split()
-    ground_truth_tokens = normalize_answer(ground_truth).split()
-    common = Counter(prediction_tokens) & Counter(ground_truth_tokens)
-    num_same = sum(common.values())
-    if num_same == 0:
-        return 0
-    precision = 1.0 * num_same / len(prediction_tokens)
-    recall = 1.0 * num_same / len(ground_truth_tokens)
-    f1 = (2 * precision * recall) / (precision + recall)
-    return f1
-
-
-def paragraph_f1_score(prediction, ground_truth):
-    if not ground_truth and not prediction:
-        # The question is unanswerable and the prediction is empty.
-        return 1.0
-    num_same = len(set(ground_truth).intersection(set(prediction)))
-    if num_same == 0:
-        return 0.0
-    precision = num_same / len(prediction)
-    recall = num_same / len(ground_truth) 
-    f1 = (2 * precision * recall) / (precision + recall)
-    return f1
-
-
-def get_answers_and_evidence(data, text_evidence_only):
-    answers_and_evidence = {}
-    for paper_data in data.values():
-        for qa_info in paper_data["qas"]:
-            question_id = qa_info["question_id"]
-            references = []
-            for annotation_info in qa_info["answers"]:
-                answer_info = annotation_info["answer"]
-                if answer_info["unanswerable"]:
-                    references.append({"answer": "Unanswerable", "evidence": [], "type": "none"})
-                else:
-                    if answer_info["extractive_spans"]:
-                        answer = ", ".join(answer_info["extractive_spans"])
-                        answer_type = "extractive"
-                    elif answer_info["free_form_answer"]:
-                        answer = answer_info["free_form_answer"]
-                        answer_type = "abstractive"
-                    elif answer_info["yes_no"]:
-                        answer = "Yes"
-                        answer_type = "boolean"
-                    elif answer_info["yes_no"] is not None:
-                        answer = "No"
-                        answer_type = "boolean"
-                    else:
-                        raise RuntimeError(f"Annotation {answer_info['annotation_id']} does not contain an answer")
-                    if text_evidence_only:
-                        evidence = [text for text in answer_info["evidence"] if "FLOAT SELECTED" not in text]
-                    else:
-                        evidence = answer_info["evidence"]
-                    references.append({"answer": answer, "evidence": evidence, "type": answer_type})
-            answers_and_evidence[question_id] = references
-
-    return answers_and_evidence
-
-
-def evaluate(gold, predicted):
-    max_answer_f1s = []
-    max_evidence_f1s = []
-    max_answer_f1s_by_type = {
-        "extractive": [],
-        "abstractive": [],
-        "boolean": [],
-        "none": [],
-    }
-    num_missing_predictions = 0
-    for question_id, references in gold.items():
-        if question_id not in predicted:
-            num_missing_predictions += 1
-            max_answer_f1s.append(0.0)
-            max_evidence_f1s.append(0.0)
-            continue
-        answer_f1s_and_types = [
-            (token_f1_score(predicted[question_id]["answer"], reference["answer"]),
-             reference["type"])
-            for reference in gold[question_id]
-        ]
-        max_answer_f1, answer_type = sorted(answer_f1s_and_types, key=lambda x: x[0], reverse=True)[0]
-        max_answer_f1s.append(max_answer_f1)
-        max_answer_f1s_by_type[answer_type].append(max_answer_f1)
-        evidence_f1s = [
-            paragraph_f1_score(predicted[question_id]["evidence"], reference["evidence"])
-            for reference in gold[question_id]
-        ]
-        max_evidence_f1s.append(max(evidence_f1s))
-
-    mean = lambda x: sum(x) / len(x) if x else 0.0
-    return {
-        "Answer F1": mean(max_answer_f1s),
-        "Answer F1 by type": {key: mean(value) for key, value in max_answer_f1s_by_type.items()},
-        "Evidence F1": mean(max_evidence_f1s),
-        "Missing predictions": num_missing_predictions
-    }
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser()
-    parser.add_argument(
-        "--predictions",
-        type=str,
-        required=True,
-        help="""JSON lines file with each line in format:
-                {'question_id': str, 'predicted_answer': str, 'predicted_evidence': List[str]}"""
-    )
-    parser.add_argument(
-        "--gold",
-        type=str,
-        required=True,
-        help="Test or dev set from the released dataset"
-    )
-    parser.add_argument(
-        "--text_evidence_only",
-        action="store_true",
-        help="If set, the evaluator will ignore evidence in figures and tables while reporting evidence f1"
-    )
-    args = parser.parse_args()
-    gold_data = json.load(open(args.gold))
-    gold_answers_and_evidence = get_answers_and_evidence(gold_data, args.text_evidence_only)
-    predicted_answers_and_evidence = {}
-    for line in open(args.predictions):
-        prediction_data = json.loads(line)
-        predicted_answers_and_evidence[prediction_data["question_id"]] = {
-            "answer": prediction_data["predicted_answer"],
-            "evidence": prediction_data["predicted_evidence"]
-        }
-    evaluation_output = evaluate(gold_answers_and_evidence, predicted_answers_and_evidence)
-    print(json.dumps(evaluation_output, indent=2))

original_data/qasper-train-dev-v0.3/README.md

@@ -1,71 +0,0 @@
-A Dataset of Information Seeking Questions and Answers Anchored in Research Papers
-----------------------------------------------------------------------------------
-
-## Version 0.3
-
-The tarball you found this README in should contain the training and development sets of Qasper version 0.3. The images of the tables and figures
-in the papers associated can be found here: https://qasper-dataset.s3.us-west-2.amazonaws.com/train_dev_figures_and_tables.tgz
-
-The full text of the papers is extracted from S2ORC (Lo et al., 2020).
-
-Each file is in JSON format, where the keys are arxiv ids, and the values are dicts containing `title`, `abstract`, `full_text`, `figures_and_tables`, and `qas` (QA pairs).
-
-## Differences from v0.2
-
-Due to an issue in the annotation interface, a small number of annotations (about 0.6%) had multiple answer types (e.g.: unanswerable and boolean; see more information on answer types in the final section of this README) in v0.2. These were manually fixed to create v0.3. These fixes affected train, development, and test sets.
-
-## Figures and tables
-
-These are new starting version 0.2. The actual images of the figures and tables can be downloaded from the link above. The JSON files contain the
-captions to those images in the `figure_and_table_captions` field.
-
-This field is a dict whose keys are file names of the images of tables and figures, and the values are their captions.
-
-For example, the paper with arxiv id `1811.00942` is in the training set, and contains the following `figures_and_tables` field:
-
-```
-"figures_and_tables": [
-      {
-        "file": "3-Table1-1.png",
-        "caption": "Table 1: Comparison of neural language models on Penn Treebank and WikiText-103."
-      },
-      {
-        "file": "4-Figure1-1.png",
-        "caption": "Figure 1: Log perplexity\u2013recall error with KN-5."
-      },
-      {
-        "file": "4-Figure2-1.png",
-        "caption": "Figure 2: Log perplexity\u2013recall error with QRNN."
-      },
-      {
-        "file": "4-Table2-1.png",
-        "caption": "Table 2: Language modeling results on performance and model quality."
-      }
-]
-``` 
-
-and when you download the `train_dev_figures_and_tables` tarball, you will see four files in `train/1811.00942`, with file names corresponding to
-the `file` fields in the list above.
-
-## Fields specific to questions:
-
- - `nlp_background` shows the experience the question writer had. The values can be `zero` (no experience), `two` (0 - 2 years of experience), `five` (2 - 5 years of experience), and `infinity` (> 5 years of experience). The field may be empty as well, indicating the writer has chosen not to share this information.
-
- - `topic_background` shows how familiar the question writer was with the topic of the paper. The values are `unfamiliar`, `familiar`, `research` (meaning that the topic is the research area of the writer), or null.
-
- - `paper_read`, when specified shows whether the questionwriter has read the paper.
-
- - `search_query`, if not empty, is the query the question writer used to find the abstract of the paper from a large pool of abstracts we made available to them.
-
-## Fields specific to answers
-
-Unanswerable answers have `unanswerable` set to true. The remaining answers have exactly one of the following fields being non-empty.
-
- - `extractive_spans` are spans in the paper which serve as the answer.
- - `free_form_answer` is a written out answer.
- - `yes_no` is true iff the answer is Yes, and false iff the answer is No.
-
-`evidence` is the set of paragraphs, figures or tables used to arrive at the answer. When the evidence is a table or a figure, it starts with the
-string `FLOAT SELECTED`, and contains the caption of the corresponding table or figure.
-
-`highlighted_evidence` is the set of sentences the answer providers selected as evidence if they chose textual evidence. The text in the `evidence` field is a mapping from these sentences to the paragraph level. That is, if you see textual evidence in the `evidence` field, it is guaranteed to be entire paragraphs, while that is not the case with `highlighted_evidence`.

original_data/qasper-train-dev-v0.3/qasper-dev-v0.3.json

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:2ae7ee62a65b1c4225791c70de80c2aad4e8998cf1fd4f09a53103db4f21af93
-size 11398686

original_data/qasper-train-dev-v0.3/qasper-train-v0.3.json

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:9458bfe76074a8fa8d1685af02bcc73537aa6d338ad20591dfaff1946bc88bf4
-size 31969387