script to create data in GEM format; the input is in format introduced originally in the ACL 19 paper on data-to-text generation with entity modeling

create_mlb_data.py

@@ -0,0 +1,177 @@
+"""
+Script creates mlb data for gem benchmark
+"""
+import argparse
+import os
+import json
+import logging
+logging.basicConfig(level=logging.INFO)
+
+
+def sort_files_key(x):
+    if "train" in x:
+        file_index = int(x[5:7].strip("."))  # get the index of the train file
+    else:
+        file_index = -1  # valid and test
+    return file_index
+
+
+def filter_summaries(summary_entry, seen_output, test_seen_output):
+    match_words = {"rain", "rains", "rained", "snow"}
+    filter = False
+    if len(summary_entry["summary"]) < 100:
+        filter = True
+    elif 100 < len(summary_entry["summary"]) < 300:
+        if len(match_words.intersection(set(summary_entry["summary"]))) > 0:
+            filter = True
+    elif "_".join(summary_entry["summary"][:50]) in seen_output:  # retaining only one instance
+        filter = True
+    elif "_".join(summary_entry["summary"][:50]) in test_seen_output:  # retaining only one instance
+        filter = True
+    return filter
+
+
+def replace_Carmona(obj):
+    def decode_dict(a_dict):
+        for key, value in a_dict.items():
+            try:
+                if value == "Roberto Hernandez":
+                    a_dict[key] = value.replace("Roberto Hernandez", "Fausto Carmona")
+            except AttributeError:
+                pass
+        return a_dict
+    return json.loads(json.dumps(obj), object_hook=decode_dict)
+
+
+def process(input_folder, type, output_folder):
+    output_name_map = {"train": "train", "valid": "validation", "test": "test"}
+    output_jsonl = open(os.path.join(output_folder, output_name_map[type] + ".jsonl"), mode="w", encoding="utf-8")
+    file_list = os.listdir(input_folder)
+    sorted_file_list = sorted(file_list, key=sort_files_key)
+    seen_output = set()
+    test_seen_output = set()
+    if type == "train":
+        for filename in sorted_file_list:
+            if "valid" in filename or "test" in filename:
+                print("test filename", filename)
+                json_file = open(os.path.join(input_folder, filename), mode="r", encoding="utf-8")
+                data = json.load(json_file)
+                for entry_index, entry in enumerate(data):
+                    test_seen_output.add("_".join(entry["summary"][:50]))
+
+    index = 1
+    for filename in sorted_file_list:
+        if type in filename:
+            print("filename", filename)
+            json_file = open(os.path.join(input_folder, filename), mode="r", encoding="utf-8")
+            data = json.load(json_file)
+            for entry_index, entry in enumerate(data):
+                logging.debug("instance %s", entry_index)
+                if type == "train":
+                    if filter_summaries(entry, seen_output, test_seen_output):
+                        continue
+                seen_output.add("_".join(entry["summary"][:50]))
+
+                summary = entry["summary"]
+                summ = " ".join(summary)
+                if "Fausto Carmona" in summ:
+                    entry = replace_Carmona(entry)
+                gem_id = "GEM-mlb_data_to_text-"+ output_name_map[type] + "-" + str(index)
+                index += 1
+
+                updated_entry = {}
+                updated_entry["home_name"] = entry["home_name"]
+                updated_entry["vis_name"] = entry["vis_name"]
+                updated_entry["home_city"] = entry["home_city"]
+                updated_entry["vis_city"] = entry["vis_city"]
+                updated_entry["summary"] = entry["summary"]
+                updated_entry["summary_eval"] = " ".join(entry["summary"]).replace("*NEWPARAGRAPH* ", "")
+                updated_entry["day"] = entry["day"]
+                updated_entry["gem_id"] = gem_id
+                updated_entry["box_score"] = []
+                box_score_keys = entry["box_score"].keys()
+                construct_box_score(box_score_keys, entry["box_score"], updated_entry["box_score"])
+                assert len(updated_entry["box_score"]) == len(
+                    entry["box_score"][list(box_score_keys)[-1]])  # checking sizes match
+                updated_entry["play_by_play"] = []
+                construct_play_by_play(entry["play_by_play"], updated_entry["play_by_play"])
+                updated_entry["vis_line"] = {}
+                updated_entry["home_line"] = {}
+                for attrib in ["team_runs", "result", "team_hits", "team_name", "team_errors", "team_city"]:
+                    updated_entry["vis_line"][attrib] = entry["vis_line"][attrib]
+                    updated_entry["home_line"][attrib] = entry["home_line"][attrib]
+                updated_entry["vis_line"]["innings"] = []
+                construct_inning_scores(entry["vis_line"]["innings"], updated_entry["vis_line"]["innings"])
+                updated_entry["home_line"]["innings"] = []
+                construct_inning_scores(entry["home_line"]["innings"], updated_entry["home_line"]["innings"])
+                json.dump(updated_entry, output_jsonl, ensure_ascii=False)
+                output_jsonl.write('\n')
+                if entry_index % 50 == 0:
+                    print("entry_index", entry_index)
+    output_jsonl.close()
+
+
+def construct_box_score(box_score_keys, box_score, box_score_list):
+    player_index = 0
+    while True:
+        box_score_object = {}
+        for _box_score_key in box_score_keys:
+            if str(player_index) not in box_score[_box_score_key]:
+                return
+            box_score_object[_box_score_key] = box_score[_box_score_key][str(player_index)]
+        box_score_list.append(box_score_object)
+        player_index += 1
+
+
+def construct_play_by_play(play_by_play, play_by_play_list):
+    inning_index = 1
+    while True:
+        if str(inning_index) not in play_by_play:
+            return
+        play_by_play_object = {}
+        for side in ["top", "bottom"]:
+            if side in play_by_play[str(inning_index)]:
+                play_by_play_object[side] = []
+                for play in play_by_play[str(inning_index)][side]:
+                    play_object = construct_play_object(play)
+                    play_by_play_object[side].append(play_object)
+
+        play_by_play_object["inning"] = inning_index
+        play_by_play_list.append(play_by_play_object)
+        inning_index += 1
+
+
+def construct_play_object(play):
+    play_object = {}
+    for attrib in ["runs", "pitcher", "o", "b", "s", "batter", "event", "event2", "home_team_runs", "away_team_runs",
+                   "rbi", "error_runs", "fielder_error"]:
+        play_object[attrib] = "N/A"
+        if attrib in play:
+            play_object[attrib] = play[attrib]
+    for attrib in ["scorers", "b1", "b2", "b3"]:
+        play_object[attrib] = ["N/A"]
+        if attrib in play:
+            play_object[attrib] = play[attrib]
+    return play_object
+
+
+def construct_inning_scores(innings, innings_list):
+    inning_index = 1
+    while True:
+        if "inn" + str(inning_index) not in innings:
+            return
+        innings_list.append({"inn": inning_index, "runs": innings["inn" + str(inning_index)]})
+        inning_index += 1
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Script for constructing the mlb dataset in GEM format')
+    parser.add_argument('-json_root', type=str,
+                        help='path of json root; download from '
+                             'https://drive.google.com/drive/folders/1G4iIE-02icAU2-5skvLlTEPWDQQj1ss4', default=None)
+    parser.add_argument('-output_folder', type=str,
+                        help='path of output file', default=None)
+    parser.add_argument('-dataset_type', type=str,
+                        help='type of dataset', default=None)
+    args = parser.parse_args()
+    process(args.json_root, args.dataset_type, args.output_folder)