def compute_subgraph_metrics(dataset, n_jobs, limit):
print("--- Subgraph Metrics ---")
print(f"Dataset: {dataset}")
print(f"Num Jobs: {n_jobs}")
print(f"Limit: {limit}")
print("---------------------------")
# paths
conf = Config(dataset)
output_fpath = f"{conf.data_root}/subgraph_metrics.csv"
json_fpaths = json_paths_iter(conf.conversations_no_embs_jsons_dir,
limit=limit)
# compute metrics
print("Computing metrics ...")
if n_jobs == 1:
metrics = [compute_metrics(json_fpath) \
for json_fpath in tqdm(json_fpaths)]
else:
parallel = Parallel(n_jobs=n_jobs, verbose=10)
metrics = parallel(
delayed(compute_metrics)(json_fpath) \
for json_fpath in json_fpaths
)
print("Output:", len(metrics))
# output to csv
print("Outputting tree metrics to CSV ...")
write_dicts_to_csv(metrics, output_fpath)
print("Done!")
def compute_dyad_metrics(dataset, n_jobs, limit):
# hard-coding some settings
toxicity_threshold = 0.531
splits_only = False
skip_root = True
print("--- Dyad Metrics ---")
print(f"Dataset: {dataset}")
print(f"Toxicity threshold: {toxicity_threshold}")
print(f"Num Jobs: {n_jobs}")
print(f"Limit: {limit}")
print("----------------------------")
conf = Config(dataset)
output_fpath = f"{conf.data_root}/dyad_metrics.csv"
json_fpaths = json_paths_iter(conf.conversations_jsons_dir, limit=limit)
# compute metrics
print("Computing metrics ...")
if n_jobs == 1:
metrics = [
process_conversation(json_fpath, toxicity_threshold, splits_only,
skip_root) for json_fpath in tqdm(json_fpaths)
]
else:
parallel = Parallel(n_jobs=n_jobs, verbose=10)
metrics = parallel(
delayed(process_conversation)(
json_fpath,
toxicity_threshold,
splits_only,
skip_root
) \
for json_fpath in json_fpaths
)
# flatten the results
metrics = list(itertools.chain.from_iterable(metrics))
print(len(metrics))
# output to CSV
fields = [
"root_tweet_id", "parent_tox", "parent_n_friends",
"parent_n_followers", "child_tox", "child_n_friends",
"child_n_followers", "dyad_type", "dyad_n_common_friends"
]
with open(output_fpath, "w") as fout:
writer = csv.writer(fout)
writer.writerow(fields)
writer.writerows(metrics)
print("Done!")
def compute_prefix_metrics(dataset, n_jobs=1, limit=None):
prefixes = [10, 20, 30, 40, 50, 60, 70, 80, 90, 100]
print("--- Prefix Metrics ---")
print(f"Dataset: {dataset}")
print(f"Num Jobs: {n_jobs}")
print(f"Limit: {limit}")
print(f"Prefixes: {prefixes}")
print("----------------------------")
# paths
conf = Config(dataset)
output_fpath = f"{conf.data_root}/prefix_metrics/{dataset}.json.gz"
output_pickle_fpath = f"{conf.data_root}/prefix_metrics/{dataset}.pkl.gz"
json_fpaths = json_paths_iter(conf.conversations_jsons_dir, limit=limit)
# compute metrics
print("Computing metrics ...")
if n_jobs == 1:
metrics = [conversation_prefix_metrics(json_fpath, prefixes) \
for json_fpath in tqdm(json_fpaths)]
else:
parallel = Parallel(n_jobs=n_jobs, verbose=10)
metrics = parallel(
delayed(conversation_prefix_metrics)(json_fpath, prefixes) \
for json_fpath in json_fpaths
)
print(f"Metrics total: {len(metrics)}")
# skip empty results
metrics = [m for m in metrics if len(m) > 0]
print(f"Metrics non-zero: {len(metrics)}")
# pickle
with gzip.open(output_pickle_fpath, "wb") as fout:
pickle.dump(metrics, fout, protocol=4)
# uJSON complains: cast numpy ints/floats to python ints/floats
for conv_metrics in metrics:
for prefix_n, prefix_metrics in conv_metrics.items():
if prefix_n != "root_tweet_id":
for group_name, group_values in prefix_metrics.items():
if group_values is not None:
group_values = sanitize_numpy_types(group_values)
# output metrics to JSON
print("Outputting results to JSON ...")
with gzip.open(output_fpath, "wt") as fout:
json.dump(metrics, fout)
print("Done!")
def compute_user_metrics(dataset, n_jobs=1, limit=None):
print("--- User Metrics ---")
print(f"Dataset: {dataset}")
print(f"Num Jobs: {n_jobs}")
print(f"Limit: {limit}")
print("----------------------------")
toxicity_threshold = 0.531
conf = Config(dataset)
json_fpaths = json_paths_iter(conf.conversations_no_embs_jsons_dir,
limit=limit)
# all_user_conv_stats = [
# compute_user_conversation_stats(json_fpath, toxicity_threshold) \
# for json_fpath in json_fpaths]
parallel = Parallel(n_jobs=n_jobs, verbose=10)
all_user_conv_stats = parallel(
delayed(compute_user_conversation_stats)(
json_fpath,
toxicity_threshold
) \
for json_fpath in json_fpaths
)
print("Aggregating user metrics ...")
user_stats = agg_user_stats(all_user_conv_stats)
user_stats_csv = [{"user_id": u_id, **u_stats} \
for u_id, u_stats in user_stats.items()]
# out_json_fpath = f"{conf.data_root}/user_metrics.json.gz"
# json.dump(user_stats, gzip.open(out_json_fpath, "wt"), indent=2)
out_csv_fpath = f"{conf.data_root}/user_metrics.csv"
write_dicts_to_csv(user_stats_csv, out_csv_fpath)
print("Done!")
def compute_toxicity_metrics(dataset, n_jobs=1, limit=None):
print("--- Toxicity Metrics ---")
print(f"Dataset: {dataset}")
print(f"Num Jobs: {n_jobs}")
print(f"Limit: {limit}")
print("----------------------------")
# paths
conf = Config(dataset)
output_fpath = f"{conf.data_root}/toxicity.csv"
# iterator
json_fpaths = json_paths_iter(
conf.conversations_no_embs_jsons_dir,
limit=limit
)
# compute metrics
print("Computing metrics ...")
if n_jobs == 1:
metrics = [toxicity_metrics(json_fpath) \
for json_fpath in tqdm(json_fpaths)]
else:
parallel = Parallel(n_jobs=n_jobs, verbose=10)
metrics = parallel(
delayed(toxicity_metrics)(json_fpath) \
for json_fpath in json_fpaths
)
print("Metrics computed:", len(metrics))
print("Outputting metrics to CSV ...")
write_dicts_to_csv(metrics, output_fpath)
print("Done!")