analyze_stats.py

#!/usr/bin/env python3
"""Parse Chatterino .log files, write JSONL to data/, and print dataset stats."""

import re
import json
import string
import sys
from pathlib import Path
from collections import Counter

ROOT = Path(__file__).parent
LOGS_DIR = ROOT / "original-logs"
DATA_DIR = ROOT / "data"

MSG_RE = re.compile(r"^\[(\d{2}:\d{2}:\d{2})\] ([^:]+): (.+)$")
DATE_HEADER_RE = re.compile(r"^# Start logging at (\d{4}-\d{2}-\d{2})")

UNICODE_EMOJI_RE = re.compile(
    "["
    "\U0001F300-\U0001F9FF"
    "\U0001FA00-\U0001FA9F"
    "\U00002600-\U000027BF"
    "\U00002300-\U000023FF"
    "\U00002B00-\U00002BFF"
    "\U0001F000-\U0001F02F"
    "\U0001F0A0-\U0001F0FF"
    "]"
)

# Common global + channel Twitch emotes
TWITCH_EMOTES = {
    "PogChamp","Pog","PogU","POGGERS","PauseChamp","PepeHands","FeelsBadMan",
    "FeelsGoodMan","Kappa","LUL","KEKW","OMEGALUL","LULW","monkaS","monkaHmm",
    "monkaGIGA","AYAYA","NotLikeThis","WutFace","VoteYea","VoteNay","KKona",
    "KKonaW","BibleThump","EZ","Sadge","Copium","COPIUM","WidepeepoHappy",
    "peepoHappy","peepoSad","peepoWTF","pepeJAM","PepeJAM","pepeLaugh",
    "PepeLaugh","pepeD","pepePoint","HYPERCLAP","Clap","TriHard","haHAA",
    "ResidentSleeper","gachiHYPER","gachiBASS","forsenCD","forsenE","GIGACHAD",
    "BASED","TOOBASED","NODDERS","YEPPERS","YEP","Chatting","Bedge","Hmm",
    "WhosThisDiva","CumDance","NOTED","xqcL",
    # hasanabi channel
    "hasL","hasS","hasPog","hasCop","hasFIST","hasKek","hasFeel","hasThink",
    "PIKMINPARTY","pepePoint","TOOBASED",
    # zackrawrr channel
    "asmongPog","asmongCozy","asmongKEK","ddHuh","XEsht",
}

STOPWORDS = set(
    "the a an and or but in on at to for of is are was were be been being "
    "have has had do does did will would could should may might shall can "
    "i you he she it we they me him her us them my your his its our their "
    "what which who where when how this that these those just like so no "
    "not dont its its im its yeah yes lol".split()
)


def classify_line(line: str) -> dict | None:
    m = MSG_RE.match(line)
    if m:
        ts, username, message = m.groups()
        return {"type": "message", "timestamp": ts, "username": username.strip(), "message": message}
    m = re.match(r"^\[(\d{2}:\d{2}:\d{2})\] (.+)$", line)
    if not m:
        return None
    ts, rest = m.groups()
    if "permanently banned" in rest:
        event = "ban"
    elif "timed out" in rest:
        event = "timeout"
    elif "subscribed" in rest or "gifted" in rest:
        event = "subscription"
    elif "watch streak" in rest:
        event = "watch_streak"
    elif "joined channel" in rest or "is live" in rest or rest == "Announcement":
        event = "system"
    else:
        event = "system"
    return {"type": event, "timestamp": ts, "text": rest}


def is_emote(token: str) -> bool:
    clean = token.strip(string.punctuation)
    if clean in TWITCH_EMOTES:
        return True
    # All-caps alpha, 3+ chars — likely a global or 7TV emote
    if clean.isalpha() and clean.isupper() and len(clean) >= 3:
        return True
    return False


def ngrams(tokens: list[str], n: int) -> list[str]:
    return [" ".join(tokens[i : i + n]) for i in range(len(tokens) - n + 1)]


def analyze_channel(channel: str) -> dict:
    logs_path = LOGS_DIR / channel
    data_path = DATA_DIR / channel
    data_path.mkdir(parents=True, exist_ok=True)

    messages = []

    for log_file in sorted(logs_path.glob("*.log")):
        # Extract date from filename e.g. hasanabi-2026-05-07.log
        parts = log_file.stem.split("-", 1)
        date = parts[1] if len(parts) == 2 else log_file.stem

        out_path = data_path / (log_file.stem + ".jsonl")
        with open(log_file, encoding="utf-8", errors="replace") as f, \
             open(out_path, "w", encoding="utf-8") as out:
            for line in f:
                line = line.rstrip("\n")
                record = classify_line(line)
                if record is None:
                    continue
                record["channel"] = channel
                record["date"] = date
                out.write(json.dumps(record, ensure_ascii=False) + "\n")
                if record["type"] == "message":
                    messages.append(record)

    return compute_stats(channel, messages)


def compute_stats(channel: str, messages: list) -> dict:
    total = len(messages)
    if total == 0:
        return {"channel": channel, "total_messages": 0}

    user_counts = Counter(m["username"] for m in messages)
    unique_users = len(user_counts)
    top_user, top_count = user_counts.most_common(1)[0]
    single_senders = sum(1 for c in user_counts.values() if c == 1)
    pct_single = 100 * single_senders / unique_users

    unicode_emoji_total = 0
    twitch_emote_total = 0
    char_lengths = []
    word_counts = []
    word_lengths_sample = []
    all_content_tokens: list[str] = []  # non-emote tokens for n-grams/TTR
    longest_msg = ""
    longest_len = 0

    for m in messages:
        msg = m["message"]
        unicode_emoji_total += len(UNICODE_EMOJI_RE.findall(msg))

        clen = len(msg)
        char_lengths.append(clen)
        if clen > longest_len:
            longest_len = clen
            longest_msg = msg

        tokens = msg.split()
        word_counts.append(len(tokens))

        for tok in tokens:
            clean = tok.strip(string.punctuation)
            if not clean:
                continue
            if is_emote(clean):
                twitch_emote_total += 1
            else:
                word_lengths_sample.append(len(clean))
                all_content_tokens.append(clean.lower())

    # Vocabulary diversity (type-token ratio on first 100k tokens)
    sample = all_content_tokens[:100_000]
    ttr = len(set(sample)) / len(sample) if sample else 0.0

    # N-grams: filter stopword-only grams
    def useful(ng: str) -> bool:
        return not all(p in STOPWORDS for p in ng.split())

    bigram_counter: Counter = Counter()
    trigram_counter: Counter = Counter()
    for m in messages:
        tokens = [t.lower().strip(string.punctuation) for t in m["message"].split() if t.strip(string.punctuation) and not is_emote(t.strip(string.punctuation))]
        bigram_counter.update(ngrams(tokens, 2))
        trigram_counter.update(ngrams(tokens, 3))

    top_bigrams = [(ng, c) for ng, c in bigram_counter.most_common(100) if useful(ng)][:10]
    top_trigrams = [(ng, c) for ng, c in trigram_counter.most_common(100) if useful(ng)][:10]

    return {
        "channel": channel,
        "total_messages": total,
        "unique_users": unique_users,
        "unicode_emoji_count": unicode_emoji_total,
        "twitch_emote_count": twitch_emote_total,
        "total_emojis": unicode_emoji_total + twitch_emote_total,
        "top_sender": {"username": top_user, "count": top_count},
        "single_message_senders": single_senders,
        "pct_single_message_senders": round(pct_single, 1),
        "repeat_senders": unique_users - single_senders,
        "avg_messages_per_user": round(total / unique_users, 2),
        "avg_message_length_chars": round(sum(char_lengths) / total, 1),
        "avg_message_length_words": round(sum(word_counts) / total, 2),
        "avg_word_length_chars": round(sum(word_lengths_sample) / len(word_lengths_sample), 2) if word_lengths_sample else 0,
        "avg_emojis_per_message": round((unicode_emoji_total + twitch_emote_total) / total, 3),
        "longest_message": {"text": longest_msg[:200], "length_chars": longest_len},
        "vocabulary_diversity_ttr": round(ttr, 4),
        "top_bigrams": top_bigrams,
        "top_trigrams": top_trigrams,
    }


def print_stats(s: dict) -> None:
    ch = s["channel"]
    print(f"\n{'='*60}")
    print(f"  {ch.upper()}")
    print(f"{'='*60}")
    print(f"Total messages        : {s['total_messages']:,}")
    print(f"Unique users          : {s['unique_users']:,}")
    print(f"Unicode emoji         : {s['unicode_emoji_count']:,}")
    print(f"Twitch emotes         : {s['twitch_emote_count']:,}")
    print(f"Total emojis          : {s['total_emojis']:,}")
    print(f"Top sender            : {s['top_sender']['username']} ({s['top_sender']['count']:,} messages)")
    print(f"Single-msg senders    : {s['single_message_senders']:,} ({s['pct_single_message_senders']}% of users)")
    print(f"Repeat senders        : {s['repeat_senders']:,}")
    print(f"Avg msgs/user         : {s['avg_messages_per_user']}")
    print(f"Avg message (chars)   : {s['avg_message_length_chars']}")
    print(f"Avg message (words)   : {s['avg_message_length_words']}")
    print(f"Avg word length       : {s['avg_word_length_chars']} chars")
    print(f"Avg emojis/message    : {s['avg_emojis_per_message']}")
    print(f"Vocabulary diversity  : {s['vocabulary_diversity_ttr']} (TTR)")
    print(f"Longest message       : {s['longest_message']['length_chars']} chars")
    print(f"  \"{s['longest_message']['text'][:120]}\"")
    print(f"\nTop bigrams:")
    for ng, c in s["top_bigrams"]:
        print(f"  {c:>6,}  {ng}")
    print(f"\nTop trigrams:")
    for ng, c in s["top_trigrams"]:
        print(f"  {c:>6,}  {ng}")


if __name__ == "__main__":
    channels = sys.argv[1:] or [d.name for d in sorted(LOGS_DIR.iterdir()) if d.is_dir()]
    results = {}
    for ch in channels:
        print(f"Processing {ch}...", flush=True)
        stats = analyze_channel(ch)
        results[ch] = stats
        print_stats(stats)

    # Save raw stats for reference
    out = ROOT / "stats.json"
    with open(out, "w", encoding="utf-8") as f:
        json.dump(results, f, indent=2, ensure_ascii=False)
    print(f"\nStats saved to {out}")