"""Arc Tagger quickstart. Run from the downloaded repo:  python example.py  [local dir or repo id, default: this dir]
(the repo id is WhissleAI/arc-tagger)"""
import json, os, sys, time
HERE = os.path.dirname(os.path.abspath(__file__)); sys.path.insert(0, HERE)
import torch
torch.set_num_threads(int(os.environ.get("ARCTAG_THREADS", "2")))
from arctag import ArcTagger

tg = ArcTagger.from_pretrained(sys.argv[1] if len(sys.argv) > 1 else HERE)

# 1) offline: plain text (utterances end after . ? !)
out = tg.tag("Hi, thanks for calling Northwind Bank, this is Maria. My name is John Patel and my number is "
             "five five five, two one three, four four eight nine.", domain="talk")
print("redacted:", out["redacted_text"])
for s in out["spans"]:
    print(f"  span {s['tag']:<34} {s['text']!r} p={s['p']}")
for u in out["utterances"]:
    print(f"  utt [{u['start']}:{u['end']}] act={u['act']} intent={u['intent']} safety={u['safety']}")

# 2) streaming: push ASR words as they arrive (interim -> revised final), endpoint at end of utterance
st = tg.stream(domain="assistant")
st.push_words(["set", "an", "alarm", "for"], final=False)                                      # interim hypothesis
up = st.push_words(["set", "an", "alarm", "for", "seven", "thirty."], revise_from=0, final=True)  # final replaces words[0:]
print("live spans:", [(s["tag"], s["text"]) for s in up["spans"]])
up = st.endpoint()
print("utterance:", up["utterances"][0]["intent"], up["utterances"][0]["act"])
res = st.finalize()
print("latency:", st.s.stats())
print(json.dumps({k: res[k] for k in ("meta", "spans", "events")}, indent=1)[:1500])