#!/bin/bash
# pb 全链: 等 dl_pb_mirror 完成 -> 冒烟 -> B 全量 -> convert -> count gate
set -x
export PYTHONPATH=/root/patientphex/retrain_t1/code
export NLTK_DATA=/root/patientphex/retrain_t1/third_party/PhenoTagger-Plus/nltk_data
PY=/opt/anaconda3/envs/pt_env310/bin/python
C=/root/patientphex/retrain_t1/code
PB=/root/patientphex/retrain_t1/third_party/PhenoBERT
O=/root/patientphex/pb_runtime
R=$C/artifacts/b_retrain
B=/root/patientphex/PatientPheX-V1-B/PatientPheX-V1-B.jsonl

# 等待权重下载进程结束
for i in $(seq 1 240); do
  pgrep -f dl_pb_mirror.sh > /dev/null || break
  sleep 30
done
pgrep -f dl_pb_mirror.sh > /dev/null && { echo PB_DL_STILL_RUNNING; exit 1; }

# 等 stanza 资源就位
for i in $(seq 1 120); do
  grep -q STANZA_TAR_DONE /root/patientphex/dl_stanza4.out 2>/dev/null && break
  sleep 30
done
grep -q STANZA_TAR_DONE /root/patientphex/dl_stanza4.out || { echo STANZA_NOT_READY; exit 1; }
ls /root/stanza_resources/en/ | head -4

ls -la $PB/phenobert/models/ $PB/phenobert/embeddings/ | head -50
[ -f $PB/phenobert/models/HPOModel_H/model_layer1.pkl ] || { echo PB_LAYER1_MISSING; exit 2; }
[ -f $PB/phenobert/models/bert_model_max_triple.pkl ] || { echo PB_BERTPKL_MISSING; exit 2; }
[ -f $PB/phenobert/embeddings/fasttext_pubmed.bin ] || { echo PB_FASTTEXT_MISSING; exit 2; }
du -sb $PB/phenobert/embeddings/fasttext_pubmed.bin
echo PB_WEIGHTS_READY

# 冒烟: 单篇文档
mkdir -p $O/smoke_in $O/smoke_out
head -c 100000 $O/corpus/$(ls $O/corpus | head -1) > /dev/null
cp $(ls -d $O/corpus/* | head -1) $O/smoke_in/
CUDA_VISIBLE_DEVICES=4 $PY $C/scripts/run_phenobert_official_compat.py \
  --phenobert $PB --input-dir $O/smoke_in --output-dir $O/smoke_out \
  --threads 16 || { echo PB_SMOKE_FAIL; exit 3; }
ls $O/smoke_out/
cat $O/smoke_out/*.json 2>/dev/null | head -c 600
echo PB_SMOKE_DONE

# B 全量
mkdir -p $O/out
CUDA_VISIBLE_DEVICES=4 $PY $C/scripts/run_phenobert_official_compat.py \
  --phenobert $PB --input-dir $O/corpus --output-dir $O/out \
  --threads 32 || { echo PB_FULL_FAIL; exit 4; }
ls $O/out | wc -l
echo PB_FULL_DONE

# convert + 2026 过滤
$PY $C/scripts/prepare_phenobert_official.py convert \
  --input $B \
  --annotations $O/out \
  --ontology /root/patientphex/PatientPheX-V1-A/hp.obo \
  --raw-output $R/B-phenobert-raw.jsonl \
  --compatible-output $R/B-phenobert-official-hpo2026.jsonl \
  --report $R/B-phenobert-conversion-report.json || { echo PB_CONVERT_FAIL; exit 5; }
/root/patientphex/judge_venv/bin/python /root/patientphex/count_ents.py $R/B-phenobert-official-hpo2026.jsonl 8110
echo PB_CHAIN_DONE