Spaces:
Running on Zero
Running on Zero
File size: 7,197 Bytes
f1ef7e2 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 | """
Run selected V5 emotion/sentiment inference on AppTek call-center samples.
This script uses the final selected V5 model through EmotionPredictor and
generates realistic call-center inference outputs.
Run smoke test from ml-services:
python -m src.inference.run_apptek_inference --max-calls 2 --max-duration-seconds 30
Run full AppTek sample:
python -m src.inference.run_apptek_inference --max-calls 20 --max-duration-seconds 60
"""
import argparse
import json
from pathlib import Path
from typing import Any, Dict, Optional
import pandas as pd
from src.inference.emotion_predictor import (
EmotionPredictor,
SELECTED_EMOTION_MODEL_VERSION,
)
PROJECT_ROOT = Path(__file__).resolve().parents[3]
ML_SERVICES_ROOT = PROJECT_ROOT / "ml-services"
APPTEK_METADATA_PATH = (
ML_SERVICES_ROOT / "data" / "processed" / "apptek" / "apptek_metadata.csv"
)
APPTEK_RESULTS_DIR = (
ML_SERVICES_ROOT / "outputs" / "apptek" / "sentiment_results"
)
APPTEK_SUMMARY_PATH = (
ML_SERVICES_ROOT / "outputs" / "apptek" / "apptek_sentiment_summary.csv"
)
def result_to_dict(result: Any) -> Dict:
"""
Convert AudioSentimentResult to dictionary for JSON saving.
Supports Pydantic v2, Pydantic v1, or dataclass-like objects.
"""
if hasattr(result, "model_dump"):
return result.model_dump()
if hasattr(result, "dict"):
return result.dict()
if hasattr(result, "__dict__"):
return result.__dict__
raise TypeError(f"Cannot convert result to dict: {type(result)}")
def run_apptek_inference(
metadata_path: Path = APPTEK_METADATA_PATH,
max_calls: Optional[int] = None,
max_duration_seconds: float = 60.0,
build_timeline: bool = True,
) -> None:
"""
Run selected emotion model on AppTek samples.
"""
if not metadata_path.exists():
raise FileNotFoundError(
f"AppTek metadata not found: {metadata_path}\n"
"Run python -m src.data.apptek_dataset first."
)
metadata = pd.read_csv(metadata_path)
if max_calls is not None:
metadata = metadata.head(max_calls)
APPTEK_RESULTS_DIR.mkdir(parents=True, exist_ok=True)
APPTEK_SUMMARY_PATH.parent.mkdir(parents=True, exist_ok=True)
predictor = EmotionPredictor(max_duration_seconds=max_duration_seconds)
summary_rows = []
print("\nRunning AppTek inference")
print("-" * 80)
print(f"Selected model: {SELECTED_EMOTION_MODEL_VERSION}")
print(f"Metadata: {metadata_path}")
print(f"Calls to process: {len(metadata)}")
print(f"Max duration per call: {max_duration_seconds} seconds")
print(f"Build timeline: {build_timeline}")
print("-" * 80)
for _, row in metadata.iterrows():
call_id = row["call_id"]
audio_path = ML_SERVICES_ROOT / row["audio_path"]
domain = row.get("selected_domain", row.get("domain", ""))
raw_domain = row.get("raw_domain", domain)
print(f"Processing {call_id} | domain={domain} | raw_domain={raw_domain} | audio={audio_path.name}")
result = predictor.analyze_audio(
audio_path=audio_path,
call_id=call_id,
build_timeline=build_timeline,
)
result_dict = result_to_dict(result)
# Add AppTek metadata context to output.
result_dict["apptek_metadata"] = {
"domain": domain,
"raw_domain": raw_domain,
"gender": row.get("gender", ""),
"accent": row.get("accent", ""),
"duration_seconds": float(row.get("duration_seconds", 0.0)),
"text_preview": str(row.get("text", ""))[:500],
}
output_path = APPTEK_RESULTS_DIR / f"{call_id}_sentiment.json"
with output_path.open("w", encoding="utf-8") as file:
json.dump(result_dict, file, indent=2)
summary_rows.append(
{
"call_id": call_id,
"domain": domain,
"raw_domain": raw_domain,
"gender": row.get("gender", ""),
"accent": row.get("accent", ""),
"duration_seconds": row.get("duration_seconds", 0.0),
"processed_duration_seconds": max_duration_seconds,
"overall_audio_sentiment": result_dict.get("overall_audio_sentiment"),
"dominant_emotion": result_dict.get("dominant_emotion"),
"negative_emotion_probability": result_dict.get(
"negative_emotion_probability"
),
"anger_probability": result_dict.get("anger_probability"),
"stress_probability": result_dict.get("stress_probability"),
"sadness_probability": result_dict.get("sadness_probability"),
"anxiety_probability": result_dict.get("anxiety_probability"),
"calm_probability": result_dict.get("calm_probability"),
"audio_escalation_score": result_dict.get("audio_escalation_score"),
"risk_level": result_dict.get("risk_level"),
"prediction_confidence": result_dict.get("prediction_confidence"),
"confidence_level": result_dict.get("confidence_level"),
"uncertain_prediction": result_dict.get("uncertain_prediction"),
"model_version": result_dict.get("model_version"),
"result_json": str(output_path.relative_to(ML_SERVICES_ROOT)),
}
)
summary_df = pd.DataFrame(summary_rows)
summary_df.to_csv(APPTEK_SUMMARY_PATH, index=False)
print("\nAppTek inference completed successfully.")
print("-" * 80)
print(f"Saved JSON results to: {APPTEK_RESULTS_DIR}")
print(f"Saved summary CSV to: {APPTEK_SUMMARY_PATH}")
print("-" * 80)
print("\nSummary preview:")
print(
summary_df[
[
"call_id",
"domain",
"dominant_emotion",
"overall_audio_sentiment",
"audio_escalation_score",
"risk_level",
"confidence_level",
]
].head()
)
def parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser(
description="Run selected V5 emotion inference on AppTek samples."
)
parser.add_argument(
"--metadata-path",
type=Path,
default=APPTEK_METADATA_PATH,
help="Path to AppTek metadata CSV.",
)
parser.add_argument(
"--max-calls",
type=int,
default=None,
help="Maximum number of AppTek calls to process.",
)
parser.add_argument(
"--max-duration-seconds",
type=float,
default=60.0,
help="Maximum duration per call to process.",
)
parser.add_argument(
"--no-timeline",
action="store_true",
help="Disable sentiment timeline generation.",
)
return parser.parse_args()
if __name__ == "__main__":
args = parse_args()
run_apptek_inference(
metadata_path=args.metadata_path,
max_calls=args.max_calls,
max_duration_seconds=args.max_duration_seconds,
build_timeline=not args.no_timeline,
) |