Spaces:
Sleeping
Sleeping
File size: 2,760 Bytes
556524e 72e2b6e 556524e 72e2b6e 556524e 72e2b6e 556524e 72e2b6e 556524e | 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 | import importlib.util
from pathlib import Path
_client_path = Path(__file__).parent.parent / "api_client.py"
_spec = importlib.util.spec_from_file_location("api_client", _client_path)
_module = importlib.util.module_from_spec(_spec)
_spec.loader.exec_module(_module)
api_predict = _module.api_predict
require_api = _module.require_api
import pandas as pd
import streamlit as st
require_api()
st.title("Compare Models")
st.caption("Side-by-side comparison of all trained models on CLINC150 test set.")
RESULTS = {
"logreg": {"accuracy": 0.8058, "macro_f1": 0.8416, "weighted_f1": 0.7985, "latency_p50_ms": 0.12},
"svm": {"accuracy": 0.8002, "macro_f1": 0.8474, "weighted_f1": 0.7795, "latency_p50_ms": 0.14},
"textcnn": {"accuracy": 0.7740, "macro_f1": 0.8244, "weighted_f1": 0.7646, "latency_p50_ms": 97.6},
"rnn": {"accuracy": 0.1818, "macro_f1": 0.0020, "weighted_f1": 0.0560, "latency_p50_ms": 124.5},
"lstm": {"accuracy": 0.7669, "macro_f1": 0.8113, "weighted_f1": 0.7555, "latency_p50_ms": 249.2},
"distilbert": {"accuracy": 0.8765, "macro_f1": 0.9019, "weighted_f1": 0.8722, "latency_p50_ms": 9.6},
}
df = pd.DataFrame(RESULTS).T
df.index.name = "model"
st.subheader("Metrics Table")
st.dataframe(
df.style.format({
"accuracy": "{:.2%}",
"macro_f1": "{:.2%}",
"weighted_f1": "{:.2%}",
"latency_p50_ms": "{:.2f} ms",
}).background_gradient(subset=["accuracy", "macro_f1", "weighted_f1"], cmap="Greens"),
use_container_width=True,
)
st.subheader("Accuracy & F1 Comparison")
chart_df = df[["accuracy", "macro_f1", "weighted_f1"]]
st.bar_chart(chart_df)
st.subheader("Latency Comparison")
st.bar_chart(df[["latency_p50_ms"]])
st.divider()
st.subheader("Side-by-Side Prediction")
st.caption("Run the same input through multiple models at once.")
text = st.text_input("Enter a query to compare", placeholder="e.g. cancel my hotel reservation")
selected_models = st.multiselect(
"Models to compare",
options=["classical", "svm", "transformer"],
default=["classical", "transformer"],
)
if st.button("Compare", type="primary", disabled=not (text and selected_models)):
cols = st.columns(len(selected_models))
for col, model_type in zip(cols, selected_models):
with col:
with st.spinner(f"running {model_type}..."):
result = api_predict(text, model_type)
st.markdown(f"**{model_type}**")
st.metric("Intent", result["intent"])
st.metric("Confidence", f"{result['confidence']:.2%}")
st.metric("Latency", f"{result['latency_ms']:.1f} ms")
if result["is_oos"]:
st.warning("flagged OOS") |