File size: 3,455 Bytes
82de80f
ddd39d8
 
 
af00309
 
 
a5cf525
 
ddd39d8
 
 
 
 
 
 
aea431b
ddd39d8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
82de80f
ddd39d8
 
 
 
 
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
import streamlit as st
import pandas as pd
import plotly.express as px

from pathlib import Path

with open(Path(__file__).parent / "custom.css") as f:
    st.markdown(f"<style>{f.read()}</style>", unsafe_allow_html=True)

st.set_page_config(layout="wide")
st.title("Thread Pulse")
st.write(
    "This instrument visualizes long-form conversational dynamics using "
    "derived, non-semantic metrics from selected example threads."
)

DATA_DIR = "/app/src/"

FILES = {
    "Anchor": f"{DATA_DIR}/Anchor_turns.csv",
    "Big Flame": f"{DATA_DIR}/BigFlame_turns.csv",
}

# ---- Load selected thread ----
thread_name = st.selectbox("Select thread", list(FILES.keys()))
csv_path = FILES[thread_name]

df = pd.read_csv(csv_path).sort_values("turn")

# ---- Controls ----
roll_window = st.slider("Rolling window (turns)", 5, 150, 25)
show_band = st.checkbox("Show mean ± variance band", value=True)
k = st.slider("Band width (σ multiplier)", 0.5, 3.0, 1.0, 0.5)

scope = st.radio(
    "Compute stability on:",
    ["GPT turns only", "All turns"],
    horizontal=True
)

st.subheader("Stability Detection")
sigma_thresh = st.slider("Stability threshold (σ)", 10.0, 300.0, 80.0, 5.0)
persist_len = st.slider("Required persistence (turns)", 10, 200, 50)

# ---- Choose series for stability stats ----
if scope == "GPT turns only":
    dstat = df[df["speaker"] == "gpt"].copy()
else:
    dstat = df.copy()

dstat["tokens_est"] = pd.to_numeric(dstat["tokens_est"], errors="coerce").fillna(0)

# Rolling mean & std
dstat["roll_mean"] = dstat["tokens_est"].rolling(roll_window, min_periods=1).mean()
dstat["roll_std"]  = dstat["tokens_est"].rolling(roll_window, min_periods=1).std().fillna(0)

# ---- Time-to-stability detection ----
stability_turn = None
roll_std = dstat["roll_std"].to_numpy()
turns = dstat["turn"].to_numpy()

if len(roll_std) > persist_len:
    for i in range(len(roll_std) - persist_len):
        std_slice = roll_std[i:i + persist_len]
        if (std_slice <= sigma_thresh).all():
            stability_turn = int(turns[i])
            break

# Variance band bounds
dstat["upper"] = dstat["roll_mean"] + (k * dstat["roll_std"])
dstat["lower"] = (dstat["roll_mean"] - (k * dstat["roll_std"])).clip(lower=0)

# ---- Plot ----
fig = px.scatter(
    df,
    x="turn",
    y="tokens_est",
    color="speaker",
    opacity=0.6,
    title="Conversation Rhythm",
)

# Rolling mean line
fig.add_scatter(
    x=dstat["turn"],
    y=dstat["roll_mean"],
    mode="lines",
    name=f"Rolling mean ({scope.lower()}, w={roll_window})",
)

# Variance band
if show_band:
    fig.add_scatter(
        x=dstat["turn"],
        y=dstat["lower"],
        mode="lines",
        line=dict(width=0),
        showlegend=False,
        name="Lower bound",
    )
    fig.add_scatter(
        x=dstat["turn"],
        y=dstat["upper"],
        mode="lines",
        fill="tonexty",
        name=f"± {k}σ band",
        opacity=0.2,
    )

# Stability marker line
if stability_turn is not None:
    fig.add_vline(
        x=stability_turn,
        line_dash="dot",
        line_color="green",
        annotation_text="Stability onset",
        annotation_position="top left",
    )

st.plotly_chart(fig, use_container_width=True)

# ---- Report ----
if stability_turn is not None:
    st.success(f"Stability detected at turn {stability_turn} (σ ≤ {sigma_thresh} for {persist_len} turns)")
else:
    st.warning("No stable regime detected under current parameters.")