More work on scorer
This commit is contained in:
@@ -0,0 +1,2 @@
|
|||||||
|
[theme]
|
||||||
|
base="light"
|
||||||
+122
-29
@@ -29,16 +29,18 @@ def load_data(file_path):
|
|||||||
return data
|
return data
|
||||||
|
|
||||||
def save_data(file_path, data):
|
def save_data(file_path, data):
|
||||||
"""Save the updated data back to JSONL file."""
|
|
||||||
with open(file_path, "w", encoding="utf-8") as f:
|
with open(file_path, "w", encoding="utf-8") as f:
|
||||||
for entry in data:
|
for entry in data:
|
||||||
for o in entry.get("output", []):
|
for o in entry.get("output", []):
|
||||||
if "content_parsed" in o:
|
if "content_parsed" in o:
|
||||||
o["content"] = json.dumps(o["content_parsed"], ensure_ascii=False)
|
o["content"] = json.dumps(
|
||||||
|
o["content_parsed"],
|
||||||
|
ensure_ascii=False
|
||||||
|
)
|
||||||
f.write(json.dumps(entry, ensure_ascii=False) + "\n")
|
f.write(json.dumps(entry, ensure_ascii=False) + "\n")
|
||||||
|
|
||||||
# --------------------------
|
# --------------------------
|
||||||
# Initialize session state
|
# Session State Init
|
||||||
# --------------------------
|
# --------------------------
|
||||||
if "data" not in st.session_state:
|
if "data" not in st.session_state:
|
||||||
st.session_state.data = load_data(DATA_FILE)
|
st.session_state.data = load_data(DATA_FILE)
|
||||||
@@ -46,35 +48,49 @@ if "data" not in st.session_state:
|
|||||||
if "current_claim" not in st.session_state:
|
if "current_claim" not in st.session_state:
|
||||||
st.session_state.current_claim = None
|
st.session_state.current_claim = None
|
||||||
|
|
||||||
|
if "drag_order" not in st.session_state:
|
||||||
|
st.session_state.drag_order = None
|
||||||
|
|
||||||
|
st.set_page_config(
|
||||||
|
page_title="Claim Visualizer",
|
||||||
|
layout="wide",
|
||||||
|
initial_sidebar_state="expanded"
|
||||||
|
)
|
||||||
st.title("Claim Visualizer")
|
st.title("Claim Visualizer")
|
||||||
|
|
||||||
# --------------------------
|
# --------------------------
|
||||||
# Sidebar
|
# Sidebar
|
||||||
# --------------------------
|
# --------------------------
|
||||||
view = st.sidebar.selectbox("Choose View", ["All Claims", "Single Claim Random"])
|
view = st.sidebar.selectbox(
|
||||||
|
"Choose View",
|
||||||
|
["All Claims", "Single Claim Random", "View Rules"]
|
||||||
|
)
|
||||||
|
|
||||||
# --------------------------
|
# --------------------------
|
||||||
# All Claims View
|
# ALL CLAIMS VIEW
|
||||||
# --------------------------
|
# --------------------------
|
||||||
if view == "All Claims":
|
if view == "All Claims":
|
||||||
st.header("All Claims")
|
st.header("All Claims")
|
||||||
for entry in st.session_state.data:
|
for entry in st.session_state.data:
|
||||||
st.subheader(f"{entry.get('text')}")
|
st.subheader(entry.get("text"))
|
||||||
|
|
||||||
for o in entry.get("output", []):
|
for o in entry.get("output", []):
|
||||||
for c in o.get("content_parsed", []):
|
for c in o.get("content_parsed", []):
|
||||||
st.markdown(f"**Event:** {c.get('event')}")
|
st.markdown(f"**Event:** {c.get('event')}")
|
||||||
st.markdown(f"**Reasoning:** {c.get('reasoningWhyRelevant')}")
|
st.markdown(f"**Reasoning:** {c.get('reasoningWhyRelevant')}")
|
||||||
st.markdown(f"**Score:** {c.get('score')}")
|
st.markdown(f"**Score:** {c.get('score')}")
|
||||||
st.markdown(f"**Human Score:** {c.get('human_score')}")
|
st.markdown(f"**Human Score:** {c.get('human_score')}")
|
||||||
|
st.markdown(f"**Extra Info:** {c.get('extra_info', '')}")
|
||||||
st.markdown("---")
|
st.markdown("---")
|
||||||
|
|
||||||
# --------------------------
|
# --------------------------
|
||||||
# Single Claim Random View (Ranking Based)
|
# SINGLE CLAIM RANDOM VIEW
|
||||||
# --------------------------
|
# --------------------------
|
||||||
elif view == "Single Claim Random":
|
|
||||||
|
|
||||||
# Select an entry that still has unscored items
|
elif view == "Single Claim Random":
|
||||||
|
# Select new entry if needed
|
||||||
if st.session_state.current_claim is None:
|
if st.session_state.current_claim is None:
|
||||||
|
|
||||||
unscored_entries = []
|
unscored_entries = []
|
||||||
|
|
||||||
for entry in st.session_state.data:
|
for entry in st.session_state.data:
|
||||||
@@ -93,6 +109,7 @@ elif view == "Single Claim Random":
|
|||||||
|
|
||||||
if unscored_entries:
|
if unscored_entries:
|
||||||
st.session_state.current_claim = random.choice(unscored_entries)
|
st.session_state.current_claim = random.choice(unscored_entries)
|
||||||
|
st.session_state.drag_order = None
|
||||||
else:
|
else:
|
||||||
st.session_state.current_claim = None
|
st.session_state.current_claim = None
|
||||||
|
|
||||||
@@ -106,43 +123,119 @@ elif view == "Single Claim Random":
|
|||||||
|
|
||||||
st.subheader(entry.get("text"))
|
st.subheader(entry.get("text"))
|
||||||
|
|
||||||
st.write("Rank events (1 = best / most relevant)")
|
# --------------------------
|
||||||
|
# Stable Drag IDs (FIX)
|
||||||
|
# --------------------------
|
||||||
|
|
||||||
rankings = []
|
claim_ids = [str(i) for i in range(len(claims))]
|
||||||
|
|
||||||
|
# Initialize order only once
|
||||||
|
if (
|
||||||
|
st.session_state.drag_order is None
|
||||||
|
or len(st.session_state.drag_order) != len(claim_ids)
|
||||||
|
):
|
||||||
|
st.session_state.drag_order = claim_ids.copy()
|
||||||
|
|
||||||
|
ordered_indices = [
|
||||||
|
int(i) for i in st.session_state.drag_order
|
||||||
|
]
|
||||||
|
|
||||||
|
# --------------------------
|
||||||
|
# Annotation Section
|
||||||
|
# --------------------------
|
||||||
|
|
||||||
|
st.subheader("Annotate Events")
|
||||||
|
|
||||||
|
for pos, idx in enumerate(ordered_indices):
|
||||||
|
c = claims[idx]
|
||||||
|
|
||||||
|
with st.container(border=True):
|
||||||
|
|
||||||
# Collect rankings
|
|
||||||
for i, c in enumerate(claims):
|
|
||||||
st.markdown(f"### Event {i+1}")
|
|
||||||
st.markdown(f"**Event:** {c.get('event')}")
|
st.markdown(f"**Event:** {c.get('event')}")
|
||||||
st.markdown(f"**Reasoning:** {c.get('reasoningWhyRelevant')}")
|
st.markdown(
|
||||||
|
f"**Reasoning:** {c.get('reasoningWhyRelevant')}"
|
||||||
rank = st.number_input(
|
|
||||||
f"Rank for event {i+1}",
|
|
||||||
min_value=1,
|
|
||||||
max_value=len(claims),
|
|
||||||
key=f"rank_{i}"
|
|
||||||
)
|
)
|
||||||
|
|
||||||
rankings.append((c, rank))
|
cols = st.columns(5)
|
||||||
|
|
||||||
|
temp = ""
|
||||||
|
|
||||||
|
with cols[0]:
|
||||||
|
a = st.checkbox("Rewording", key = "R" + str(idx) + c.get('event') )
|
||||||
|
temp += "REWORDING " if a else ""
|
||||||
|
|
||||||
|
with cols[1]:
|
||||||
|
a = st.checkbox("Not Specific", key = "S" + str(idx) + c.get('event') )
|
||||||
|
temp += "NSPECIFIC " if a else ""
|
||||||
|
|
||||||
|
with cols[2]:
|
||||||
|
a = st.checkbox("Time Incorrect", key = "T" + str(idx) + c.get('event') )
|
||||||
|
temp += "TINCORRECT " if a else ""
|
||||||
|
|
||||||
|
with cols[3]:
|
||||||
|
a = st.checkbox("Story?", key = "Y" + str(idx) + c.get('event') )
|
||||||
|
temp += "STORY " if a else ""
|
||||||
|
|
||||||
|
with cols[4]:
|
||||||
|
a = st.checkbox("Perfect", key = "P" + str(idx) + c.get('event') )
|
||||||
|
temp += "PERFECT " if a else ""
|
||||||
|
|
||||||
|
c["extra_info"] = temp
|
||||||
|
|
||||||
|
# ---- MOVE BUTTONS ----
|
||||||
|
move_cols = st.columns(2)
|
||||||
|
|
||||||
|
with move_cols[0]:
|
||||||
|
if st.button(
|
||||||
|
"Up",
|
||||||
|
key="UP" + str(idx) + c.get("event")
|
||||||
|
):
|
||||||
|
if pos > 0:
|
||||||
|
order = st.session_state.drag_order
|
||||||
|
order[pos], order[pos - 1] = order[pos - 1], order[pos]
|
||||||
|
st.session_state.drag_order = order
|
||||||
|
st.rerun()
|
||||||
|
|
||||||
|
with move_cols[1]:
|
||||||
|
if st.button(
|
||||||
|
"Down",
|
||||||
|
key="DOWN" + str(idx) + c.get("event")
|
||||||
|
):
|
||||||
|
if pos < len(st.session_state.drag_order) - 1:
|
||||||
|
order = st.session_state.drag_order
|
||||||
|
order[pos], order[pos + 1] = order[pos + 1], order[pos]
|
||||||
|
st.session_state.drag_order = order
|
||||||
|
st.rerun()
|
||||||
|
|
||||||
|
|
||||||
|
# --------------------------
|
||||||
|
# Submit Ranking
|
||||||
|
# --------------------------
|
||||||
|
|
||||||
if st.button("Submit Ranking"):
|
if st.button("Submit Ranking"):
|
||||||
|
|
||||||
# Sort by rank (ascending = best)
|
n = len(ordered_indices)
|
||||||
rankings.sort(key=lambda x: x[1])
|
|
||||||
|
|
||||||
n = len(rankings)
|
for rank_position, idx in enumerate(ordered_indices):
|
||||||
|
|
||||||
|
claim_obj = claims[idx]
|
||||||
|
|
||||||
# Convert ranking -> normalized score
|
|
||||||
for idx, (claim_obj, _) in enumerate(rankings):
|
|
||||||
if n == 1:
|
if n == 1:
|
||||||
score = 1.0
|
score = 1.0
|
||||||
else:
|
else:
|
||||||
score = 1 - (idx / (n - 1))
|
score = 1 - (rank_position / (n - 1))
|
||||||
|
|
||||||
claim_obj["human_score"] = round(score, 3)
|
claim_obj["human_score"] = round(score, 3)
|
||||||
|
|
||||||
save_data(DATA_FILE, st.session_state.data)
|
save_data(DATA_FILE, st.session_state.data)
|
||||||
st.success("Ranking converted to scores and saved!")
|
|
||||||
|
print("Ranking converted to scores and saved!")
|
||||||
|
|
||||||
st.session_state.current_claim = None
|
st.session_state.current_claim = None
|
||||||
|
st.session_state.drag_order = None
|
||||||
|
|
||||||
st.rerun()
|
st.rerun()
|
||||||
|
|
||||||
|
elif view == "View Rules":
|
||||||
|
with open("rules.txt", "r", encoding="utf-8") as f:
|
||||||
|
st.write(f.readlines())
|
||||||
|
|||||||
@@ -1,4 +1,4 @@
|
|||||||
1. Proposed trigger events rewording the initial prompt should penalised
|
1. Proposed trigger events rewording the initial prompt should penalised
|
||||||
2. Proposed trigger events should be specific enough to create resonable analysis
|
2. Proposed trigger events should be specific enough to create resonable analysis
|
||||||
3. Proposed trigger events should be time-specific. A trigger events that happened after the event is invalid
|
3. Proposed trigger events should be time-specific. A trigger events that happened after the event is invalid
|
||||||
4. Proposed trigger events should not describe a story of propogation
|
4. Proposed trigger events should not describe a story of propogation (discussions on debunk articles is invalid)
|
||||||
Reference in New Issue
Block a user