19arjun89's picture
Update app.py
2c6a8fa verified
import os
import json
from collections import Counter
import math
import gradio as gr
import pandas as pd
import plotly.express as px
import pycountry
from datasets import load_dataset
# =========================
# Config
# =========================
VISITS_URL = os.getenv(
"VISITS_URL",
"https://huggingface.co/datasets/19arjun89/ai_recruiting_agent_usage/resolve/main/usage/visits_enriched.jsonl",
)
# Set this as a HF Space SECRET named MAPBOX_TOKEN
MAPBOX_TOKEN = os.getenv("MAPBOX_TOKEN", "").strip()
# Path to your GeoJSON (commit into the Space repo)
GEOJSON_PATH = os.getenv("GEOJSON_PATH", "countries.geojson")
# IMPORTANT: Set this to match the property name inside your GeoJSON features.
# Common values: "properties.ISO_A3" or "properties.ADM0_A3"
GEOJSON_FEATURE_ID_KEY = "properties.ISO3166-1-Alpha-3"
MAX_ROWS = int(os.getenv("MAX_ROWS", "500000"))
# =========================
# Helpers
# =========================
def normalize_country_name(country: str | None) -> str | None:
if not country or not isinstance(country, str):
return None
c = country.strip()
if not c or c.lower() == "unknown":
return None
return c
def iso2_to_iso3(country_code: str | None) -> str | None:
"""Convert ISO-2 -> ISO-3 for map matching."""
if not country_code or not isinstance(country_code, str):
return None
c2 = country_code.strip().upper()
if len(c2) != 2:
return None
try:
rec = pycountry.countries.get(alpha_2=c2)
return rec.alpha_3 if rec else None
except Exception:
return None
def load_rows_streaming():
ds = load_dataset(
"json",
data_files=VISITS_URL,
split="train",
streaming=True,
)
for i, row in enumerate(ds):
yield row
if i + 1 >= MAX_ROWS:
break
def load_geojson(path: str) -> dict:
with open(path, "r", encoding="utf-8") as f:
return json.load(f)
def patch_geojson_iso_codes(countries_geojson: dict) -> dict:
"""
Some GeoJSON files have ISO codes as '-99'. Patch them using the 'name' field.
Updates:
properties['ISO3166-1-Alpha-2']
properties['ISO3166-1-Alpha-3']
"""
fixed = 0
for feat in countries_geojson.get("features", []):
props = feat.get("properties", {}) or {}
iso3 = str(props.get("ISO3166-1-Alpha-3", "") or "").strip()
iso2 = str(props.get("ISO3166-1-Alpha-2", "") or "").strip()
name = str(props.get("name", "") or "").strip()
needs_fix = (iso3 == "-99" or iso2 == "-99" or not iso3 or not iso2)
if not needs_fix or not name:
continue
try:
rec = pycountry.countries.search_fuzzy(name)[0]
props["ISO3166-1-Alpha-3"] = rec.alpha_3
props["ISO3166-1-Alpha-2"] = rec.alpha_2
fixed += 1
except Exception:
# leave as-is if we can't resolve
pass
print(f"DEBUG patched GeoJSON features: {fixed}")
return countries_geojson
# =========================
# Main report builder
# =========================
def build_report():
if not MAPBOX_TOKEN:
# We can still run, but Mapbox will not render nicely without token.
# We'll still build a figure (it may appear blank/limited).
pass
countries_geojson = patch_geojson_iso_codes(load_geojson(GEOJSON_PATH))
# Counters for clean reconciliation
scanned = 0
skipped_session_start = 0
missing_country = 0
invalid_country_code = 0
# Table (country name) and map (iso3)
country_counts = Counter()
iso3_counts = Counter()
iso3_to_name = {}
for row in load_rows_streaming():
scanned += 1
event_type = str(row.get("event", "") or "").strip().lower()
if event_type == "session_start":
skipped_session_start += 1
continue
country = normalize_country_name(row.get("final_country"))
if not country:
missing_country += 1
continue
# Count it for the table FIRST (all usage events with a valid country name)
country_counts[country] += 1
iso3 = iso2_to_iso3(row.get("final_country_code"))
if not iso3:
invalid_country_code += 1
continue
# Count it for the map only (requires ISO3)
iso3_counts[iso3] += 1
iso3_to_name.setdefault(iso3, country)
# Build table dataframe
table_df = (
pd.DataFrame([{"country": k, "usage events": v} for k, v in country_counts.items()])
.sort_values("usage events", ascending=False)
.reset_index(drop=True)
)
# Build map dataframe
map_df = (
pd.DataFrame(
[
{"iso3": iso3, "country": iso3_to_name.get(iso3, iso3), "usage events": cnt}
for iso3, cnt in iso3_counts.items()
]
)
.sort_values("usage events", ascending=False)
.reset_index(drop=True)
)
# Log scale for nicer color spread (keeps small countries visible)
map_df["usage_log"] = map_df["usage events"].clip(lower=1).apply(lambda x: math.log10(x))
# Reconciliation
rows_mappable = int(map_df["usage events"].sum()) # note: this is TOTAL events, not rows
mappable_rows_count = int(sum(iso3_counts.values())) # count of rows after filters (events counted)
table_rows_counted = int(sum(country_counts.values()))
accounted = skipped_session_start + missing_country + invalid_country_code + mappable_rows_count
# If you want “Rows mappable” to mean “rows that made it to map”, use mappable_rows_count
# If you want “Total usage events” (same thing here), use table_df sum.
# Map figure
if map_df.empty:
fig = px.scatter(title="No mappable data found")
fig.update_layout(height=740, margin=dict(l=0, r=0, t=40, b=0))
summary = (
f"Rows scanned: {scanned:,}\n"
f"- Rows counted in table: {table_rows_counted:,}\n"
f"- Rows mapped: {mappable_rows_count:,}\n"
f"- Session starts skipped: {skipped_session_start:,}\n"
f"- Missing country: {missing_country:,}\n"
f"- Invalid country code: {invalid_country_code:,}\n\n"
f"Accounted rows: {accounted:,} / {scanned:,}\n"
f"Countries (table): {len(table_df):,}\n"
f"Total usage events: {int(table_df['usage events'].sum()) if len(table_df) else 0:,}"
)
return fig, table_df.head(50), summary
# Mapbox choropleth using GeoJSON
px.set_mapbox_access_token(MAPBOX_TOKEN)
map_df["iso3"] = map_df["iso3"].astype(str).str.upper()
fig = px.choropleth_mapbox(
map_df,
geojson=countries_geojson,
featureidkey=GEOJSON_FEATURE_ID_KEY,
locations="iso3",
color="usage_log",
hover_name="country",
hover_data={"usage events": True, "iso3": True},
labels={"usage_log": "Usage intensity (log10)", "usage events": "Usage events"},
mapbox_style="open-street-map",
opacity=0.75,
zoom=1.5,
center={"lat": 15, "lon": 0},
)
fig.update_traces(
# Use a clean hover card
hovertemplate=(
"<b>%{customdata[0]}</b><br>"
"Usage events: %{customdata[1]:,}<br>"
"<extra></extra>"
),
# customdata lets us show real counts even though color is log-scaled
customdata=map_df[["country", "usage events"]].to_numpy(),
)
fig.update_traces(
marker_line_width=0.8,
marker_line_color="rgba(255,255,255,0.85)", # nice on light basemaps
)
# Full-bleed layout
fig.update_layout(
height=740,
margin=dict(l=0, r=0, t=0, b=0),
)
# Dashboard title
fig.add_annotation(
text="Usage Events by Country",
x=0.01,
y=0.95,
xref="paper",
yref="paper",
xanchor="left",
yanchor="top",
showarrow=False,
font=dict(size=20),
)
fig.update_layout(coloraxis_showscale=False)
# Summary text (clean math)
summary = (
f"Rows scanned: {scanned:,}\n"
f"- Session starts skipped: {skipped_session_start:,}\n"
f"- Missing country: {missing_country:,}\n"
f"- Invalid country code: {invalid_country_code:,}\n"
f"- Rows mapped: {mappable_rows_count:,}\n\n"
f"Accounted rows: {accounted:,} / {scanned:,}\n"
f"Countries (table): {len(table_df):,}\n"
f"Countries (map): {map_df['iso3'].nunique():,}\n"
f"Total usage events: {int(table_df['usage events'].sum()) if len(table_df) else 0:,}"
)
table_out = table_df.head(50).copy()
table_out.insert(0, "refreshed_at_utc", pd.Timestamp.utcnow().strftime("%Y-%m-%d %H:%M:%S"))
return fig, table_out, summary
# =========================
# UI
# =========================
with gr.Blocks(title="AI Recruiting Agent — Usage Map") as demo:
gr.Markdown(
"# AI Recruiting Agent — Usage by Country (Mapbox)\n"
"This Space reads **only** `visits_enriched.jsonl`, excludes `event=session_start`, "
"and plots **usage events** by country.\n\n"
)
run = gr.Button("Generate map")
summary = gr.Markdown()
plot = gr.Plot()
table = gr.Dataframe(label="Top countries", interactive=False)
run.click(
fn=build_report,
inputs=[],
outputs=[plot, table, summary],
)
demo.launch()