app.py

# app.py  —  San José Crash Safety Explorer
import os
import streamlit as st
import pandas as pd
import requests
import altair as alt
import pydeck as pdk
from datetime import datetime, timedelta

### If CrashMap is not working on the local system- Uncomment the belwo Mapbox demo token

# if not os.getenv("MAPBOX_API_KEY"):
#     pdk.settings.mapbox_api_key = (
#         "pk.eyJ1IjoibWFwYm94IiwiYSI6ImNpejY4Nm44NTA4emYycXBndHRqbnB2N3gifQ."
#         "w5iJj8yOTFoNUG0GI_LhwA"
#     )

### CONSTANTS & DATA HELPERS
RESOURCE_ID = "15408d78-9734-4ea1-b3e5-a0f99568dd9b"
BASE_URL    = "https://data.sanjoseca.gov/api/3/action/datastore_search"
PAGE_LIMIT  = 1_000

@st.cache_data(ttl=3_600, show_spinner=False)
def fetch_data() -> pd.DataFrame:
    records, offset = [], 0
    while True:
        params = {"resource_id": RESOURCE_ID, "limit": PAGE_LIMIT, "offset": offset}
        try:
            batch = requests.get(BASE_URL, params=params, timeout=20).json()["result"]["records"]
        except Exception:
            st.error("Cannot reach the San José Open‑Data API.")
            return pd.DataFrame(columns=["CrashDateTime"])
        if not batch:
            break
        records.extend(batch)
        offset += PAGE_LIMIT
    return pd.DataFrame.from_records(records)

def preprocess(df: pd.DataFrame) -> pd.DataFrame:
    if df.empty:
        return df
    df["CrashDateTime"] = pd.to_datetime(df["CrashDateTime"], errors="coerce")
    df["Latitude"]      = pd.to_numeric(df["Latitude"],  errors="coerce")
    df["Longitude"]     = pd.to_numeric(df["Longitude"], errors="coerce")
    for col in ["MinorInjuries", "ModerateInjuries", "SevereInjuries", "FatalInjuries"]:
        df[col] = pd.to_numeric(df[col], errors="coerce").fillna(0)
    df["SpeedingFlag"]   = df["SpeedingFlag"].fillna("Unknown")
    df["Weather"]        = df["Weather"].fillna("Unknown")
    df["RoadwaySurface"] = df["RoadwaySurface"].fillna("Unknown")
    df["CrashTime"]      = df["CrashDateTime"].dt.strftime("%Y‑%m‑%d %H:%M")
    return df.dropna(subset=["CrashDateTime"]).copy()

def time_filter(df: pd.DataFrame, months: str) -> pd.DataFrame:
    if months == "All" or df.empty:
        return df
    end = datetime.now()
    start = end - timedelta(days=int(months) * 30)
    return df[(df["CrashDateTime"] >= start) & (df["CrashDateTime"] <= end)]

### PAGE HEADER, MOTIVATION, DATASET INFO, REFRESH
st.set_page_config("San José Crash Dashboard", layout="wide", page_icon="🚦")

st.title("🚦 San José Crash Safety Explorer")

st.markdown("""
### Motivation  
San José’s ambitious **Vision Zero** program aims to eliminate traffic fatalities and serious injuries.  
Real‑time insights into *when, where, and why* crashes occur help planners, engineers, and residents target the most effective counter‑measures.
""")

left, right = st.columns([5, 1])
with left:
    st.markdown("""
**Dataset**: [City of San José Crashes Data](https://data.sanjoseca.gov/)  
Updated nightly · 60 + attributes per crash (location, weather, speeding flag, injury counts, etc.)
""")
with right:
    if st.button("🔁 Refresh"):
        st.cache_data.clear()

### FILTERS
months = st.selectbox(
    "Select time range to explore",
    ["6", "9", "12", "All"],
    index=3,
    format_func=lambda m: f"{m} months" if m != "All" else "All time"
)
st.divider()

### DATA LOAD
df = time_filter(preprocess(fetch_data()), months)
if df.empty:
    st.warning("No data available for the selected period.")
    st.stop()

df["year"]          = df["CrashDateTime"].dt.year
df["TotalInjuries"] = df[["MinorInjuries", "ModerateInjuries",
                          "SevereInjuries", "FatalInjuries"]].sum(axis=1)

def two_col_layout(chart, narrative_md):
    vis, story = st.columns([3, 2], gap="large")
    with vis:
        st.altair_chart(chart, use_container_width=True)
    with story:
        st.markdown(narrative_md, unsafe_allow_html=True)

### NAVIGATION TABS
tabs = st.tabs([
    "Weather & Injuries",
    "Speeding Trend",
    "Crash Map",
    "Road Surface",
    "Injury Correlation"
])

# ───────── TAB 1 — WEATHER × INJURIES ─────────
with tabs[0]:
    w_df = (
        df.groupby("Weather")[["MinorInjuries", "ModerateInjuries",
                               "SevereInjuries", "FatalInjuries"]]
          .sum().reset_index()
          .rename(columns={"MinorInjuries":"Minor","ModerateInjuries":"Moderate",
                           "SevereInjuries":"Severe","FatalInjuries":"Fatal"})
    )
    w_long = w_df.melt("Weather", var_name="InjuryType", value_name="Count")
    sel_type = alt.selection_point(fields=["InjuryType"], bind="legend")
    sel_weather = alt.selection_point(fields=["Weather"], toggle="event")
    w_chart = (
        alt.Chart(w_long)
           .mark_bar()
           .encode(
               x="Weather:N",
               y="Count:Q",
               color="InjuryType:N",
               opacity=alt.condition(sel_type & sel_weather, alt.value(1), alt.value(0.25)),
               tooltip=["Weather","InjuryType","Count"])
           .add_params(sel_type, sel_weather)
           .properties(height=380)
    )
    two_col_layout(
        w_chart,
        """
### Weather‑Linked Injury Patterns  
*Rain & drizzle* elevate **moderate injuries**, while fatal crashes stay steady—indicating speed, not just weather, drives lethality.

**Use for** seasonal safety messaging, wet‑weather friction treatments, and enforcement surge planning.
"""
    )

# ───────── TAB 2 — SPEEDING TREND ─────────
with tabs[1]:
    s_df = df.groupby(["year","SpeedingFlag"]).size().reset_index(name="Count")
    sel_flag = alt.selection_point(fields=["SpeedingFlag"], bind="legend")
    s_chart = (
        alt.Chart(s_df)
           .mark_bar(size=22)
           .encode(
               x="year:O",
               y="Count:Q",
               color="SpeedingFlag:N",
               opacity=alt.condition(sel_flag, alt.value(1), alt.value(0.25)),
               tooltip=["year","SpeedingFlag","Count"])
           .add_params(sel_flag)
           .properties(height=380)
    )
    two_col_layout(
        s_chart,
        """
### Speeding‑Related Collisions Over Time  
Red bars show crashes **flagged for speeding**. A mild downward trend since 2020 suggests progress from speed‑limit reductions and targeted enforcement.

**Actionable insight** – focus radar enforcement on outlier years or sudden upticks.
"""
    )

# ───────── TAB 3 — CRASH MAP ─────────
with tabs[2]:
    # st.header("Crash Locations in San José")
    # st.markdown("Zoom in to spot areas with high crash density.")

    map_df = df.dropna(subset=["Latitude", "Longitude"])
    if map_df.empty:
        st.info("No geocoded crashes for this period.")
    else:
        deck = pdk.Deck(
            map_style="mapbox://styles/mapbox/light-v9",
            initial_view_state=pdk.ViewState(
                latitude=map_df["Latitude"].mean(),
                longitude=map_df["Longitude"].mean(),
                zoom=11,
                pitch=50,
            ),
            layers=[
                pdk.Layer(
                    "ScatterplotLayer",
                    data=map_df,
                    get_position="[Longitude, Latitude]",
                    get_radius=40,
                    get_fill_color="[200, 30, 0, 160]",
                    pickable=True,
                )
            ],
            tooltip={"html": "<b>📍 Crash</b><br/>{CrashTime}<br/>Total injuries: {TotalInjuries}"}
        )
        vis, story = st.columns([3, 2], gap="large")
        with vis:
            st.pydeck_chart(deck, use_container_width=True)
        with story:
            st.markdown("""
### City‑wide Crash Hotspots  
Pins cluster along freeway ramps and busy arterials.  
Hover to reveal timestamp and injury tally; zoom for intersection‑scale detail.

*Guides infrastructure fixes – protected turns, lane re‑striping, signal timing.*
""")

# ───────── TAB 4 — ROAD SURFACE ─────────
with tabs[3]:
    surf_df = df.groupby("RoadwaySurface").size().reset_index(name="Count")
    sel_surf = alt.selection_point(fields=["RoadwaySurface"], bind="legend")
    surf_chart = (
        alt.Chart(surf_df)
           .mark_bar()
           .encode(
               x="Count:Q",
               y=alt.Y("RoadwaySurface:N", sort="-x"),
               color="RoadwaySurface:N",
               opacity=alt.condition(sel_surf, alt.value(1), alt.value(0.25)),
               tooltip=["RoadwaySurface","Count"])
           .add_params(sel_surf)
           .properties(height=380)
    )
    two_col_layout(
        surf_chart,
        """
### Surface Condition & Crash Frequency  
While asphalt dominates totals, **wet/oily** pavement poses disproportionate risk.

**Maintenance cue** – schedule surface re‑texturing or drainage upgrades right after early‑season rains.
"""
    )

# ───────── TAB 5 — INJURY CORRELATION ─────────
with tabs[4]:
    # st.header("Correlation Among Injury Stats")
    # st.markdown("Numeric relationships between injury types and year.")

    num_df = df[["MinorInjuries", "ModerateInjuries", "SevereInjuries",
                 "FatalInjuries", "TotalInjuries"]]
    corr = num_df.corr().round(2)
    corr_df = corr.reset_index().melt(id_vars="index")

    heatmap = (
        alt.Chart(corr_df)
           .mark_rect()
           .encode(
               x=alt.X("index:N", title=None),
               y=alt.Y("variable:N", title=None),
               color=alt.Color("value:Q", scale=alt.Scale(scheme="redblue")),
               tooltip=["index", "variable", "value"]
           )
           .properties(height=450)
    )

    two_col_layout(
        heatmap,
        """
### Injury Metric Inter‑Relationships  
Heat‑map reveals how injury categories co‑vary with each other and with **year**.

* Strong red = positive correlation (metrics rise together).  
* Strong blue = negative correlation (one rises as the other falls).  
* Near‑white = little or no relationship.

Use this matrix to spot redundant metrics or unexpected links worth deeper investigation.
"""
    )

### CONCLUSION & FOOTER
st.divider()
st.markdown("""
## Conclusion  
Leveraging open data and lightweight web visualisation, this dashboard pinpoints **where**, **when**, and **under what conditions** San José crashes occur.  
Insights support:

* **Vision Zero engineering** – identify corridors for protected turns, lane diets, and friction treatment.  
* **Targeted enforcement** – deploy speed emphasis patrols where speeding spikes.  
* **Public outreach** – craft seasonal safety messages (e.g., first‑rain reminders).

Continued monitoring keeps stakeholders aligned on progress toward the city’s goal of *zero traffic deaths*.
""")
st.caption("© City of San José Open‑Data • Dashboard for IS 445 / Group2 - Aniruddha Jenifer Radha Sanidhya Smruti")