Merge pull request #7 from boettiger-lab/feat/update_app

fixing chatbot and bar charts

Author: cassiebuhler

app/app.py

@@ -1,8 +1,12 @@
 import streamlit as st
 from cng.h3 import *
-from ibis import _
 import importlib
 from datetime import time
+import traceback
+import ibis.selectors as s
+from ibis import _
+import ibis 
+import openai
 
 st.set_page_config(layout="wide",
                    page_title="TPL Conservation Almanac",
@@ -50,6 +54,13 @@
         county_choice = 'All'
     st.divider()
 
+with st.sidebar:
+    with st.popover("ℹ️ Help"):
+        st.markdown(help_message)
+    if st.button("🧹 Clear Filters", type="secondary", help = 'Reset all the filters to their default state.'):
+        st.rerun()
+    st.divider()
+
 legend, position, bg_color, fontsize, shape_type, controls = get_legend(paint, leafmap_choice)
 # get all the ids that correspond to the filter
 gdf = filter_data(tpl_table, state_choice, county_choice, year_range)
@@ -65,6 +76,7 @@
             '''
             Mapping queries: 
             - Show me the most expensive protected site
+            - Show me protected areas with high levels of carbon in Florida
             - Show me sites owned, managed or sponsored by the Trust for Public Land
             '''
 
@@ -165,11 +177,11 @@ def run_sql(query, llm_choice):
 
                     # extract ids, columns, bounds if present
                     if "fid" in llm_output.columns and not llm_output.empty:
-                        ids = list(set(llm_output['fid'].tolist()))
+                        unique_ids = list(set(llm_output['fid'].tolist()))
                         llm_cols = extract_columns(sql_query)
-                        bounds = llm_output.total_bounds.tolist()
+                        llm_bounds = llm_output.total_bounds.tolist()
                     else:
-                        ids, llm_cols = [], []
+                        unique_ids, llm_cols = [], []
                         not_mapping = True
 
         except Exception as e:
@@ -196,13 +208,16 @@ def run_sql(query, llm_choice):
 
 # define PMTiles style dict (if we didn't already do so using the chatbot)
 if 'style' not in locals(): 
-    if one_state:
+    if one_state or ('llm_output' in locals()):
         # filter to ids in that state 
         style = tpl_style(unique_ids, paint, pmtiles)
     else: 
         # selected all states, so no need to filter 
         style=tpl_style_default(paint, pmtiles)
-    bounds = get_bounds(state_choice, county_choice, m)
+    if 'llm_output' in locals():
+        bounds = llm_bounds
+    else:
+        bounds = get_bounds(state_choice, county_choice, m)
 
 # add pmtiles to map (using user-specified module)
 if leafmap_choice == "maplibregl":
@@ -277,7 +292,7 @@ def run_sql(query, llm_choice):
     get_bar(gdf_tpl, style_choice, 'year', 'total_amount', paint,'Year','Acquisition Cost ($)',"Yearly investment ($) in protected area")
 
 with col2:
-    gdf_landvote = group_data(gdf_landvote.filter(_.status == 'Pass'), 'Measure Cost')
+    gdf_landvote = group_data(gdf_landvote.filter(gdf_landvote.status == 'Pass'), 'Measure Cost')
     get_bar(gdf_landvote, style_choice, 'year', 'total_amount', paint, 'Year','Funds Approved ($)','Yearly funds from conservation ballot measures')
 
 st.divider()

app/system_prompt.txt

@@ -65,7 +65,19 @@ example_assistant: {{"sql_query":
     LEFT JOIN carbon
     USING (h8)
     GROUP BY "fid", "site", geom"
-    ORDER BY "mean_carbon" DESC LIMIT 10;
+    ORDER BY "mean_carbon" DESC LIMIT 100;
 "explanation":"I joined `conservation_almanac` with `carbon` to retrieve irrecoverable carbon levels for protected areas in the `conservation_almanac`. I returned 10 areas with highest levels of carbon.  
 }}
 
+## Example:
+example_user: "Show me protected areas with high species richness"
+example_assistant: {{"sql_query": 
+    SELECT "fid", "geom", "site", AVG("richness") AS "mean_richness"
+    FROM conservation_almanac
+    LEFT JOIN richness
+    USING (h8)
+    GROUP BY "fid", "site", geom"
+    ORDER BY "mean_richness" DESC LIMIT 100;
+"explanation":"I joined `conservation_almanac` with `richness` to retrieve species richness for protected areas in the `conservation_almanac`. I returned 100 areas with highest levels of carbon.  
+}}
+

app/utils.py

@@ -5,6 +5,8 @@
 import re
 from leafmap.foliumap import PMTilesMapLibreTooltip
 from branca.element import Template
+import pandas as pd
+import datetime
 
 def get_counties(state_selection):
     if state_selection != 'All':
@@ -112,6 +114,14 @@ def tpl_style(ids, paint, pmtiles):
     }
     return style
 
+
+
+    
+def extract_columns(sql_query):
+    # Find all substrings inside double quotes
+    columns = list(dict.fromkeys(re.findall(r'"(.*?)"', sql_query)))
+    return columns
+    
 def get_colorbar(gdf, paint):
     """
     Extracts color hex codes and value range (vmin, vmax) from paint

app/variables.py

@@ -5,7 +5,7 @@
 from cng.utils import *
 from cng.h3 import *
 from minio import Minio
-import streamlit 
+import streamlit as st
 from datetime import datetime, timedelta
 import streamlit
 import re
@@ -16,9 +16,9 @@
 set_secrets(con)
 
 # Get signed URLs to access license-controlled layers
-# key = st.secrets["MINIO_KEY"]
-# secret = st.secrets["MINIO_SECRET"]
-# client = Minio("minio.carlboettiger.info", key, secret)
+key = st.secrets["MINIO_KEY"]
+secret = st.secrets["MINIO_SECRET"]
+client = Minio("minio.carlboettiger.info", key, secret)
 
 pmtiles = "https://minio.carlboettiger.info/public-tpl/conservation_almanac/tpl.pmtiles"
 tpl_z8_url = "https://minio.carlboettiger.info/public-tpl/conservation_almanac/z8/tpl_h3_z8.parquet"
@@ -159,6 +159,11 @@
 #     ['Forestry','Historical','Unknown','Other','Farming','Recreation','Environment','Scenic','RAN'],
 # }
 
+help_message = '''
+- ❌ Safari/iOS not fully supported. For Safari/iOS users, change the **Leafmap module** from MapLibre to Folium in **(Map Settings)** below. 
+- 📊 Use this sidebar to color-code the map by different attributes **(Group by)**
+- 💬 For a more tailored experience, query our dataset of protected areas and their precomputed metrics for each of the displayed layers, using the experimental chatbot. The language model tries to answer natural language questions by drawing only from curated datasets (listed below).
+'''
 
 #maplibregl tooltip 
 tooltip_cols = ['fid','state','site','sponsor','program','county','year','manager',
@@ -249,9 +254,9 @@
     openrouter_api = st.secrets["OPENROUTER_API_KEY"]
 
 llm_options = {
+    "gpt-oss-20b": ChatOpenAI(model = "openai/gpt-oss-20b:free", api_key=openrouter_api, base_url = "https://openrouter.ai/api/v1",  temperature=0),
     "mistral-small-3.2-24b-instruct": ChatOpenAI(model = "mistralai/mistral-small-3.2-24b-instruct:free", api_key=openrouter_api, base_url = "https://openrouter.ai/api/v1",  temperature=0),
     "devstral-small-2505": ChatOpenAI(model = "mistralai/devstral-small-2505:free", api_key=openrouter_api, base_url = "https://openrouter.ai/api/v1",  temperature=0),
-    "gpt-oss-20b": ChatOpenAI(model = "openai/gpt-oss-20b:free", api_key=openrouter_api, base_url = "https://openrouter.ai/api/v1",  temperature=0),
     "deepseek-r1t2-chimera": ChatOpenAI(model = "tngtech/deepseek-r1t2-chimera:free", api_key=openrouter_api, base_url = "https://openrouter.ai/api/v1",  temperature=0),
     "kimi-dev-72b": ChatOpenAI(model = "moonshotai/kimi-dev-72b:free", api_key=openrouter_api, base_url = "https://openrouter.ai/api/v1",  temperature=0),
     "hunyuan-a13b-instruct": ChatOpenAI(model = "tencent/hunyuan-a13b-instruct:free", api_key=openrouter_api, base_url = "https://openrouter.ai/api/v1",  temperature=0),