App.py

import os
from langchain_community.utilities import OpenWeatherMapAPIWrapper
import streamlit as st
import vertexai
from langchain_google_vertexai import VertexAI
from langchain.prompts import PromptTemplate
from langchain.chains import LLMChain, SequentialChain
import numpy as np
import requests
import deepl
import time

os.environ["GOOGLE_APPLICATION_CREDENTIALS"]="key.json" # place the key JSON file in the same folder as your notebook
PROJECT_ID = "" # use your project id 
REGION = ""  #
BUCKET_URI = ""  # create your own bucket

# PROJECT_ID = "genai-and-lllm" # use your project id
# REGION = "us-central1"  #
# BUCKET_URI = f"gs://gen-ai-storage-bucket-for-class"  # create your own bucket


vertexai.init(project=PROJECT_ID, location=REGION, staging_bucket=BUCKET_URI)

llm1 = VertexAI(
    model_name="text-bison@001",
    max_output_tokens=256,
    temperature=0.5,
    top_p=0.8,
    top_k=40,
    verbose=True,
)

llm2 = VertexAI(
    model_name="gemini-pro",
    max_output_tokens=256,
    temperature=0.5,
    top_p=0.8,
    top_k=40,
    verbose=True,
)

deepL_key = "379fa3bc-7867-46d9-bace-750f6fd673b3:fx"
translator = deepl.Translator(deepL_key)

os.environ["OPENWEATHERMAP_API_KEY"] = '5708941459d96133b31f54f2f15bd9aa'

weather = OpenWeatherMapAPIWrapper()

#weather_data = weather.run("Columbus")


st.sidebar.image('weather_logo.png')
st.sidebar.title("Weather Worldwide")

###############
# Sidebar for Country selection
countries_and_cities = {
    'United States': ['Pittsburgh', 'Boston',  'Chicago', 'Dallas', 'Los Angeles', 'Miami', 'New York', 'Seattle', 'Use this to test errors from API :)'],
    'Canada': ['Jasper', 'Toronto', 'Vancouver', 'Montreal'],
    'United Kingdom': ['London', 'Manchester', 'Liverpool', 'Birmingham'],
    'Japan': ['Tokyo', 'Osaka', 'Kyoto', 'Sapporo'],
    'France': ['Paris', 'Lyon', 'Marseille', 'Nice'],
    'Spain': ['Madrid', 'Barcelona', 'Valencia', 'Seville'],
    'India': ['New Delhi', 'Mumbai', 'Bengaluru', 'Kolkata']
}
languages = ['BG - Bulgarian', 'CS - Czech', 'DA - Danish', 'DE - German', 'EL - Greek', 'EN-GB - English (British)', 
             'EN-US - English (American)', 'ES - Spanish', 'ET - Estonian', 'FI - Finnish', 'FR - French', 'HU - Hungarian', 'ID - Indonesian', 
             'IT - Italian', 'JA - Japanese', 'KO - Korean', 'LT - Lithuanian', 'LV - Latvian', 'NB - Norwegian (Bokmål)', 'NL - Dutch', 
             'PL - Polish', 'PT - Portuguese (unspecified variant for backward compatibility; please select PT-BR or PT-PT instead)', 
             'PT-BR - Portuguese (Brazilian)', 'PT-PT - Portuguese (all Portuguese varieties excluding Brazilian Portuguese)', 
             'RO - Romanian', 'RU - Russian', 'SK - Slovak', 'SL - Slovenian', 'SV - Swedish', 'TR - Turkish', 'UK - Ukrainian', 
             'ZH - Chinese (simplified)']

st.sidebar.header("Please select your location:")
country = st.sidebar.selectbox('Select a country:', list(countries_and_cities.keys()))
cities = countries_and_cities[country]
city = st.sidebar.selectbox('Select a city:', cities)
###############
#Select Weather Boxes
st.header("Please select your weather interest:")
Temperature_Box = st.checkbox(label="Temperature (Required)", value=True, disabled=True)
col1, col2 = st.columns(2)
Humidity_Box = col1.checkbox(label="Humidity") 
Cloud_Cover_Box = col1.checkbox(label="Cloud Cover")
Precipitation_Box = col2.checkbox(label="Precipitation")
Wind_Box = col2.checkbox(label="Wind Speed/Direction")
Lang_Box = st.sidebar.selectbox(label="Language", options=languages, index=6)
##############
#Select Model
st.sidebar.header("Change the model (optional):")
LLM_choice = st.sidebar.selectbox("Model choice:", [ "Gemini Pro", "Text Bison"])


#The following runs after the user selects their desired info. By default, they will get the Temp in New York, USA (prepopulated)
generate_result = st.button("Give me the Weather details!")
if generate_result:
    weather_data = None

    #Send API call based on selected country/city. If the try doesn't work with the API, send error message to user
    try:
        weather_data = weather.run(f"{city},{country}")
    # Process weather_data
    except Exception as e:
        st.error(f"\u26A0 An error occurred with the weather API: \"{e}\"  \n\nWe apologize for the inconvenience!  \nPlease try again later.")

    #If an exception was triggered above then weather_data will still be none and this won't run
    if weather_data:
        with st.spinner(f"Getting weather for {city}, {country} in {Lang_Box}."):
            time.sleep(8)
        st.success('Done!')
        # st.write(f"Getting weather for {city}, {country} in {Lang_Box}.")

        # Parse what the user wants to see using mask
        user_wants = np.array([Temperature_Box, Humidity_Box, Cloud_Cover_Box, Precipitation_Box, Wind_Box])
        strings= np.array(["Temperature", "Humidity", "Cloud Cover", "Precipitation", "Wind Speed/Direction"])
        selected_strings = strings[user_wants]


        #Depending on what they want, we join the strings correctly to pass to LLM in a coherent sentence
        if len(selected_strings) > 2:
            output = ", ".join(selected_strings[:-1]) + " and " + selected_strings[-1] #if we have more than 2 we want to add "and" for the last
        elif len(selected_strings) == 2:
            output = " and ".join(selected_strings) #if we have only 2 we want to add "and" 
        else:
            output = selected_strings[0] # do nothing 

        #Provided the sentence from above, we pass it to the prompt template 
        prompt_template_weather = PromptTemplate(
            input_variables=['output', 'weather_data'],
            template= 'The user wants to know the {output} from the following: {weather_data}.'
            )
            
        
        #Now we access our chain
        if LLM_choice =="Text Bison":
            llm=llm1
        else:
            llm=llm2

        chain = LLMChain(llm=llm, prompt=prompt_template_weather) #Create our chain and print to Streamlit
        result = chain.run(output=output, weather_data=weather_data)
        if Lang_Box[:2] == "EN" or Lang_Box[:2] == "PT":
            target = Lang_Box[:5]
        else:
            target = Lang_Box[:2]
            
        result = translator.translate_text(text = result, target_lang=target, preserve_formatting=True)
        st.write(result)