# Import the core rpy2 interface to communicate with R from Python
import rpy2.robjects as robjects

# Import the package manager interface from rpy2
# This allows us to load and use R packages directly within Python
from rpy2.robjects.packages import importr

# 1. Initialize R's internal package manager
# 'utils' provides functions like install.packages() in R.
utils = importr('utils')

# 2. Define the packages required for the workflow
# These match the R code you executed: data retrieval, wrangling,
# compositional analysis, structural break detection, Bayesian regression, and visualization.
packages = (
    'tidyquant',     # Financial data retrieval (e.g., Bitcoin prices)
    'tidyverse',     # Data wrangling and transformation
    'compositions',  # Compositional data transformations
    'DirichletReg',  # Dirichlet regression models
    'strucchange',   # Structural break detection
    'MCMCpack',      # Bayesian regression and MCMC methods
    'zoo',           # Rolling window functions for volatility
    'ggplot2'        # Visualization
)

print("Checking R packages...")
for p in packages:
    # Check if the package is already installed in R
    check_cmd = f'"{p}" %in% rownames(installed.packages())'
    is_installed_vector = robjects.r(check_cmd)
    
    # Convert R logical vector to Python boolean
    if is_installed_vector is not None and bool(is_installed_vector[0]):
        print(f"Package '{p}' is already available.")
    else:
        # If not installed, install from CRAN
        print(f"Installing {p} in R. This might take a minute...")
        utils.install_packages(p, repos="https://cloud.r-project.org")

print("\nEnvironment Ready. You can now run your Azure AI Agent code.")

# Import Python's built-in library for interacting with the operating system
import os

# Import httpx, a modern HTTP client for Python
# Used here to handle network requests to Azure OpenAI
import httpx

# Import the Azure OpenAI client
# This allows Python to communicate with Azure-hosted OpenAI models
from openai import AzureOpenAI

# Import the core rpy2 interface to communicate with R from Python
import rpy2.robjects as robjects

# Import pandas2ri for seamless conversion between pandas DataFrames and R data frames
from rpy2.robjects import pandas2ri

# Import localconverter to manage conversion rules between Python and R objects
from rpy2.robjects.conversion import localconverter

# Import display utilities from IPython
# Used to show images (e.g., plots generated by R and saved as PNG files) directly in notebooks
from IPython.display import Image, display

# Connect to Azure OpenAI service
# Configure the client with API version, endpoint, and your API key.
# httpx.Client is used here with SSL verification disabled for simplicity.
client = AzureOpenAI(
    api_version="2025-01-01-preview",
    azure_endpoint="YOUR_URL_ENDPOINT",
    api_key="YOUR_API_KEY",  # replace with your actual key
    http_client=httpx.Client(verify=False, trust_env=False)
)

def run_btc_agent(user_request):
    # Define system instructions for the agent.
    # These instructions enforce raw R code output only.
    # The agent is guided to fetch Bitcoin data, calculate volatility,
    # detect regime shifts, run Bayesian regression, and visualize results.
    system_instructions = (
    "You are a Financial Modeling Agent. Output must be raw R code only, with no explanations, "
    "and never use markdown fences, language tags, or commentary. "
    "Your task is to fetch Bitcoin prices for the last 3 years, calculate volatility, detect regime shifts, "
    "and run Bayesian regression using MCMCpack. Follow these rules:\n"
    "1. Load libraries: tidyquant, tidyverse, compositions, DirichletReg, strucchange, MCMCpack, zoo, ggplot2.\n"
    "2. Fetch Bitcoin data with tq_get('BTC-USD', from = Sys.Date() - years(3), to = Sys.Date()).\n"
    "3. Preprocess: select date and close, rename close -> price, compute returns, "
    "calculate volatility with rollapply(sd), multiply by 100 to express as percentage, drop NA values.\n"
    "4. Regime analysis: use breakpoints(volatility ~ 1, data = btc_data), assign regimes with breakfactor().\n"
    "5. Bayesian regression: run MCMCregress(price ~ regime, data = btc_data) only if regime has at least 2 levels.\n"
    "6. Visualization: create a dual-axis plot with ggplot2. "
    "   • Left y-axis: price (blue line, currency format with dollar sign).\n"
    "   • Right y-axis: volatility (red line, percentage format).\n"
    "   • Rescale volatility for plotting: use btc_data$volatility explicitly in max() calls to avoid closure errors.\n"
    "   • Example: geom_line(aes(y = volatility * max(btc_data$price)/max(btc_data$volatility, na.rm=TRUE))).\n"
    "   • Use scale_y_continuous(name='Price (USD)', labels=scales::dollar_format(), "
    "sec.axis = sec_axis(~ . * max(btc_data$volatility, na.rm=TRUE)/max(btc_data$price), name='Volatility (%)')).\n"
    "   • Lines must be thicker and visually clear.\n"
    "   • Axis text and labels must be bold and bigger, Roboto Slab font.\n"
    "   • Axis titles must be color-coded: blue for Price, red for Volatility. "
    "     Use theme(axis.title.y = element_text(color='blue', face='bold'), "
    "axis.title.y.right = element_text(color='red', face='bold')).\n"
    "   • Title: 'Bitcoin Price & Volatility Analysis'.\n"
    "   • Remove x axis title.\n"
    "   • Remove legend with theme(legend.position = 'none').\n"
    "7. Export: save plot with ggsave('btc_model_plot.png', width=10, height=6, dpi=300)"
)

    # Send the request to Azure OpenAI with system + user instructions
    response = client.chat.completions.create(
        model="gpt-4o",  # name of your deployed model in Azure
        messages=[
            {"role": "system", "content": system_instructions},
            {"role": "user", "content": user_request}
        ]
    )

    # Extract the agent’s R code output
    agent_code = response.choices[0].message.content.strip()
    if agent_code.startswith("```"):
        agent_code = "\n".join(agent_code.split("\n")[1:-1])

    print("-" * 40)
    print(agent_code)
    print("-" * 40)

    try:
        # Execute the R code if it looks valid
        if "library(" in agent_code or "<-" in agent_code:
            with localconverter(robjects.default_converter + pandas2ri.converter):
                cwd = os.getcwd().replace("\\", "/")
                robjects.r(f'setwd(\"{cwd}\")')
                robjects.r(agent_code.strip())

                # Display the generated plot if it exists
                if os.path.exists("btc_model_plot.png"):
                    display(Image(filename="btc_model_plot.png"))
        else:
            print("Skipping execution: output does not look like valid R code.")
    except Exception as e:
        print(f"Agent Error: {e}")

# Run the agent with a specific request
run_btc_agent("Model Bitcoin prices for the last 3 years and provide a volatility analysis.")