#' Time Series Analysis (Smart Aggregation for Date & Numeric)
#'
#' @param dataset Dataset name
#' @param var_num The numeric variable to analyze (Y)
#' @param var_date The time variable (X). Can be Date or Numeric.
#' @param agg_level Aggregation level ("day", "week", "month", "quarter", "year")
#' @param agg_fun Aggregation function ("mean", "sum", "count")
#' @param ma_n Window size for Moving Average
#' @param fc_periods Number of periods to forecast
#' @param methods Vector of methods: "ma", "ets", "acf"
#' @param data_filter Filter expression
#' @param envir Environment
#'
#' @return A list of results
#' @export
time_series <- function(dataset, var_num, var_date = NULL,
                        agg_level = "month", agg_fun = "mean",
                        ma_n = 3, fc_periods = 0, methods = c("ma"),
                        data_filter = "", envir = parent.frame()) {
  
  # === 1. 数据准备 ===
  vars <- c(var_num)
  if (!is.null(var_date) && var_date != "") vars <- c(vars, var_date)
  
  df_raw <- get_data(dataset, vars, filt = data_filter, na.rm = TRUE, envir = envir)
  
  # === 2. 智能聚合逻辑 ===
  # 检查 X 轴数据类型
  is_date_col <- !is.null(var_date) && var_date != "" && inherits(df_raw[[var_date]], c("Date", "POSIXt"))
  is_num_col  <- !is.null(var_date) && var_date != "" && is.numeric(df_raw[[var_date]])
  
  df <- df_raw # 默认回退
  
  if (is_date_col || is_num_col) {
    requireNamespace("dplyr")
    
    if (is_date_col) {
      # A. 如果是日期：使用 lubridate 对齐
      requireNamespace("lubridate")
      df_raw$ts_group_col <- lubridate::floor_date(df_raw[[var_date]], unit = agg_level)
      step_num <- 1 # 仅用于后续判断
      
    } else if (is_num_col) {
      # B. 如果是数字：执行分箱 (Binning)
      # 将 "Month" 映射为 "30个单位", "Week" -> "7个单位"
      step_num <- switch(agg_level, 
                         "week" = 7, 
                         "month" = 30, 
                         "quarter" = 90, 
                         "year" = 365, 
                         1) # default day=1
      
      # 核心算法：向下取整分箱
      df_raw$ts_group_col <- floor(df_raw[[var_date]] / step_num) * step_num
    }
    
    # 执行聚合
    df <- df_raw %>%
      dplyr::group_by(ts_group_col) %>%
      dplyr::summarise(
        y_agg = if (agg_fun == "count") dplyr::n() 
        else if (agg_fun == "sum") sum(.data[[var_num]], na.rm = TRUE)
        else mean(.data[[var_num]], na.rm = TRUE)
      ) %>%
      dplyr::arrange(ts_group_col) %>%
      dplyr::rename(!!var_date := ts_group_col, !!var_num := y_agg) %>%
      as.data.frame()
    
  } else {
    # 如果没选变量，生成 Index
    if (is.null(var_date) || var_date == "") {
      var_date <- "Index"
      df$Index <- 1:nrow(df)
      step_num <- 1
    }
  }
  
  y_val <- df[[var_num]]
  res_list <- list()
  metrics <- data.frame(Method = character(), RMSE = numeric(), MAE = numeric(), stringsAsFactors = FALSE)
  
  get_accuracy <- function(actual, fitted, name) {
    err <- actual - fitted
    c(name, round(sqrt(mean(err^2, na.rm=TRUE)), 3), round(mean(abs(err), na.rm=TRUE), 3))
  }
  
  # === 3. 计算模块 ===
  
  # --- Moving Average ---
  if ("ma" %in% methods) {
    # 确保窗口不大于数据长度
    safe_ma_n <- min(ma_n, length(y_val))
    ma_weights <- rep(1/safe_ma_n, safe_ma_n)
    df$ma_line <- as.numeric(stats::filter(y_val, ma_weights, sides = 1))
    
    valid <- !is.na(df$ma_line)
    if (sum(valid) > 0) {
      metrics[nrow(metrics)+1,] <- get_accuracy(y_val[valid], df$ma_line[valid], "Moving Average")
    }
  }
  
  # --- ETS (Holt-Winters) ---
  if ("ets" %in% methods) {
    # 尝试根据聚合粒度设定频率
    freq <- 1
    if (is_date_col) {
      freq <- switch(agg_level, "month"=12, "quarter"=4, "week"=52, 7)
    } else if (is_num_col) {
      # 对于数值型，尝试推断周期，默认无季节性
      freq <- 1 
    }
    
    if (length(y_val) < 2 * freq) freq <- 1
    ts_obj <- ts(y_val, frequency = freq)
    
    ets_model <- tryCatch(HoltWinters(ts_obj), 
                          error = function(e) tryCatch(HoltWinters(ts_obj, gamma=FALSE), error = function(e) NULL))
    
    if (!is.null(ets_model)) {
      fitted_raw <- as.numeric(ets_model$fitted[, 1])
      pad <- rep(NA, length(y_val) - length(fitted_raw))
      df$ets_line <- c(pad, fitted_raw)
      
      # 记录参数
      res_list$ets_params <- list(alpha = ets_model$alpha, beta = ets_model$beta, gamma = ets_model$gamma)
      
      # 记录误差
      metrics[nrow(metrics)+1,] <- get_accuracy(y_val[!is.na(df$ets_line)], df$ets_line[!is.na(df$ets_line)], "Exp. Smoothing")
      
      # === 预测 (Forecast) ===
      if (fc_periods > 0) {
        fc_out <- predict(ets_model, n.ahead = fc_periods, prediction.interval = TRUE, level = 0.95)
        
        last_time <- tail(df[[var_date]], 1)
        
        # 生成未来时间轴 (区分日期和数字)
        if (inherits(last_time, "Date") || inherits(last_time, "POSIXt")) {
          future_dates <- seq(from = last_time, by = agg_level, length.out = fc_periods + 1)[-1]
        } else {
          # 对于数字，按 step_num 递增
          # 如果之前判定是数字列，使用之前的 step_num
          step <- if(exists("step_num")) step_num else 1
          future_dates <- seq(from = last_time + step, by = step, length.out = fc_periods)
        }
        
        res_list$forecast <- data.frame(
          date = future_dates,
          fit = as.numeric(fc_out[, "fit"]),
          upr = as.numeric(fc_out[, "upr"]),
          lwr = as.numeric(fc_out[, "lwr"])
        )
      }
    }
  }
  
  # --- ACF ---
  if ("acf" %in% methods) {
    # 增加 tryCatch 防止数据太少报错
    acf_out <- tryCatch(acf(y_val, plot = FALSE, na.action = na.pass), error = function(e) NULL)
    if (!is.null(acf_out)) {
      res_list$acf_df <- data.frame(lag = acf_out$lag, acf = acf_out$acf)
    }
  }
  
  # === 4. 返回 ===
  res_list$dataset <- df
  res_list$metrics <- metrics
  res_list$var_num <- var_num
  res_list$var_date <- var_date
  res_list$agg_level <- agg_level
  res_list$agg_fun <- agg_fun
  res_list$methods <- methods
  res_list$df_name <- if (is_string(dataset)) dataset else deparse(substitute(dataset))
  
  as.list(res_list) %>% add_class("time_series")
}

#' Summary method
#' @export
summary.time_series <- function(object, ...) {
  cat("Time Series Analysis\n")
  cat("Data      :", object$df_name, "\n")
  cat("Variable  :", object$var_num, "\n")
  cat("Time Var  :", object$var_date, "\n")
  cat("Aggregation:", object$agg_level, "(", object$agg_fun, ")\n\n")
  
  if (nrow(object$metrics) > 0) {
    cat("Model Accuracy (Goodness of Fit):\n")
    print(object$metrics, row.names = FALSE)
    cat("---\n")
  }
  
  if ("ets" %in% object$methods && !is.null(object$ets_params)) {
    p <- object$ets_params
    cat("Exp. Smoothing Parameters:\n")
    cat(sprintf("Alpha: %.3f, Beta: %.3f, Gamma: %.3f\n", p$alpha, ifelse(is.null(p$beta), 0, p$beta), ifelse(is.null(p$gamma), 0, p$gamma)))
  }
  
  if (!is.null(object$forecast)) {
    cat("\nForecast (Next", nrow(object$forecast), "periods):\n")
    print(head(object$forecast, 5))
  }
}

#' Plot method
#' @export
plot.time_series <- function(x, plots = c("original", "ma", "ets"), shiny = FALSE, custom = FALSE, ...) {
  
  plot_list <- list()
  
  if (any(c("original", "ma", "ets") %in% plots)) {
    
    p <- ggplot(x$dataset, aes(x = .data[[x$var_date]], y = .data[[x$var_num]])) +
      labs(title = paste("Time Series:", x$var_num), subtitle = paste("Aggregated by", x$agg_level)) +
      theme_minimal()
    
    if ("original" %in% plots) {
      p <- p + geom_line(color = "black", alpha = 0.3) + 
        geom_point(color = "black", alpha = 0.3, size = 1.5)
    }
    
    if ("ma" %in% plots && "ma_line" %in% names(x$dataset)) {
      p <- p + geom_line(aes(y = ma_line, color = "Moving Avg"), linewidth = 0.8)
    }
    
    if ("ets" %in% plots && "ets_line" %in% names(x$dataset)) {
      p <- p + geom_line(aes(y = ets_line, color = "Exp Smoothing"), linewidth = 0.8)
    }
    
    if ("ets" %in% plots && !is.null(x$forecast)) {
      p <- p + 
        geom_ribbon(data = x$forecast, aes(x = date, ymin = lwr, ymax = upr, y = fit), fill = "red", alpha = 0.15, inherit.aes = FALSE) +
        geom_line(data = x$forecast, aes(x = date, y = fit, color = "Forecast"), linetype = "dashed", linewidth = 0.8, inherit.aes = FALSE)
    }
    
    p <- p + scale_color_manual(name = "Legend", 
                                values = c("Moving Avg" = "blue", "Exp Smoothing" = "red", "Forecast" = "red"))
    
    plot_list[[1]] <- p
  }
  
  if ("acf" %in% plots && "acf" %in% x$methods && !is.null(x$acf_df)) {
    ci <- 0.95
    clim <- qnorm((1 + ci)/2) / sqrt(nrow(x$dataset))
    p_acf <- ggplot(x$acf_df, aes(x = lag, y = acf)) +
      geom_segment(aes(xend = lag, yend = 0), color = "darkblue", linewidth = 1) +
      geom_hline(yintercept = c(clim, -clim), linetype = "dashed", color = "blue") +
      geom_hline(yintercept = 0) +
      labs(title = "Autocorrelation (ACF)", y = "ACF", x = "Lag") + theme_minimal()
    plot_list[[length(plot_list) + 1]] <- p_acf
  }
  
  if (length(plot_list) > 0) {
    patchwork::wrap_plots(plot_list, ncol = 1) %>%
      (function(obj) if (shiny) obj else print(obj))
  }
}