将月份转换为连续整数：Kernel Regression数据预处理及逆向转换需求

云原生阿岚

2026-5-29

Hey there! Let’s work through this problem together. You’ve got monthly observation counts spanning 2000 to 2018, need to convert those dates to sequential integers (where Jan 2000 = 1) for kernel regression, then map those integers back to dates for plotting. Here’s a straightforward R solution tailored to your needs:

Step 1: Clean and Structure Your Raw Data

First, let’s turn that space-separated raw data into a proper data frame with year, month, and observation count columns:

# Your raw input data
raw_data <- c(2005, 07, 2, 2005, 10, 4, 2005, 12, 2, 2006, 01, 4, 2006, 02, 1, 2006, 07, 2, 2006, 08, 1, 2006, 10, 3)

# Reshape into a 3-column data frame
obs_df <- data.frame(
  year = raw_data[seq(1, length(raw_data), 3)],
  month = raw_data[seq(2, length(raw_data), 3)],
  count = raw_data[seq(3, length(raw_data), 3)]
)

Step 2: Convert Year/Month to a Date Vector

Next, we’ll create a formal date column (using the first day of each month as a consistent anchor):

# Combine year, month, and day (first of the month) into a date
obs_df$date <- as.Date(paste(obs_df$year, obs_df$month, "01", sep = "-"))

Step 3: Generate Sequential Integers for Kernel Regression

We’ll calculate the integer offset from your start date (Jan 2000) so that each month gets a unique sequential number:

# Define your starting date (Jan 2000 = integer 1)
start_date <- as.Date("2000-01-01")

# Install lubridate if you haven't already
# install.packages("lubridate")

# Calculate sequential integer: (year difference * 12) + month difference + 1
obs_df$seq_int <- as.integer(
  (lubridate::year(obs_df$date) - lubridate::year(start_date)) * 12 +
  (lubridate::month(obs_df$date) - lubridate::month(start_date)) + 1
)

# Check the result (Jan 2006 = 73, which makes sense: 6 years *12 =72 +1 =73)
head(obs_df)

Step 4: Map Integers Back to Dates for Plotting

Once you’ve run your kernel regression, you’ll need to convert those sequential integers back to dates for visualization. Here’s a reusable function to do that:

# Function to convert sequential integer to date
int_to_date <- function(seq_int, start_date = as.Date("2000-01-01")) {
  months_since_start <- seq_int - 1
  years <- months_since_start %/% 12
  months <- months_since_start %% 12 + 1
  as.Date(paste(lubridate::year(start_date) + years, months, "01", sep = "-"))
}

# Test it: integer 13 should return Jan 2001
int_to_date(13) # Output: "2001-01-01"

# Example: Convert a full sequence of integers (2000-2018 = 216 months)
pred_seq <- seq(1, 216, 1)
pred_dates <- int_to_date(pred_seq)

Bonus: Plotting Your Regression Results

If you’re using ggplot2 for plotting, here’s how to integrate the converted dates:

# Install ggplot2 if you haven't already
# install.packages("ggplot2")
library(ggplot2)

# Example: Simulate kernel regression predictions (replace with your actual results)
pred_df <- data.frame(
  seq_int = pred_seq,
  pred_count = rnorm(216, mean = 2, sd = 0.5)
)
pred_df$date <- int_to_date(pred_df$seq_int)

# Plot observed data + regression line
ggplot() +
  geom_point(data = obs_df, aes(x = date, y = count), color = "darkblue", alpha = 0.7) +
  geom_line(data = pred_df, aes(x = date, y = pred_count), color = "firebrick", linewidth = 1) +
  labs(x = "Date", y = "Observation Count", title = "Kernel Regression of Monthly Observations") +
  theme_minimal()

内容的提问来源于stack exchange，提问作者Sergio Olalla