No public description

publications/a_grammar_of_data_analysis/comparison.R

@@ -0,0 +1,131 @@
+# ==============================================================================
+# comparison.R
+#
+# Description:
+#   This script replicates the data analysis examples from the paper
+#   "A Grammar of Data Analysis" using R and the tidyverse (dplyr, readr, tidyr).
+#   It serves as a comparison to the Python/Meterstick implementations.
+#
+# Examples covered:
+#   1. Baseball Data Analysis (pitch distribution)
+#   2. Churn Rate (A/B test comparison)
+#   3. Difference-in-Differences (minimum wage study)
+#
+# Data Sources:
+#   - Reads from a local 'datasets/' directory by default.
+#   - Can be configured to read directly from GitHub by uncommenting the
+#     relevant lines in each example.
+# ==============================================================================
+
+library(dplyr)
+library(readr)
+library(tidyr)
+
+# ==============================================================================
+# Example 1: Baseball Data Analysis
+# ==============================================================================
+cat("\n--- Running Example 1: Baseball Data Analysis ---\n")
+
+# Read directly from GitHub:
+# df_baseball <- read_csv("https://raw.githubusercontent.com/google/meterstick/refs/heads/master/publications/a_grammar_of_data_analysis/datasets/baseball_pitches_to_SEA_batters.csv")
+# player_names <- read_csv("https://raw.githubusercontent.com/google/meterstick/refs/heads/master/publications/a_grammar_of_data_analysis/datasets/baseball_player_names.csv")
+
+# Or read from a local directory:
+df_baseball <- read_csv("datasets/baseball_pitches_to_SEA_batters.csv")
+player_names <- read_csv("datasets/baseball_player_names.csv")
+
+player_names <- player_names |>
+  mutate(Name = paste(first_name, last_name, sep = " "))
+
+df_baseball <- df_baseball |>
+  left_join(player_names |> select(id, Name), by = c("pitcher_id" = "id"))
+df_baseball <- df_baseball |>
+  rename(Name_pitcher = Name) |>
+  left_join(player_names |> select(id, Name), by = c("batter_id" = "id")) |>
+  rename(Name_batter = Name)
+
+baseball_res <- df_baseball |>
+  group_by(Name_batter, Name_pitcher, pitch_type) |>
+  count() %>%
+  group_by(Name_pitcher, Name_batter) %>%
+  mutate(n / sum(n))
+
+# Show a sample to compare with Python output
+sample_baseball <- baseball_res |>
+  filter(Name_batter == "Dee Gordon", Name_pitcher == "A.J. Cole")
+print(sample_baseball)
+
+
+# ==============================================================================
+# Example 2: Churn Rate (Online Retailer)
+# ==============================================================================
+cat("\n--- Running Example 2: Churn Rate ---\n")
+
+# Read directly from GitHub:
+# df_churn <- read_delim("https://raw.githubusercontent.com/google/meterstick/refs/heads/master/publications/a_grammar_of_data_analysis/datasets/churn_rate.csv", delim = ";")
+
+# Or read from a local directory:
+df_churn <- read_delim("datasets/churn_rate.csv", delim = ";")
+
+df_churn <- df_churn |> mutate(lost = as.logical(lost))
+
+df_by_expt <- df_churn |>
+  group_by(region, experiment) |>
+  summarize(churn = sum(lost, na.rm = TRUE) / n())
+
+df_treated <- df_by_expt |> filter(experiment != "control")
+df_control <- df_by_expt |> filter(experiment == "control")
+
+churn_change <- df_treated |>
+  inner_join(df_control, by = "region", suffix = c("_treated", "_control")) |>
+  mutate(churn_diff = 100 * (churn_treated / churn_control - 1)) |>
+  select(region, experiment = experiment_treated, churn_diff)
+
+print(churn_change)
+
+
+# ==============================================================================
+# Example 3: Difference-in-Differences (Minimum Wage)
+# ==============================================================================
+cat("\n--- Running Example 3: Difference-in-Differences ---\n")
+
+# Read directly from GitHub:
+# df_minwage <- read_delim("https://raw.githubusercontent.com/google/meterstick/refs/heads/master/publications/a_grammar_of_data_analysis/datasets/minimum_wage.csv", delim = ";")
+
+# Or read from a local directory:
+df_minwage <- read_delim("datasets/minimum_wage.csv", delim = ";")
+
+df_minwage <- df_minwage |>
+  mutate(
+    STATE_NAME = ifelse(STATE == 1, "NJ", "PA"),
+    CHAIN = case_when(
+      CHAIN == 1 ~ "Burger King",
+      CHAIN == 2 ~ "KFC",
+      CHAIN == 3 ~ "Roy Rogers",
+      CHAIN == 4 ~ "Wendy's",
+      TRUE ~ as.character(CHAIN)
+    ),
+    EMPTOT = EMPPT * 0.5 + EMPFT + NMGRS,
+    EMPTOT2 = EMPPT2 * 0.5 + EMPFT2 + NMGRS2
+  )
+
+df_long <- df_minwage |>
+  pivot_longer(
+    cols = c(EMPTOT, EMPTOT2),
+    names_to = "PERIOD",
+    values_to = "EMP"
+  ) |>
+  mutate(
+    PERIOD = recode(PERIOD, EMPTOT = "Before", EMPTOT2 = "After")
+  )
+
+did_res <- df_long |>
+  group_by(STATE_NAME, PERIOD) |>
+  summarize(EMP_RATE = mean(EMP, na.rm = TRUE), .groups = "drop") |>
+  pivot_wider(names_from = STATE_NAME, values_from = EMP_RATE) |>
+  mutate(DIFF = NJ - PA) |>
+  select(PERIOD, DIFF) |>
+  pivot_wider(names_from = PERIOD, values_from = DIFF) |>
+  mutate(DIFF_OF_DIFFS = After - Before)
+
+print(did_res)