Upload R_analysis_Romain_PRUDHON.ipynb

R_analysis_Romain_PRUDHON.ipynb

@@ -0,0 +1,518 @@
+{
+  "cells": [
+    {
+      "cell_type": "markdown",
+      "id": "75fd9cc6",
+      "metadata": {
+        "id": "75fd9cc6"
+      },
+      "source": [
+        "# **🤖 Benchmarking & Modeling**"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "id": "fb807724",
+      "metadata": {
+        "id": "fb807724"
+      },
+      "source": [
+        "## **1.** 📦 Setup"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": 6,
+      "id": "d40cd131",
+      "metadata": {
+        "id": "d40cd131",
+        "outputId": "f520f27c-3e9c-470d-c36b-0267ff1c4949",
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        }
+      },
+      "outputs": [
+        {
+          "output_type": "stream",
+          "name": "stderr",
+          "text": [
+            "Installing packages into ‘/usr/local/lib/R/site-library’\n",
+            "(as ‘lib’ is unspecified)\n",
+            "\n",
+            "Warning message in install.packages(c(\"readr\", \"dplyr\", \"stringr\", \"tidyr\", \"lubridate\", :\n",
+            "“installation of package ‘forecast’ had non-zero exit status”\n"
+          ]
+        }
+      ],
+      "source": [
+        "install.packages(c(\n",
+        "  \"readr\",\"dplyr\",\"stringr\",\"tidyr\",\"lubridate\",\n",
+        "  \"ggplot2\",\"forecast\",\"broom\",\"jsonlite\"\n",
+        "), repos=\"https://cloud.r-project.org\")"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "#do above first and then Runtime → Restart session and then do:\n",
+        "library(readr)\n",
+        "library(dplyr)\n",
+        "library(stringr)\n",
+        "library(tidyr)\n",
+        "library(lubridate)\n",
+        "library(ggplot2)\n",
+        "library(forecast)\n",
+        "library(broom)\n",
+        "library(jsonlite)"
+      ],
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "FR_YZUD9J_Py",
+        "outputId": "6b801e29-0053-469f-b4c9-05d8ead19904"
+      },
+      "id": "FR_YZUD9J_Py",
+      "execution_count": 1,
+      "outputs": [
+        {
+          "output_type": "stream",
+          "name": "stderr",
+          "text": [
+            "\n",
+            "Attaching package: ‘dplyr’\n",
+            "\n",
+            "\n",
+            "The following objects are masked from ‘package:stats’:\n",
+            "\n",
+            "    filter, lag\n",
+            "\n",
+            "\n",
+            "The following objects are masked from ‘package:base’:\n",
+            "\n",
+            "    intersect, setdiff, setequal, union\n",
+            "\n",
+            "\n",
+            "\n",
+            "Attaching package: ‘lubridate’\n",
+            "\n",
+            "\n",
+            "The following objects are masked from ‘package:base’:\n",
+            "\n",
+            "    date, intersect, setdiff, union\n",
+            "\n",
+            "\n"
+          ]
+        }
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "id": "f01d02e7",
+      "metadata": {
+        "id": "f01d02e7"
+      },
+      "source": [
+        "## **2.** ✅️ Load & inspect inputs"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": 3,
+      "id": "29e8f6ce",
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "29e8f6ce",
+        "outputId": "9af6e4c9-59d9-4475-a106-beada0b3338c"
+      },
+      "outputs": [
+        {
+          "output_type": "stream",
+          "name": "stdout",
+          "text": [
+            "Loaded: 1000 rows (title-level), 18 rows (monthly)\n"
+          ]
+        }
+      ],
+      "source": [
+        "\n",
+        "must_exist <- function(path, label) {\n",
+        "  if (!file.exists(path)) stop(paste0(\"Missing \", label, \": \", path))\n",
+        "}\n",
+        "\n",
+        "TITLE_PATH <- \"synthetic_title_level_features.csv\"\n",
+        "MONTH_PATH <- \"synthetic_monthly_revenue_series.csv\"\n",
+        "\n",
+        "must_exist(TITLE_PATH, \"TITLE_PATH\")\n",
+        "must_exist(MONTH_PATH, \"MONTH_PATH\")\n",
+        "\n",
+        "df_title <- read_csv(TITLE_PATH, show_col_types = FALSE)\n",
+        "df_month <- read_csv(MONTH_PATH, show_col_types = FALSE)\n",
+        "\n",
+        "cat(\"Loaded:\", nrow(df_title), \"rows (title-level),\", nrow(df_month), \"rows (monthly)\n",
+        "\")"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": 4,
+      "id": "9fd04262",
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "9fd04262",
+        "outputId": "b7c028ba-157c-4d4f-db6c-9098104adec5"
+      },
+      "outputs": [
+        {
+          "output_type": "stream",
+          "name": "stdout",
+          "text": [
+            "\u001b[90m# A tibble: 1 × 6\u001b[39m\n",
+            "      n na_avg_revenue na_price na_rating na_share_pos na_share_neg\n",
+            "  \u001b[3m\u001b[90m<int>\u001b[39m\u001b[23m          \u001b[3m\u001b[90m<int>\u001b[39m\u001b[23m    \u001b[3m\u001b[90m<int>\u001b[39m\u001b[23m     \u001b[3m\u001b[90m<int>\u001b[39m\u001b[23m        \u001b[3m\u001b[90m<int>\u001b[39m\u001b[23m        \u001b[3m\u001b[90m<int>\u001b[39m\u001b[23m\n",
+            "\u001b[90m1\u001b[39m  \u001b[4m1\u001b[24m000              0        0      \u001b[4m1\u001b[24m000            0            0\n",
+            "Monthly rows after parsing: 18 \n"
+          ]
+        }
+      ],
+      "source": [
+        "\n",
+        "# ---------- helpers ----------\n",
+        "safe_num <- function(x) {\n",
+        "  # strips anything that is not digit or dot\n",
+        "  suppressWarnings(as.numeric(str_replace_all(as.character(x), \"[^0-9.]\", \"\")))\n",
+        "}\n",
+        "\n",
+        "parse_rating <- function(x) {\n",
+        "  # Accept: 4, \"4\", \"4.0\", \"4/5\", \"4 out of 5\", \"⭐⭐⭐⭐\", etc.\n",
+        "  x <- as.character(x)\n",
+        "  x <- str_replace_all(x, \"⭐\", \"\")\n",
+        "  x <- str_to_lower(x)\n",
+        "  x <- str_replace_all(x, \"stars?\", \"\")\n",
+        "  x <- str_replace_all(x, \"out of\", \"/\")\n",
+        "  x <- str_replace_all(x, \"\\\\s+\", \"\")\n",
+        "  x <- str_replace_all(x, \"[^0-9./]\", \"\")\n",
+        "  suppressWarnings(as.numeric(str_extract(x, \"^[0-9.]+\")))\n",
+        "}\n",
+        "\n",
+        "parse_month <- function(x) {\n",
+        "  x <- as.character(x)\n",
+        "  # try YYYY-MM-DD, then YYYY-MM\n",
+        "  out <- suppressWarnings(ymd(x))\n",
+        "  if (mean(is.na(out)) > 0.5) out <- suppressWarnings(ymd(paste0(x, \"-01\")))\n",
+        "  na_idx <- which(is.na(out))\n",
+        "  if (length(na_idx) > 0) out[na_idx] <- suppressWarnings(ymd(paste0(x[na_idx], \"-01\")))\n",
+        "  out\n",
+        "}\n",
+        "\n",
+        "# ---------- normalize keys ----------\n",
+        "df_title <- df_title %>% mutate(title = str_squish(as.character(title)))\n",
+        "df_month <- df_month %>% mutate(month = as.character(month))\n",
+        "\n",
+        "# ---------- parse numeric columns defensively ----------\n",
+        "need_cols_title <- c(\"title\",\"avg_revenue\",\"total_revenue\",\"price\",\"rating\",\"share_positive\",\"share_negative\",\"share_neutral\")\n",
+        "missing_title <- setdiff(need_cols_title, names(df_title))\n",
+        "if (length(missing_title) > 0) stop(paste0(\"df_title missing columns: \", paste(missing_title, collapse=\", \")))\n",
+        "\n",
+        "df_title <- df_title %>%\n",
+        "  mutate(\n",
+        "    avg_revenue    = safe_num(avg_revenue),\n",
+        "    total_revenue  = safe_num(total_revenue),\n",
+        "    price          = safe_num(price),\n",
+        "    rating         = parse_rating(rating),\n",
+        "    share_positive = safe_num(share_positive),\n",
+        "    share_negative = safe_num(share_negative),\n",
+        "    share_neutral  = safe_num(share_neutral)\n",
+        "  )\n",
+        "\n",
+        "# basic sanity stats\n",
+        "hyg <- df_title %>%\n",
+        "  summarise(\n",
+        "    n = n(),\n",
+        "    na_avg_revenue = sum(is.na(avg_revenue)),\n",
+        "    na_price = sum(is.na(price)),\n",
+        "    na_rating = sum(is.na(rating)),\n",
+        "    na_share_pos = sum(is.na(share_positive)),\n",
+        "    na_share_neg = sum(is.na(share_negative))\n",
+        "  )\n",
+        "\n",
+        "print(hyg)\n",
+        "\n",
+        "# monthly parsing\n",
+        "need_cols_month <- c(\"month\",\"total_revenue\")\n",
+        "missing_month <- setdiff(need_cols_month, names(df_month))\n",
+        "if (length(missing_month) > 0) stop(paste0(\"df_month missing columns: \", paste(missing_month, collapse=\", \")))\n",
+        "\n",
+        "df_month2 <- df_month %>%\n",
+        "  mutate(\n",
+        "    month = parse_month(month),\n",
+        "    total_revenue = safe_num(total_revenue)\n",
+        "  ) %>%\n",
+        "  filter(!is.na(month)) %>%\n",
+        "  arrange(month)\n",
+        "\n",
+        "cat(\"Monthly rows after parsing:\", nrow(df_month2), \"\\n\")"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "id": "b8971bc4",
+      "metadata": {
+        "id": "b8971bc4"
+      },
+      "source": [
+        "## **3.** 💾 Folder for R outputs for Hugging Face"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": 5,
+      "id": "dfaa06b1",
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "dfaa06b1",
+        "outputId": "85005150-1d3f-4c16-ffbe-ca7b91b379ab"
+      },
+      "outputs": [
+        {
+          "output_type": "stream",
+          "name": "stdout",
+          "text": [
+            "R outputs will be written to: /content/artifacts/r \n"
+          ]
+        }
+      ],
+      "source": [
+        "\n",
+        "ART_DIR  <- \"artifacts\"\n",
+        "R_FIG_DIR <- file.path(ART_DIR, \"r\", \"figures\")\n",
+        "R_TAB_DIR <- file.path(ART_DIR, \"r\", \"tables\")\n",
+        "\n",
+        "dir.create(R_FIG_DIR, recursive = TRUE, showWarnings = FALSE)\n",
+        "dir.create(R_TAB_DIR, recursive = TRUE, showWarnings = FALSE)\n",
+        "\n",
+        "cat(\"R outputs will be written to:\", normalizePath(file.path(ART_DIR, \"r\"), winslash = \"/\"), \"\n",
+        "\")"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "id": "f880c72d",
+      "metadata": {
+        "id": "f880c72d"
+      },
+      "source": [
+        "## **4.** 🔮 Forecast book sales benchmarking with `accuracy()`"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "We benchmark **three** models on a holdout window (last *h* months):\n",
+        "- ARIMA + Fourier (seasonality upgrade)\n",
+        "- ETS\n",
+        "- Naive baseline\n",
+        "\n",
+        "Then we export:\n",
+        "- `accuracy_table.csv`\n",
+        "- `forecast_compare.png`\n",
+        "- `rmse_comparison.png`"
+      ],
+      "metadata": {
+        "id": "R0JZlzKegmzW"
+      },
+      "id": "R0JZlzKegmzW"
+    },
+    {
+      "cell_type": "code",
+      "execution_count": 6,
+      "id": "62e87992",
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/",
+          "height": 82
+        },
+        "id": "62e87992",
+        "outputId": "533ff3fe-2c85-4312-ab04-36007d0cbe1b"
+      },
+      "outputs": [
+        {
+          "output_type": "stream",
+          "name": "stdout",
+          "text": [
+            "✅ Saved: artifacts/r/tables/accuracy_table.csv\n",
+            "✅ Saved: artifacts/r/figures/rmse_comparison.png\n"
+          ]
+        },
+        {
+          "output_type": "display_data",
+          "data": {
+            "text/html": [
+              "<strong>agg_record_805943657:</strong> 2"
+            ],
+            "text/markdown": "**agg_record_805943657:** 2",
+            "text/latex": "\\textbf{agg\\textbackslash{}\\_record\\textbackslash{}\\_805943657:} 2",
+            "text/plain": [
+              "agg_record_805943657 \n",
+              "                   2 "
+            ]
+          },
+          "metadata": {}
+        },
+        {
+          "output_type": "stream",
+          "name": "stdout",
+          "text": [
+            "✅ Saved: artifacts/r/figures/forecast_compare.png\n"
+          ]
+        }
+      ],
+      "source": [
+        "\n",
+        "# Build monthly ts\n",
+        "start_year <- year(min(df_month2$month, na.rm = TRUE))\n",
+        "start_mon  <- month(min(df_month2$month, na.rm = TRUE))\n",
+        "\n",
+        "y <- ts(df_month2$total_revenue, frequency = 12, start = c(start_year, start_mon))\n",
+        "\n",
+        "# holdout size: min(6, 20% of series), at least 1\n",
+        "h_test <- min(6, max(1, floor(length(y) / 5)))\n",
+        "train_ts <- head(y, length(y) - h_test)\n",
+        "test_ts  <- tail(y, h_test)\n",
+        "\n",
+        "# Model A: ARIMA + Fourier\n",
+        "K <- 2\n",
+        "xreg_train <- fourier(train_ts, K = K)\n",
+        "fit_arima <- auto.arima(train_ts, xreg = xreg_train)\n",
+        "xreg_future <- fourier(train_ts, K = K, h = h_test)\n",
+        "fc_arima <- forecast(fit_arima, xreg = xreg_future, h = h_test)\n",
+        "\n",
+        "# Model B: ETS\n",
+        "fit_ets <- ets(train_ts)\n",
+        "fc_ets  <- forecast(fit_ets, h = h_test)\n",
+        "\n",
+        "# Model C: Naive baseline\n",
+        "fc_naive <- naive(train_ts, h = h_test)\n",
+        "\n",
+        "# accuracy() tables\n",
+        "acc_arima <- as.data.frame(accuracy(fc_arima, test_ts))\n",
+        "acc_ets   <- as.data.frame(accuracy(fc_ets, test_ts))\n",
+        "acc_naive <- as.data.frame(accuracy(fc_naive, test_ts))\n",
+        "\n",
+        "accuracy_tbl <- bind_rows(\n",
+        "  acc_arima %>% mutate(model = \"ARIMA+Fourier\"),\n",
+        "  acc_ets   %>% mutate(model = \"ETS\"),\n",
+        "  acc_naive %>% mutate(model = \"Naive\")\n",
+        ") %>% relocate(model)\n",
+        "\n",
+        "write_csv(accuracy_tbl, file.path(R_TAB_DIR, \"accuracy_table.csv\"))\n",
+        "cat(\"✅ Saved: artifacts/r/tables/accuracy_table.csv\\n\")\n",
+        "\n",
+        "# RMSE bar chart\n",
+        "p_rmse <- ggplot(accuracy_tbl, aes(x = reorder(model, RMSE), y = RMSE)) +\n",
+        "  geom_col() +\n",
+        "  coord_flip() +\n",
+        "  labs(title = \"Forecast model comparison (RMSE on holdout)\", x = \"\", y = \"RMSE\") +\n",
+        "  theme_minimal()\n",
+        "\n",
+        "ggsave(file.path(R_FIG_DIR, \"rmse_comparison.png\"), p_rmse, width = 8, height = 4, dpi = 160)\n",
+        "cat(\"✅ Saved: artifacts/r/figures/rmse_comparison.png\\n\")\n",
+        "\n",
+        "# Side-by-side forecast plots (simple, no extra deps)\n",
+        "png(file.path(R_FIG_DIR, \"forecast_compare.png\"), width = 1200, height = 500)\n",
+        "par(mfrow = c(1, 3))\n",
+        "plot(fc_arima, main = \"ARIMA + Fourier\", xlab = \"Time\", ylab = \"Total revenue\"); lines(test_ts, col = \"black\")\n",
+        "plot(fc_ets, main = \"ETS\", xlab = \"Time\", ylab = \"Total revenue\"); lines(test_ts, col = \"black\")\n",
+        "plot(fc_naive, main = \"Naive\", xlab = \"Time\", ylab = \"Total revenue\"); lines(test_ts, col = \"black\")\n",
+        "dev.off()\n",
+        "cat(\"✅ Saved: artifacts/r/figures/forecast_compare.png\\n\")"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "id": "30bc017b",
+      "metadata": {
+        "id": "30bc017b"
+      },
+      "source": [
+        "## **5.** 💾 Some R metadata for Hugging Face"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": 7,
+      "id": "645cb12b",
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "645cb12b",
+        "outputId": "d454e0a5-b986-45ca-9069-65a1c21ce540"
+      },
+      "outputs": [
+        {
+          "output_type": "stream",
+          "name": "stdout",
+          "text": [
+            "✅ Saved: artifacts/r/tables/r_meta.json\n",
+            "DONE. R artifacts written to: artifacts/r \n"
+          ]
+        }
+      ],
+      "source": [
+        "# =========================================================\n",
+        "# Metadata export (aligned with current notebook objects)\n",
+        "# =========================================================\n",
+        "\n",
+        "meta <- list(\n",
+        "\n",
+        "  # ---------------------------\n",
+        "  # Dataset footprint\n",
+        "  # ---------------------------\n",
+        "  n_titles = nrow(df_title),\n",
+        "  n_months = nrow(df_month2),\n",
+        "\n",
+        "  # ---------------------------\n",
+        "  # Forecasting info\n",
+        "  # (only if these objects exist in your forecasting section)\n",
+        "  # ---------------------------\n",
+        "  forecasting = list(\n",
+        "    holdout_h = h_test,\n",
+        "    arima_order = forecast::arimaorder(fit_arima),\n",
+        "    ets_method = fit_ets$method\n",
+        "  )\n",
+        ")\n",
+        "\n",
+        "jsonlite::write_json(\n",
+        "  meta,\n",
+        "  path = file.path(R_TAB_DIR, \"r_meta.json\"),\n",
+        "  pretty = TRUE,\n",
+        "  auto_unbox = TRUE\n",
+        ")\n",
+        "\n",
+        "cat(\"✅ Saved: artifacts/r/tables/r_meta.json\\n\")\n",
+        "cat(\"DONE. R artifacts written to:\", file.path(ART_DIR, \"r\"), \"\\n\")\n"
+      ]
+    }
+  ],
+  "metadata": {
+    "colab": {
+      "provenance": []
+    },
+    "kernelspec": {
+      "name": "ir",
+      "display_name": "R"
+    },
+    "language_info": {
+      "name": "R"
+    }
+  },
+  "nbformat": 4,
+  "nbformat_minor": 5
+}