|
| 1 | +{ |
| 2 | + "cells": [ |
| 3 | + { |
| 4 | + "cell_type": "markdown", |
| 5 | + "id": "0", |
| 6 | + "metadata": { |
| 7 | + "marimo": { |
| 8 | + "config": { |
| 9 | + "hide_code": true |
| 10 | + } |
| 11 | + } |
| 12 | + }, |
| 13 | + "source": [ |
| 14 | + "# Motivation" |
| 15 | + ] |
| 16 | + }, |
| 17 | + { |
| 18 | + "cell_type": "code", |
| 19 | + "execution_count": null, |
| 20 | + "id": "1", |
| 21 | + "metadata": {}, |
| 22 | + "outputs": [], |
| 23 | + "source": [ |
| 24 | + "from datetime import datetime\n", |
| 25 | + "\n", |
| 26 | + "import pandas as pd\n", |
| 27 | + "\n", |
| 28 | + "df = pd.DataFrame(\n", |
| 29 | + " {\n", |
| 30 | + " \"date\": [datetime(2020, 1, 1), datetime(2020, 1, 8), datetime(2020, 2, 3)],\n", |
| 31 | + " \"price\": [1, 4, 3],\n", |
| 32 | + " }\n", |
| 33 | + ")\n", |
| 34 | + "df" |
| 35 | + ] |
| 36 | + }, |
| 37 | + { |
| 38 | + "cell_type": "code", |
| 39 | + "execution_count": null, |
| 40 | + "id": "2", |
| 41 | + "metadata": {}, |
| 42 | + "outputs": [], |
| 43 | + "source": [ |
| 44 | + "def monthly_aggregate_pandas(user_df):\n", |
| 45 | + " return user_df.resample(\"MS\", on=\"date\")[[\"price\"]].mean()\n", |
| 46 | + "\n", |
| 47 | + "monthly_aggregate_pandas(df)" |
| 48 | + ] |
| 49 | + }, |
| 50 | + { |
| 51 | + "cell_type": "markdown", |
| 52 | + "id": "3", |
| 53 | + "metadata": { |
| 54 | + "marimo": { |
| 55 | + "config": { |
| 56 | + "hide_code": true |
| 57 | + } |
| 58 | + } |
| 59 | + }, |
| 60 | + "source": [ |
| 61 | + "# Dataframe-agnostic data science\n", |
| 62 | + "\n", |
| 63 | + "\n", |
| 64 | + "## Bad solution: just convert to pandas" |
| 65 | + ] |
| 66 | + }, |
| 67 | + { |
| 68 | + "cell_type": "code", |
| 69 | + "execution_count": null, |
| 70 | + "id": "4", |
| 71 | + "metadata": {}, |
| 72 | + "outputs": [], |
| 73 | + "source": [ |
| 74 | + "import duckdb\n", |
| 75 | + "import polars as pl\n", |
| 76 | + "import pyarrow as pa\n", |
| 77 | + "import pyspark\n", |
| 78 | + "import pyspark.sql.functions as F\n", |
| 79 | + "from pyspark.sql import SparkSession\n" |
| 80 | + ] |
| 81 | + }, |
| 82 | + { |
| 83 | + "cell_type": "code", |
| 84 | + "execution_count": null, |
| 85 | + "id": "5", |
| 86 | + "metadata": {}, |
| 87 | + "outputs": [], |
| 88 | + "source": [ |
| 89 | + "def monthly_aggregate_bad(user_df):\n", |
| 90 | + " if isinstance(user_df, pd.DataFrame):\n", |
| 91 | + " df = user_df\n", |
| 92 | + " elif isinstance(user_df, pl.DataFrame):\n", |
| 93 | + " df = user_df.to_pandas()\n", |
| 94 | + " elif isinstance(user_df, duckdb.DuckDBPyRelation):\n", |
| 95 | + " df = user_df.df()\n", |
| 96 | + " elif isinstance(user_df, pa.Table):\n", |
| 97 | + " df = user_df.to_pandas()\n", |
| 98 | + " elif isinstance(user_df, pyspark.sql.dataframe.DataFrame):\n", |
| 99 | + " df = user_df.toPandas()\n", |
| 100 | + " else:\n", |
| 101 | + " raise TypeError(\"Unsupported DataFrame type: cannot convert to pandas\")\n", |
| 102 | + "\n", |
| 103 | + " return df.resample(\"MS\", on=\"date\")[[\"price\"]].mean()\n" |
| 104 | + ] |
| 105 | + }, |
| 106 | + { |
| 107 | + "cell_type": "code", |
| 108 | + "execution_count": null, |
| 109 | + "id": "6", |
| 110 | + "metadata": {}, |
| 111 | + "outputs": [], |
| 112 | + "source": [ |
| 113 | + "data = {\n", |
| 114 | + " \"date\": [datetime(2020, 1, 1), datetime(2020, 1, 8), datetime(2020, 2, 3)],\n", |
| 115 | + " \"price\": [1, 4, 3],\n", |
| 116 | + "}" |
| 117 | + ] |
| 118 | + }, |
| 119 | + { |
| 120 | + "cell_type": "code", |
| 121 | + "execution_count": null, |
| 122 | + "id": "7", |
| 123 | + "metadata": {}, |
| 124 | + "outputs": [], |
| 125 | + "source": [ |
| 126 | + "# pandas\n", |
| 127 | + "pandas_df = pd.DataFrame(data)\n", |
| 128 | + "monthly_aggregate_bad(pandas_df)\n", |
| 129 | + "\n", |
| 130 | + "# polars\n", |
| 131 | + "polars_df = pl.DataFrame(data)\n", |
| 132 | + "monthly_aggregate_bad(polars_df)\n", |
| 133 | + "\n", |
| 134 | + "# duckdb\n", |
| 135 | + "duckdb_df = duckdb.from_df(pandas_df)\n", |
| 136 | + "monthly_aggregate_bad(duckdb_df)\n", |
| 137 | + "\n", |
| 138 | + "# pyspark\n", |
| 139 | + "spark = SparkSession.builder.getOrCreate()\n", |
| 140 | + "spark_df = spark.createDataFrame(pandas_df)\n", |
| 141 | + "monthly_aggregate_bad(spark_df)\n", |
| 142 | + "\n", |
| 143 | + "# pyarrow\n", |
| 144 | + "arrow_table = pa.table(data)\n", |
| 145 | + "monthly_aggregate_bad(arrow_table)" |
| 146 | + ] |
| 147 | + }, |
| 148 | + { |
| 149 | + "cell_type": "markdown", |
| 150 | + "id": "8", |
| 151 | + "metadata": { |
| 152 | + "marimo": { |
| 153 | + "config": { |
| 154 | + "hide_code": true |
| 155 | + } |
| 156 | + } |
| 157 | + }, |
| 158 | + "source": [ |
| 159 | + "## Unmaintainable solution: different branches for each library" |
| 160 | + ] |
| 161 | + }, |
| 162 | + { |
| 163 | + "cell_type": "code", |
| 164 | + "execution_count": null, |
| 165 | + "id": "9", |
| 166 | + "metadata": {}, |
| 167 | + "outputs": [], |
| 168 | + "source": [ |
| 169 | + "def monthly_aggregate_unmaintainable(user_df):\n", |
| 170 | + " if isinstance(user_df, pd.DataFrame):\n", |
| 171 | + " result = user_df.resample(\"MS\", on=\"date\")[[\"price\"]].mean()\n", |
| 172 | + " elif isinstance(user_df, pl.DataFrame):\n", |
| 173 | + " result = (\n", |
| 174 | + " user_df.group_by(pl.col(\"date\").dt.truncate(\"1mo\"))\n", |
| 175 | + " .agg(pl.col(\"price\").mean())\n", |
| 176 | + " .sort(\"date\")\n", |
| 177 | + " )\n", |
| 178 | + " elif isinstance(user_df, pyspark.sql.dataframe.DataFrame):\n", |
| 179 | + " result = (\n", |
| 180 | + " user_df.withColumn(\"date_month\", F.date_trunc(\"month\", F.col(\"date\")))\n", |
| 181 | + " .groupBy(\"date_month\")\n", |
| 182 | + " .agg(F.mean(\"price\").alias(\"price_mean\"))\n", |
| 183 | + " .orderBy(\"date_month\")\n", |
| 184 | + " )\n", |
| 185 | + " # TODO: more branches for DuckDB, PyArrow, Dask, etc... :sob:\n", |
| 186 | + " return result\n" |
| 187 | + ] |
| 188 | + }, |
| 189 | + { |
| 190 | + "cell_type": "code", |
| 191 | + "execution_count": null, |
| 192 | + "id": "10", |
| 193 | + "metadata": {}, |
| 194 | + "outputs": [], |
| 195 | + "source": [ |
| 196 | + "# pandas\n", |
| 197 | + "monthly_aggregate_unmaintainable(pandas_df)\n", |
| 198 | + "\n", |
| 199 | + "# polars\n", |
| 200 | + "monthly_aggregate_unmaintainable(polars_df)\n", |
| 201 | + "\n", |
| 202 | + "# pyspark\n", |
| 203 | + "monthly_aggregate_unmaintainable(spark_df)" |
| 204 | + ] |
| 205 | + }, |
| 206 | + { |
| 207 | + "cell_type": "markdown", |
| 208 | + "id": "11", |
| 209 | + "metadata": { |
| 210 | + "marimo": { |
| 211 | + "config": { |
| 212 | + "hide_code": true |
| 213 | + } |
| 214 | + } |
| 215 | + }, |
| 216 | + "source": [ |
| 217 | + "## Best solution: Narwhals as a unified dataframe interface" |
| 218 | + ] |
| 219 | + }, |
| 220 | + { |
| 221 | + "cell_type": "code", |
| 222 | + "execution_count": null, |
| 223 | + "id": "12", |
| 224 | + "metadata": {}, |
| 225 | + "outputs": [], |
| 226 | + "source": [ |
| 227 | + "import narwhals as nw\n", |
| 228 | + "from narwhals.typing import IntoFrameT\n", |
| 229 | + "\n", |
| 230 | + "\n", |
| 231 | + "def monthly_aggregate(user_df: IntoFrameT) -> IntoFrameT:\n", |
| 232 | + " return (\n", |
| 233 | + " nw.from_native(user_df)\n", |
| 234 | + " .group_by(nw.col(\"date\").dt.truncate(\"1mo\"))\n", |
| 235 | + " .agg(nw.col(\"price\").mean())\n", |
| 236 | + " .sort(\"date\")\n", |
| 237 | + " .to_native()\n", |
| 238 | + " )\n" |
| 239 | + ] |
| 240 | + }, |
| 241 | + { |
| 242 | + "cell_type": "code", |
| 243 | + "execution_count": null, |
| 244 | + "id": "13", |
| 245 | + "metadata": {}, |
| 246 | + "outputs": [], |
| 247 | + "source": [ |
| 248 | + "# pandas\n", |
| 249 | + "monthly_aggregate(pandas_df)\n", |
| 250 | + "\n", |
| 251 | + "# polars\n", |
| 252 | + "monthly_aggregate(polars_df)\n", |
| 253 | + "\n", |
| 254 | + "# duckdb\n", |
| 255 | + "monthly_aggregate(duckdb_df)\n", |
| 256 | + "\n", |
| 257 | + "# pyarrow\n", |
| 258 | + "monthly_aggregate(arrow_table)\n", |
| 259 | + "\n", |
| 260 | + "# pyspark\n", |
| 261 | + "monthly_aggregate(spark_df)" |
| 262 | + ] |
| 263 | + }, |
| 264 | + { |
| 265 | + "cell_type": "markdown", |
| 266 | + "id": "14", |
| 267 | + "metadata": { |
| 268 | + "marimo": { |
| 269 | + "config": { |
| 270 | + "hide_code": true |
| 271 | + } |
| 272 | + } |
| 273 | + }, |
| 274 | + "source": [ |
| 275 | + "## Bonus - can we generate SQL?" |
| 276 | + ] |
| 277 | + }, |
| 278 | + { |
| 279 | + "cell_type": "code", |
| 280 | + "execution_count": null, |
| 281 | + "id": "15", |
| 282 | + "metadata": {}, |
| 283 | + "outputs": [], |
| 284 | + "source": [ |
| 285 | + "from sqlframe.duckdb import DuckDBSession\n", |
| 286 | + "\n", |
| 287 | + "sqlframe = DuckDBSession()\n", |
| 288 | + "sqlframe_df = sqlframe.createDataFrame(pandas_df)\n", |
| 289 | + "sqlframe_result = monthly_aggregate(sqlframe_df)\n", |
| 290 | + "print(sqlframe_result.sql(dialect=\"databricks\"))" |
| 291 | + ] |
| 292 | + } |
| 293 | + ], |
| 294 | + "metadata": { |
| 295 | + "language_info": { |
| 296 | + "name": "python" |
| 297 | + } |
| 298 | + }, |
| 299 | + "nbformat": 4, |
| 300 | + "nbformat_minor": 5 |
| 301 | +} |
0 commit comments