Skip to content

Instantly share code, notes, and snippets.

@beckernick

beckernick/10min_cudf_cupy.ipynb

Last active March 19, 2021 17:01
Show Gist options
  • Save beckernick/e16b8a0617859ec413019fc0d4035841 to your computer and use it in GitHub Desktop.
Save beckernick/e16b8a0617859ec413019fc0d4035841 to your computer and use it in GitHub Desktop.
Download ZIP
10 Minutes to cuDF and CuPy
Display the source blob
Display the rendered blob
Raw
10min_cudf_cupy.ipynb
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# 10 Minutes to cuDF and CuPy\n",
"\n",
"This notebook provides introductory examples of how you can use cuDF and CuPy together to take advantage of CuPy array functionality (such as advanced linear algebra operations)."
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [],
"source": [
"import timeit\n",
"\n",
"import cupy as cp\n",
"import cudf"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Converting a cuDF DataFrame to a CuPy Array\n",
"\n",
"If we want to convert a cuDF DataFrame to a CuPy ndarray, There are multiple ways to do it:\n",
"\n",
"1. We can use the [dlpack](https://github.com/dmlc/dlpack) interface.\n",
"\n",
"2. We can also use `DataFrame.values`.\n",
"\n",
"3. We can also convert via the [CUDA array interface](https://numba.pydata.org/numba-doc/dev/cuda/cuda_array_interface.html) by using cuDF's `as_gpu_matrix` and CuPy's `asarray` functionality."
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"44.1 µs ± 689 ns per loop (mean ± std. dev. of 7 runs, 10000 loops each)\n",
"209 µs ± 2.77 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)\n",
"208 µs ± 3.14 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)\n"
]
}
],
"source": [
"nelem = 10000\n",
"df = cudf.DataFrame({'a':range(nelem),\n",
" 'b':range(500, nelem + 500),\n",
" 'c':range(1000, nelem + 1000)}\n",
" )\n",
"\n",
"%timeit arr_cupy = cp.fromDlpack(df.to_dlpack())\n",
"%timeit arr_cupy = df.values\n",
"%timeit arr_cupy = cp.asarray(df.as_gpu_matrix())"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"array([[ 0, 500, 1000],\n",
" [ 1, 501, 1001],\n",
" [ 2, 502, 1002],\n",
" ...,\n",
" [ 9997, 10497, 10997],\n",
" [ 9998, 10498, 10998],\n",
" [ 9999, 10499, 10999]])"
]
},
"execution_count": 3,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"arr_cupy = cp.fromDlpack(df.to_dlpack())\n",
"arr_cupy"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Converting a cuDF Series to a CuPy Array"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"There are also multiple ways to convert a cuDF Series to a CuPy array:\n",
"\n",
"1. We can pass the Series to `cupy.asarray` as cuDF Series exposes [`__cuda_array_interface__`](https://docs-cupy.chainer.org/en/stable/reference/interoperability.html).\n",
"2. We can leverage the dlpack interface `to_dlpack()`. \n",
"3. We can also use `Series.values` \n"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"22.1 µs ± 518 ns per loop (mean ± std. dev. of 7 runs, 10000 loops each)\n",
"58.3 µs ± 647 ns per loop (mean ± std. dev. of 7 runs, 10000 loops each)\n",
"80.2 µs ± 647 ns per loop (mean ± std. dev. of 7 runs, 10000 loops each)\n"
]
}
],
"source": [
"col = 'a'\n",
"\n",
"%timeit cola_cupy = cp.asarray(df[col])\n",
"%timeit cola_cupy = cp.fromDlpack(df[col].to_dlpack())\n",
"%timeit cola_cupy = df[col].values"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"array([ 0, 1, 2, ..., 9997, 9998, 9999])"
]
},
"execution_count": 5,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"cola_cupy = cp.asarray(df[col])\n",
"cola_cupy"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"From here, we can proceed with normal CuPy workflows, such as reshaping the array, getting the diagonal, or calculating the norm."
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"array([[ 0, 1, 2, ..., 197, 198, 199],\n",
" [ 200, 201, 202, ..., 397, 398, 399],\n",
" [ 400, 401, 402, ..., 597, 598, 599],\n",
" ...,\n",
" [9400, 9401, 9402, ..., 9597, 9598, 9599],\n",
" [9600, 9601, 9602, ..., 9797, 9798, 9799],\n",
" [9800, 9801, 9802, ..., 9997, 9998, 9999]])"
]
},
"execution_count": 6,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"reshaped_arr = cola_cupy.reshape(50, 200)\n",
"reshaped_arr"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"array([ 0, 201, 402, 603, 804, 1005, 1206, 1407, 1608, 1809, 2010,\n",
" 2211, 2412, 2613, 2814, 3015, 3216, 3417, 3618, 3819, 4020, 4221,\n",
" 4422, 4623, 4824, 5025, 5226, 5427, 5628, 5829, 6030, 6231, 6432,\n",
" 6633, 6834, 7035, 7236, 7437, 7638, 7839, 8040, 8241, 8442, 8643,\n",
" 8844, 9045, 9246, 9447, 9648, 9849])"
]
},
"execution_count": 7,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"reshaped_arr.diagonal()"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"array(577306.967739)"
]
},
"execution_count": 8,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"cp.linalg.norm(reshaped_arr)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Converting a CuPy Array to a cuDF DataFrame\n",
"\n",
"We can also convert a CuPy ndarray to a cuDF DataFrame. Like before, there are multiple ways to do it:\n",
"\n",
"1. **Easiest;** We can directly use the `DataFrame` constructor.\n",
"\n",
"2. We can use CUDA array interface with the `DataFrame` constructor.\n",
"\n",
"3. We can also use the [dlpack](https://github.com/dmlc/dlpack) interface.\n",
"\n",
"For the latter two cases, we'll need to make sure that our CuPy array is Fortran contiguous in memory (if it's not already). We can either transpose the array or simply coerce it to be Fortran contiguous beforehand."
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"13.1 ms ± 193 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)\n"
]
}
],
"source": [
"%timeit reshaped_df = cudf.DataFrame(reshaped_arr)"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>0</th>\n",
" <th>1</th>\n",
" <th>2</th>\n",
" <th>3</th>\n",
" <th>4</th>\n",
" <th>5</th>\n",
" <th>6</th>\n",
" <th>7</th>\n",
" <th>8</th>\n",
" <th>9</th>\n",
" <th>...</th>\n",
" <th>190</th>\n",
" <th>191</th>\n",
" <th>192</th>\n",
" <th>193</th>\n",
" <th>194</th>\n",
" <th>195</th>\n",
" <th>196</th>\n",
" <th>197</th>\n",
" <th>198</th>\n",
" <th>199</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>0</td>\n",
" <td>1</td>\n",
" <td>2</td>\n",
" <td>3</td>\n",
" <td>4</td>\n",
" <td>5</td>\n",
" <td>6</td>\n",
" <td>7</td>\n",
" <td>8</td>\n",
" <td>9</td>\n",
" <td>...</td>\n",
" <td>190</td>\n",
" <td>191</td>\n",
" <td>192</td>\n",
" <td>193</td>\n",
" <td>194</td>\n",
" <td>195</td>\n",
" <td>196</td>\n",
" <td>197</td>\n",
" <td>198</td>\n",
" <td>199</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>200</td>\n",
" <td>201</td>\n",
" <td>202</td>\n",
" <td>203</td>\n",
" <td>204</td>\n",
" <td>205</td>\n",
" <td>206</td>\n",
" <td>207</td>\n",
" <td>208</td>\n",
" <td>209</td>\n",
" <td>...</td>\n",
" <td>390</td>\n",
" <td>391</td>\n",
" <td>392</td>\n",
" <td>393</td>\n",
" <td>394</td>\n",
" <td>395</td>\n",
" <td>396</td>\n",
" <td>397</td>\n",
" <td>398</td>\n",
" <td>399</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>400</td>\n",
" <td>401</td>\n",
" <td>402</td>\n",
" <td>403</td>\n",
" <td>404</td>\n",
" <td>405</td>\n",
" <td>406</td>\n",
" <td>407</td>\n",
" <td>408</td>\n",
" <td>409</td>\n",
" <td>...</td>\n",
" <td>590</td>\n",
" <td>591</td>\n",
" <td>592</td>\n",
" <td>593</td>\n",
" <td>594</td>\n",
" <td>595</td>\n",
" <td>596</td>\n",
" <td>597</td>\n",
" <td>598</td>\n",
" <td>599</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>600</td>\n",
" <td>601</td>\n",
" <td>602</td>\n",
" <td>603</td>\n",
" <td>604</td>\n",
" <td>605</td>\n",
" <td>606</td>\n",
" <td>607</td>\n",
" <td>608</td>\n",
" <td>609</td>\n",
" <td>...</td>\n",
" <td>790</td>\n",
" <td>791</td>\n",
" <td>792</td>\n",
" <td>793</td>\n",
" <td>794</td>\n",
" <td>795</td>\n",
" <td>796</td>\n",
" <td>797</td>\n",
" <td>798</td>\n",
" <td>799</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>800</td>\n",
" <td>801</td>\n",
" <td>802</td>\n",
" <td>803</td>\n",
" <td>804</td>\n",
" <td>805</td>\n",
" <td>806</td>\n",
" <td>807</td>\n",
" <td>808</td>\n",
" <td>809</td>\n",
" <td>...</td>\n",
" <td>990</td>\n",
" <td>991</td>\n",
" <td>992</td>\n",
" <td>993</td>\n",
" <td>994</td>\n",
" <td>995</td>\n",
" <td>996</td>\n",
" <td>997</td>\n",
" <td>998</td>\n",
" <td>999</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"<p>5 rows × 200 columns</p>\n",
"</div>"
],
"text/plain": [
" 0 1 2 3 4 5 6 7 8 9 ... 190 191 192 193 \\\n",
"0 0 1 2 3 4 5 6 7 8 9 ... 190 191 192 193 \n",
"1 200 201 202 203 204 205 206 207 208 209 ... 390 391 392 393 \n",
"2 400 401 402 403 404 405 406 407 408 409 ... 590 591 592 593 \n",
"3 600 601 602 603 604 605 606 607 608 609 ... 790 791 792 793 \n",
"4 800 801 802 803 804 805 806 807 808 809 ... 990 991 992 993 \n",
"\n",
" 194 195 196 197 198 199 \n",
"0 194 195 196 197 198 199 \n",
"1 394 395 396 397 398 399 \n",
"2 594 595 596 597 598 599 \n",
"3 794 795 796 797 798 799 \n",
"4 994 995 996 997 998 999 \n",
"\n",
"[5 rows x 200 columns]"
]
},
"execution_count": 10,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"reshaped_df = cudf.DataFrame(reshaped_arr)\n",
"reshaped_df.head()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We can check whether our array is Fortran contiguous by using cupy.isfortran or looking at the [flags](https://docs-cupy.chainer.org/en/stable/reference/generated/cupy.ndarray.html#cupy.ndarray.flags) of the array."
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"False"
]
},
"execution_count": 11,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"cp.isfortran(reshaped_arr)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"In this case, we'll need to convert it before going to a cuDF DataFrame. In the next two cells, we create the DataFrame by leveraging dlpack and the CUDA array interface, respectively."
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"4.9 ms ± 26.4 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)\n"
]
}
],
"source": [
"%%timeit\n",
"\n",
"fortran_arr = cp.asfortranarray(reshaped_arr)\n",
"reshaped_df = cudf.DataFrame(fortran_arr)"
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"5.1 ms ± 23.3 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)\n"
]
}
],
"source": [
"%%timeit\n",
"\n",
"fortran_arr = cp.asfortranarray(reshaped_arr)\n",
"reshaped_df = cudf.from_dlpack(fortran_arr.toDlpack())"
]
},
{
"cell_type": "code",
"execution_count": 14,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>0</th>\n",
" <th>1</th>\n",
" <th>2</th>\n",
" <th>3</th>\n",
" <th>4</th>\n",
" <th>5</th>\n",
" <th>6</th>\n",
" <th>7</th>\n",
" <th>8</th>\n",
" <th>9</th>\n",
" <th>...</th>\n",
" <th>190</th>\n",
" <th>191</th>\n",
" <th>192</th>\n",
" <th>193</th>\n",
" <th>194</th>\n",
" <th>195</th>\n",
" <th>196</th>\n",
" <th>197</th>\n",
" <th>198</th>\n",
" <th>199</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>0</td>\n",
" <td>1</td>\n",
" <td>2</td>\n",
" <td>3</td>\n",
" <td>4</td>\n",
" <td>5</td>\n",
" <td>6</td>\n",
" <td>7</td>\n",
" <td>8</td>\n",
" <td>9</td>\n",
" <td>...</td>\n",
" <td>190</td>\n",
" <td>191</td>\n",
" <td>192</td>\n",
" <td>193</td>\n",
" <td>194</td>\n",
" <td>195</td>\n",
" <td>196</td>\n",
" <td>197</td>\n",
" <td>198</td>\n",
" <td>199</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>200</td>\n",
" <td>201</td>\n",
" <td>202</td>\n",
" <td>203</td>\n",
" <td>204</td>\n",
" <td>205</td>\n",
" <td>206</td>\n",
" <td>207</td>\n",
" <td>208</td>\n",
" <td>209</td>\n",
" <td>...</td>\n",
" <td>390</td>\n",
" <td>391</td>\n",
" <td>392</td>\n",
" <td>393</td>\n",
" <td>394</td>\n",
" <td>395</td>\n",
" <td>396</td>\n",
" <td>397</td>\n",
" <td>398</td>\n",
" <td>399</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>400</td>\n",
" <td>401</td>\n",
" <td>402</td>\n",
" <td>403</td>\n",
" <td>404</td>\n",
" <td>405</td>\n",
" <td>406</td>\n",
" <td>407</td>\n",
" <td>408</td>\n",
" <td>409</td>\n",
" <td>...</td>\n",
" <td>590</td>\n",
" <td>591</td>\n",
" <td>592</td>\n",
" <td>593</td>\n",
" <td>594</td>\n",
" <td>595</td>\n",
" <td>596</td>\n",
" <td>597</td>\n",
" <td>598</td>\n",
" <td>599</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>600</td>\n",
" <td>601</td>\n",
" <td>602</td>\n",
" <td>603</td>\n",
" <td>604</td>\n",
" <td>605</td>\n",
" <td>606</td>\n",
" <td>607</td>\n",
" <td>608</td>\n",
" <td>609</td>\n",
" <td>...</td>\n",
" <td>790</td>\n",
" <td>791</td>\n",
" <td>792</td>\n",
" <td>793</td>\n",
" <td>794</td>\n",
" <td>795</td>\n",
" <td>796</td>\n",
" <td>797</td>\n",
" <td>798</td>\n",
" <td>799</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>800</td>\n",
" <td>801</td>\n",
" <td>802</td>\n",
" <td>803</td>\n",
" <td>804</td>\n",
" <td>805</td>\n",
" <td>806</td>\n",
" <td>807</td>\n",
" <td>808</td>\n",
" <td>809</td>\n",
" <td>...</td>\n",
" <td>990</td>\n",
" <td>991</td>\n",
" <td>992</td>\n",
" <td>993</td>\n",
" <td>994</td>\n",
" <td>995</td>\n",
" <td>996</td>\n",
" <td>997</td>\n",
" <td>998</td>\n",
" <td>999</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"<p>5 rows × 200 columns</p>\n",
"</div>"
],
"text/plain": [
" 0 1 2 3 4 5 6 7 8 9 ... 190 191 192 193 \\\n",
"0 0 1 2 3 4 5 6 7 8 9 ... 190 191 192 193 \n",
"1 200 201 202 203 204 205 206 207 208 209 ... 390 391 392 393 \n",
"2 400 401 402 403 404 405 406 407 408 409 ... 590 591 592 593 \n",
"3 600 601 602 603 604 605 606 607 608 609 ... 790 791 792 793 \n",
"4 800 801 802 803 804 805 806 807 808 809 ... 990 991 992 993 \n",
"\n",
" 194 195 196 197 198 199 \n",
"0 194 195 196 197 198 199 \n",
"1 394 395 396 397 398 399 \n",
"2 594 595 596 597 598 599 \n",
"3 794 795 796 797 798 799 \n",
"4 994 995 996 997 998 999 \n",
"\n",
"[5 rows x 200 columns]"
]
},
"execution_count": 14,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"fortran_arr = cp.asfortranarray(reshaped_arr)\n",
"reshaped_df = cudf.DataFrame(fortran_arr)\n",
"reshaped_df.head()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Converting a CuPy Array to a cuDF Series\n",
"\n",
"To convert an array to a Series, we can directly pass the array to the `Series` constructor."
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"0 0\n",
"1 201\n",
"2 402\n",
"3 603\n",
"4 804\n",
"dtype: int64"
]
},
"execution_count": 15,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"cudf.Series(reshaped_arr.diagonal()).head()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Interweaving CuDF and CuPy for Smooth PyData Workflows\n",
"\n",
"RAPIDS libraries and the entire GPU PyData ecosystem are developing quickly, but sometimes a one library may not have the functionality you need. One example of this might be taking the row-wise sum (or mean) of a Pandas DataFrame. cuDF's support for row-wise operations isn't mature, so you'd need to either transpose the DataFrame or write a UDF and explicitly calculate the sum across each row. Transposing could lead to hundreds of thousands of columns (which cuDF wouldn't perform well with) depending on your data's shape, and writing a UDF can be time intensive.\n",
"\n",
"By leveraging the interoperability of the GPU PyData ecosystem, this operation becomes very easy. Let's take the row-wise sum of our previously reshaped cuDF DataFrame."
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>0</th>\n",
" <th>1</th>\n",
" <th>2</th>\n",
" <th>3</th>\n",
" <th>4</th>\n",
" <th>5</th>\n",
" <th>6</th>\n",
" <th>7</th>\n",
" <th>8</th>\n",
" <th>9</th>\n",
" <th>...</th>\n",
" <th>190</th>\n",
" <th>191</th>\n",
" <th>192</th>\n",
" <th>193</th>\n",
" <th>194</th>\n",
" <th>195</th>\n",
" <th>196</th>\n",
" <th>197</th>\n",
" <th>198</th>\n",
" <th>199</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>0</td>\n",
" <td>1</td>\n",
" <td>2</td>\n",
" <td>3</td>\n",
" <td>4</td>\n",
" <td>5</td>\n",
" <td>6</td>\n",
" <td>7</td>\n",
" <td>8</td>\n",
" <td>9</td>\n",
" <td>...</td>\n",
" <td>190</td>\n",
" <td>191</td>\n",
" <td>192</td>\n",
" <td>193</td>\n",
" <td>194</td>\n",
" <td>195</td>\n",
" <td>196</td>\n",
" <td>197</td>\n",
" <td>198</td>\n",
" <td>199</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>200</td>\n",
" <td>201</td>\n",
" <td>202</td>\n",
" <td>203</td>\n",
" <td>204</td>\n",
" <td>205</td>\n",
" <td>206</td>\n",
" <td>207</td>\n",
" <td>208</td>\n",
" <td>209</td>\n",
" <td>...</td>\n",
" <td>390</td>\n",
" <td>391</td>\n",
" <td>392</td>\n",
" <td>393</td>\n",
" <td>394</td>\n",
" <td>395</td>\n",
" <td>396</td>\n",
" <td>397</td>\n",
" <td>398</td>\n",
" <td>399</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>400</td>\n",
" <td>401</td>\n",
" <td>402</td>\n",
" <td>403</td>\n",
" <td>404</td>\n",
" <td>405</td>\n",
" <td>406</td>\n",
" <td>407</td>\n",
" <td>408</td>\n",
" <td>409</td>\n",
" <td>...</td>\n",
" <td>590</td>\n",
" <td>591</td>\n",
" <td>592</td>\n",
" <td>593</td>\n",
" <td>594</td>\n",
" <td>595</td>\n",
" <td>596</td>\n",
" <td>597</td>\n",
" <td>598</td>\n",
" <td>599</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>600</td>\n",
" <td>601</td>\n",
" <td>602</td>\n",
" <td>603</td>\n",
" <td>604</td>\n",
" <td>605</td>\n",
" <td>606</td>\n",
" <td>607</td>\n",
" <td>608</td>\n",
" <td>609</td>\n",
" <td>...</td>\n",
" <td>790</td>\n",
" <td>791</td>\n",
" <td>792</td>\n",
" <td>793</td>\n",
" <td>794</td>\n",
" <td>795</td>\n",
" <td>796</td>\n",
" <td>797</td>\n",
" <td>798</td>\n",
" <td>799</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>800</td>\n",
" <td>801</td>\n",
" <td>802</td>\n",
" <td>803</td>\n",
" <td>804</td>\n",
" <td>805</td>\n",
" <td>806</td>\n",
" <td>807</td>\n",
" <td>808</td>\n",
" <td>809</td>\n",
" <td>...</td>\n",
" <td>990</td>\n",
" <td>991</td>\n",
" <td>992</td>\n",
" <td>993</td>\n",
" <td>994</td>\n",
" <td>995</td>\n",
" <td>996</td>\n",
" <td>997</td>\n",
" <td>998</td>\n",
" <td>999</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"<p>5 rows × 200 columns</p>\n",
"</div>"
],
"text/plain": [
" 0 1 2 3 4 5 6 7 8 9 ... 190 191 192 193 \\\n",
"0 0 1 2 3 4 5 6 7 8 9 ... 190 191 192 193 \n",
"1 200 201 202 203 204 205 206 207 208 209 ... 390 391 392 393 \n",
"2 400 401 402 403 404 405 406 407 408 409 ... 590 591 592 593 \n",
"3 600 601 602 603 604 605 606 607 608 609 ... 790 791 792 793 \n",
"4 800 801 802 803 804 805 806 807 808 809 ... 990 991 992 993 \n",
"\n",
" 194 195 196 197 198 199 \n",
"0 194 195 196 197 198 199 \n",
"1 394 395 396 397 398 399 \n",
"2 594 595 596 597 598 599 \n",
"3 794 795 796 797 798 799 \n",
"4 994 995 996 997 998 999 \n",
"\n",
"[5 rows x 200 columns]"
]
},
"execution_count": 16,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"reshaped_df.head()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We can just transform it into a CuPy array and use the `axis` argument of `sum`."
]
},
{
"cell_type": "code",
"execution_count": 17,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"array([ 19900, 59900, 99900, 139900, 179900, 219900, 259900,\n",
" 299900, 339900, 379900, 419900, 459900, 499900, 539900,\n",
" 579900, 619900, 659900, 699900, 739900, 779900, 819900,\n",
" 859900, 899900, 939900, 979900, 1019900, 1059900, 1099900,\n",
" 1139900, 1179900, 1219900, 1259900, 1299900, 1339900, 1379900,\n",
" 1419900, 1459900, 1499900, 1539900, 1579900, 1619900, 1659900,\n",
" 1699900, 1739900, 1779900, 1819900, 1859900, 1899900, 1939900,\n",
" 1979900])"
]
},
"execution_count": 17,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"new_arr = cp.fromDlpack(reshaped_df.to_dlpack())\n",
"new_arr.sum(axis=1)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"With just that single line, we're able to seamlessly move between data structures in this ecosystem, giving us enormous flexibility without sacrificing speed."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Converting a cuDF DataFrame to a CuPy Sparse Matrix\n",
"\n",
"We can also convert a DataFrame or Series to a CuPy sparse matrix. We might want to do this if downstream processes expect CuPy sparse matrices as an input.\n",
"\n",
"The sparse matrix data structure is defined by three dense arrays. We'll define a small helper function for cleanliness."
]
},
{
"cell_type": "code",
"execution_count": 18,
"metadata": {},
"outputs": [],
"source": [
"def cudf_to_cupy_sparse_matrix(data, sparseformat='column'):\n",
" \"\"\"Converts a cuDF object to a CuPy Sparse Column matrix.\n",
" \"\"\"\n",
" if sparseformat not in ('row', 'column',):\n",
" raise ValueError(\"Let's focus on column and row formats for now.\")\n",
" \n",
" _sparse_constructor = cp.sparse.csc_matrix\n",
" if sparseformat == 'row':\n",
" _sparse_constructor = cp.sparse.csr_matrix\n",
"\n",
" return _sparse_constructor(cp.fromDlpack(data.to_dlpack()))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We can define a sparsely populated DataFrame to illustrate this conversion to either sparse matrix format."
]
},
{
"cell_type": "code",
"execution_count": 19,
"metadata": {},
"outputs": [],
"source": [
"df = cudf.DataFrame()\n",
"nelem = 10000\n",
"nonzero = 1000\n",
"for i in range(20):\n",
" arr = cp.random.normal(5, 5, nelem)\n",
" arr[cp.random.choice(arr.shape[0], nelem-nonzero, replace=False)] = 0\n",
" df['a' + str(i)] = arr"
]
},
{
"cell_type": "code",
"execution_count": 20,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>a0</th>\n",
" <th>a1</th>\n",
" <th>a2</th>\n",
" <th>a3</th>\n",
" <th>a4</th>\n",
" <th>a5</th>\n",
" <th>a6</th>\n",
" <th>a7</th>\n",
" <th>a8</th>\n",
" <th>a9</th>\n",
" <th>a10</th>\n",
" <th>a11</th>\n",
" <th>a12</th>\n",
" <th>a13</th>\n",
" <th>a14</th>\n",
" <th>a15</th>\n",
" <th>a16</th>\n",
" <th>a17</th>\n",
" <th>a18</th>\n",
" <th>a19</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" <td>0.00000</td>\n",
" <td>0.0</td>\n",
" <td>16.822959</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" <td>0.00000</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>6.618972</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" <td>2.25678</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" <td>0.00000</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" <td>0.0</td>\n",
" <td>2.715802</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>4.296568</td>\n",
" <td>0.00000</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" <td>0.0</td>\n",
" <td>0.000000</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>0.0</td>\n",
" <td>4.865495</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"</div>"
],
"text/plain": [
" a0 a1 a2 a3 a4 a5 a6 a7 a8 a9 a10 a11 \\\n",
"0 0.0 0.0 0.000000 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.000000 0.00000 \n",
"1 0.0 0.0 0.000000 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.000000 0.00000 \n",
"2 0.0 0.0 6.618972 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.000000 2.25678 \n",
"3 0.0 0.0 0.000000 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.000000 0.00000 \n",
"4 0.0 0.0 0.000000 0.0 0.0 0.0 0.0 0.0 0.0 0.0 4.296568 0.00000 \n",
"\n",
" a12 a13 a14 a15 a16 a17 a18 a19 \n",
"0 0.0 16.822959 0.0 0.000000 0.0 0.0 0.0 0.000000 \n",
"1 0.0 0.000000 0.0 0.000000 0.0 0.0 0.0 0.000000 \n",
"2 0.0 0.000000 0.0 0.000000 0.0 0.0 0.0 0.000000 \n",
"3 0.0 0.000000 0.0 2.715802 0.0 0.0 0.0 0.000000 \n",
"4 0.0 0.000000 0.0 0.000000 0.0 0.0 0.0 4.865495 "
]
},
"execution_count": 20,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"df.head()"
]
},
{
"cell_type": "code",
"execution_count": 21,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<cupyx.scipy.sparse.csc.csc_matrix at 0x7f25e49466a0>"
]
},
"execution_count": 21,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"sparse_data = cudf_to_cupy_sparse_matrix(df)\n",
"sparse_data"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"From here, we could continue our workflow with a CuPy sparse matrix.\n",
"\n",
"For a full list of the functionality built into these libraries, we encourage you to check out the API docs for [cuDF](https://docs.rapids.ai/api/cudf/nightly/) and [CuPy](https://docs-cupy.chainer.org/en/stable/index.html)."
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.8.6"
}
},
"nbformat": 4,
"nbformat_minor": 4
}
@kaish114
Copy link

kaish114 commented May 7, 2020

image
How can I solve this??

@kaish114
Copy link

kaish114 commented May 7, 2020

How can I solve this??

I need help converting titanic.csv dataset(cudf) to cupy ndarray.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment