Live coded notebooks for the GPRV workshop. Check out https://github.com/dfm/gprv for more self contained versions.

live-intro-to-jax.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "b102d5cc-0756-45d8-ae50-abcf03538936",
   "metadata": {},
   "source": [
    "# Introduction to `jax`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "e627de02-c33d-4dbc-a287-539c95ecb0b7",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/homebrew/Caskroom/miniforge/base/envs/tinygp/lib/python3.9/site-packages/jax/_src/lib/__init__.py:32: UserWarning: JAX on Mac ARM machines is experimental and minimally tested. Please see https://github.com/google/jax/issues/5501 in the event of problems.\n",
      "  warnings.warn(\"JAX on Mac ARM machines is experimental and minimally tested. \"\n"
     ]
    }
   ],
   "source": [
    "import jax\n",
    "\n",
    "jax.config.update(\"jax_enable_x64\", True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "2e832a0c-cb6b-47b3-9d5b-f83837bdaafe",
   "metadata": {},
   "outputs": [],
   "source": [
    "import jax.numpy as jnp"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "8ff69de5-8f8a-44cc-8c6f-e8e66807108f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DeviceArray([0. , 0.5, 1. , 1.5, 2. ], dtype=float64)"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x = jnp.linspace(0, 2, 5)\n",
    "x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "37ea9bd7-dfa5-45ba-a8f6-27018461f090",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DeviceArray([0.        , 0.47942554, 0.84147098, 0.99749499, 0.90929743],            dtype=float64)"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "jnp.sin(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "c578f014-e232-4834-a207-6ef518e97a50",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "\n",
    "y = np.linspace(0, 2, 5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "d04e872f-d9c1-4914-80ac-7c912cec721f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DeviceArray([0.        , 0.47942554, 0.84147098, 0.99749499, 0.90929743],            dtype=float64)"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "jnp.sin(y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "048206c1-71b0-4bd7-af7e-6b663f309045",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0.        , 0.47942554, 0.84147098, 0.99749499, 0.90929743])"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.sin(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "id": "1505361c-677a-4f66-b437-cd29cd67bbb9",
   "metadata": {},
   "outputs": [],
   "source": [
    "from functools import partial\n",
    "\n",
    "@partial(jax.jit, backend=\"cpu\")\n",
    "def func(x):\n",
    "    arg = jnp.sin(x)\n",
    "    return 1.5 + jnp.exp(arg)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "7082e98a-22d8-4c85-834c-18af44fe7ecd",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DeviceArray([2.5       , 3.1151463 , 3.81977682, 4.21148102, 3.98257773],            dtype=float64)"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "func(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "d257b2ad-e12f-4ac0-9361-aa6608ab369b",
   "metadata": {},
   "outputs": [],
   "source": [
    "grad_func = jax.grad(func)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "id": "68574e9d-69ab-4743-884a-843003f7b23e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DeviceArray(1.41742422, dtype=float64, weak_type=True)"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "grad_func(0.5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "id": "ffa28e94-1ae0-4617-bc73-404f344c3fe9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DeviceArray([ 1.        ,  1.41742422,  1.25338077,  0.19180258,\n",
       "             -1.03311687], dtype=float64)"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "jax.vmap(grad_func)(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "id": "0e6c7d7b-8e68-4022-8f69-565c15eee9fe",
   "metadata": {},
   "outputs": [],
   "source": [
    "def func(params):\n",
    "    arg = jnp.sin(params[\"a\"])\n",
    "    return params[\"b\"] + jnp.exp(arg)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "id": "e1104a4b-eddc-4e24-8a20-3e6f2730ac22",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DeviceArray([1.        , 2.1151463 , 3.31977682, 4.21148102, 4.48257773],            dtype=float64)"
      ]
     },
     "execution_count": 57,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "params = {\n",
    "    \"a\": np.linspace(0, 2, 5),\n",
    "    \"b\": jnp.linspace(0, 2, 5),\n",
    "}\n",
    "func(params)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "id": "6f528892-f419-42c3-bfe4-34733660aa43",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DeviceArray([1.        , 2.1151463 , 3.31977682, 4.21148102, 4.48257773],            dtype=float64)"
      ]
     },
     "execution_count": 58,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "func(params)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "id": "31f97436-719a-4036-914c-4e469299a9df",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'a': DeviceArray([ 1.        ,  1.41742422,  1.25338077,  0.19180258,\n",
       "              -1.03311687], dtype=float64),\n",
       " 'b': DeviceArray([1., 1., 1., 1., 1.], dtype=float64)}"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "jax.vmap(jax.grad(func))(params)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "e4071c23-3e78-4608-8e53-0b760d5ccb0c",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.9.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}

live-intro-to-tinygp.ipynb