-
-
Save justinhou95/e116a82ba9ef9974840abdde634d07a0 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
{ | |
"cells": [ | |
{ | |
"cell_type": "markdown", | |
"id": "0a3f14e3", | |
"metadata": {}, | |
"source": [ | |
"## Optimal Stopping\n", | |
"\\begin{equation}\n", | |
"\t\tp_{n} = \\max\\{f_{n}, \\mathbb{E}_{Q}[\\alpha p_{n+1}\\vert \\mathcal{F}_{n}]\\}\n", | |
"\\end{equation}\n", | |
"\n", | |
"Approximate $\\mathbb{E}_{Q}[\\alpha p_{n+1}\\vert \\mathcal{F}_{n}]$ with $c_{n,\\theta}(X_1,\\dots,X_n)$\n", | |
"such that at each time $n$\n", | |
"\\begin{equation}\n", | |
"\t\t\\min_{\\theta} \\mathbb{E}_{Q} \\Big[\\big(c_{n,\\theta}(X_1,\\dots,X_n) - \\alpha p_{n+1}\\big)^{2}\\Big]\n", | |
"\t\\end{equation}\n", | |
" \n", | |
"\\begin{equation}\n", | |
"\t\t\\min_{\\theta} \\sum_{i=1}^{m} \\Big[\\big(c_{n,\\theta}(x^i_1,\\dots,x^i_n) - \\alpha p^{i}_{n+1}\\big)^{2}\\Big]\n", | |
"\t\\end{equation}\n", | |
"\n", | |
"\n", | |
"\n" | |
] | |
}, | |
{ | |
"cell_type": "markdown", | |
"id": "ccde6b59", | |
"metadata": {}, | |
"source": [ | |
"\n", | |
"### Backward induction\n", | |
"$$\n", | |
"p^i_n = g(x^i_n)\\cdot s_{n,\\theta}(x^i_n) + \\alpha p^i_{n+1}\\cdot\\big(1-s_{n,\\theta}(x^i_n)\\big)\n", | |
"$$\n", | |
"- $p^i_n$: values\n", | |
"- $g(x^i_n)$ : immediate_exercise_value\n", | |
"- $\\alpha p^i_{n+1}$: discounted_next_values\n", | |
"- $s_{n,\\theta}(\\cdot)$: stop(...)\n", | |
"- $s_{n,\\theta}(x^i_n)$: stopping_rule\n", | |
"\n", | |
"#### DOS (Deep Optimal Stopping):\n", | |
"$$s_{n,\\theta}(x^i_n) = f_{\\theta}(x^{i}_n) \\in [0,1]$$\n", | |
"- $f_{\\theta}(\\cdot)$: a neural network\n", | |
"\n", | |
"#### LSM (Least Squares Monte Carlo):\n", | |
"$$\n", | |
"s_{n,\\theta}(x^i_n) = \\mathbb{1}_{\\{g(x^i_n) \\geq c_{n,\\theta}(x^i_n)\\}}\n", | |
"$$\n", | |
"- $c_{n,\\theta}(x^i_n)$: continuation_value\n", | |
"- $c_{n,\\theta}(\\cdot)$: calculate_continuation_value(...)\n", | |
"$$\n", | |
"c_{n,\\theta}(\\cdot) = \\theta_n^{T}\\phi(\\cdot) = \\sum_{j=1}^{K}\\theta_{n,j} \\phi_{j}(\\cdot)\n", | |
"$$\n", | |
"\n", | |
"- **LSM** (Least Square Monte Carlo): $\\phi_j$ be polynomials\n", | |
"- **NLSM** (Neural Least Square Monte Carlo): $\\phi_j$ be neural networks\n", | |
"- **RLSM** (Randomized Least Square Monte Carlo): $\\phi_j$ be randomized neural networks\n", | |
"- **RRLSM** (Randomized Recurrent Least Square Monte Carlo): $\\phi_j$ be randomized neural networks taking one more hidden input $h_n$\n", | |
"\\begin{equation}\n", | |
"\\left\\{\\begin{aligned}\n", | |
" h_n &= \\sigma(A_x x_n + A_h h_{n-1} + b)\\\\\n", | |
" c_{n,\\theta}(h_n) &= A_n^{T}h_n + b_n = \\theta_n^{T}\\phi(x_n,h_{n-1})\n", | |
"\\end{aligned}\n", | |
"\\right.\n", | |
"\\end{equation}\n" | |
] | |
}, | |
{ | |
"cell_type": "markdown", | |
"id": "496d7d48", | |
"metadata": {}, | |
"source": [ | |
"### Reinforcement Learning \n", | |
"$$\n", | |
"p^i_n = \\max \\{ g(x^i_n), c_{n,\\theta}(x^i_1,\\dots,x^i_n) \\}\n", | |
"$$\n", | |
"\n", | |
"\\begin{equation}\n", | |
"\t\t\\min_{\\theta} \\mathbb{E}_{Q} \\Big[\\sum_{n=1}^{N}\\big(c_{n,\\theta}(X_1,\\dots,X_n) - \\alpha p_{n+1}\\big)^{2}\\Big]\n", | |
"\t\\end{equation}\n", | |
" \n", | |
"\\begin{equation}\n", | |
"\t\t\\min_{\\theta} \\Big[\\sum_{i=1}^{m}\\sum_{n=1}^{N}\\big(c_{n,\\theta}(x^i_1,\\dots,x^i_n) - \\alpha p^{i}_{n+1}\\big)^{2}\\Big]\n", | |
"\t\\end{equation}\n", | |
"\n", | |
" \n", | |
"#### Fitted Q-Iteration\n", | |
"\n", | |
"$$\n", | |
"c_{n,\\theta}(\\cdot) = \\theta^{T}\\phi(\\cdot, n) = \\sum_{j=1}^{K}\\theta_{j} \\phi_{j}(\\cdot,n)\n", | |
"$$\n", | |
"\n", | |
"\n", | |
"\n", | |
"\n", | |
"\n", | |
"- $\\phi_{j}$: self.bf\n", | |
"\n", | |
"\n", | |
"- **FQI** (fitted Q-Iteration): $\\phi_j$ \n", | |
"be polynomials\n", | |
"- **RFQI** (randomized fitted Q-Iteration): $\\phi_j$ be randomized neural networks\n", | |
"\n", | |
"$$\n", | |
"\\phi_{j}(x,n) = (\\sigma(A\\tilde{x} + b) , 1)\n", | |
"$$\n", | |
"where\n", | |
"$$\n", | |
"\\tilde{x} = (x, n, N-n) \\in \\mathbb{R}^{d+2}\n", | |
"$$" | |
] | |
}, | |
{ | |
"cell_type": "code", | |
"execution_count": null, | |
"id": "1033a887", | |
"metadata": {}, | |
"outputs": [], | |
"source": [] | |
} | |
], | |
"metadata": { | |
"kernelspec": { | |
"display_name": "Python [conda env:deephedge] *", | |
"language": "python", | |
"name": "conda-env-deephedge-py" | |
}, | |
"language_info": { | |
"codemirror_mode": { | |
"name": "ipython", | |
"version": 3 | |
}, | |
"file_extension": ".py", | |
"mimetype": "text/x-python", | |
"name": "python", | |
"nbconvert_exporter": "python", | |
"pygments_lexer": "ipython3", | |
"version": "3.9.7" | |
} | |
}, | |
"nbformat": 4, | |
"nbformat_minor": 5 | |
} |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment