current work on differentiating the profit function
youainti
5 years ago
parent
1ad55eb603
commit
90a4a56baf
@ -0,0 +1,460 @@
|
|||||||
|
{
|
||||||
|
"cells": [
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 1,
|
||||||
|
"id": "similar-ebony",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"import torch\n",
|
||||||
|
"import combined as c\n",
|
||||||
|
"import NeuralNetworkSpecifications as nns"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 2,
|
||||||
|
"id": "spread-hygiene",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [
|
||||||
|
{
|
||||||
|
"name": "stdout",
|
||||||
|
"output_type": "stream",
|
||||||
|
"text": [
|
||||||
|
"tensor([[[6., 1., 0.]],\n",
|
||||||
|
"\n",
|
||||||
|
" [[2., 0., 4.]],\n",
|
||||||
|
"\n",
|
||||||
|
" [[7., 6., 9.]],\n",
|
||||||
|
"\n",
|
||||||
|
" [[3., 6., 9.]],\n",
|
||||||
|
"\n",
|
||||||
|
" [[9., 1., 2.]]])\n"
|
||||||
|
]
|
||||||
|
}
|
||||||
|
],
|
||||||
|
"source": [
|
||||||
|
"batch_size,states,choices = 5,3,1\n",
|
||||||
|
"constellations = states -1 #determined by debris tracking\n",
|
||||||
|
"max_start_state = 10\n",
|
||||||
|
"\n",
|
||||||
|
"stocks = torch.randint(max_start_state,(batch_size,1,constellations),dtype=torch.float32)\n",
|
||||||
|
"debris = torch.randint(max_start_state,(batch_size,1,1),dtype=torch.float32)\n",
|
||||||
|
"\n",
|
||||||
|
"s = c.States(stocks, debris)\n",
|
||||||
|
"\n",
|
||||||
|
"print(s.values)"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 3,
|
||||||
|
"id": "attended-making",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [
|
||||||
|
{
|
||||||
|
"data": {
|
||||||
|
"text/plain": [
|
||||||
|
"tensor([6.3344e-07, 4.6190e-07])"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
"execution_count": 3,
|
||||||
|
"metadata": {},
|
||||||
|
"output_type": "execute_result"
|
||||||
|
}
|
||||||
|
],
|
||||||
|
"source": [
|
||||||
|
"constellation_collision_risk = 1e-6 * torch.rand(constellations)\n",
|
||||||
|
"constellation_collision_risk"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 4,
|
||||||
|
"id": "strategic-american",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"debris_decay_rate = 0.1\n",
|
||||||
|
"launch_debris = 0.05\n",
|
||||||
|
"debris_autocatalysis_rate = 1.4\n",
|
||||||
|
"\n",
|
||||||
|
"benefit_weight0 = torch.tensor([1.0,-0.02])\n",
|
||||||
|
"benefit_weight1 = torch.tensor([0.0,1.0])"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 5,
|
||||||
|
"id": "hired-consent",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"pm = c.PhysicalModel(10\n",
|
||||||
|
" , constellation_collision_risk #constellations_collision_risk #as tensor\n",
|
||||||
|
" , debris_decay_rate #debris_decay_rate\n",
|
||||||
|
" , launch_debris #launch_debris\n",
|
||||||
|
" , debris_autocatalysis_rate #debris_autocatalysis_rate\n",
|
||||||
|
" )"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 6,
|
||||||
|
"id": "copyrighted-tackle",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [
|
||||||
|
{
|
||||||
|
"ename": "TypeError",
|
||||||
|
"evalue": "__init__() missing 1 required positional argument: 'launch_cost'",
|
||||||
|
"output_type": "error",
|
||||||
|
"traceback": [
|
||||||
|
"\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
|
||||||
|
"\u001b[0;31mTypeError\u001b[0m Traceback (most recent call last)",
|
||||||
|
"\u001b[0;32m<ipython-input-6-54bb8ddad0e2>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m 1\u001b[0m \u001b[0mlaunch_cost\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;36m5\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 2\u001b[0;31m ea0 = c.LinearProfit(\n\u001b[0m\u001b[1;32m 3\u001b[0m \u001b[0;36m0\u001b[0m \u001b[0;31m#constellation index\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m 4\u001b[0m \u001b[0;34m,\u001b[0m\u001b[0;36m0.95\u001b[0m \u001b[0;31m#discount\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m 5\u001b[0m \u001b[0;34m,\u001b[0m\u001b[0mbenefit_weight0\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
|
||||||
|
"\u001b[0;31mTypeError\u001b[0m: __init__() missing 1 required positional argument: 'launch_cost'"
|
||||||
|
]
|
||||||
|
}
|
||||||
|
],
|
||||||
|
"source": [
|
||||||
|
"launch_cost = 5\n",
|
||||||
|
"ea0 = c.LinearProfit(\n",
|
||||||
|
" 0 #constellation index\n",
|
||||||
|
" ,0.95 #discount\n",
|
||||||
|
" ,benefit_weight0\n",
|
||||||
|
" ,launch_cost #launch_cost\n",
|
||||||
|
" )\n",
|
||||||
|
"ea1 = c.LinearProfit(\n",
|
||||||
|
" 1 #constellation index\n",
|
||||||
|
" ,0.95 #discount\n",
|
||||||
|
" ,benefit_weight1\n",
|
||||||
|
" ,launch_cost #launch_cost\n",
|
||||||
|
" )"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 7,
|
||||||
|
"id": "accepted-namibia",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"enn = nns.EstimandNN(batch_size\n",
|
||||||
|
" ,states\n",
|
||||||
|
" ,choices\n",
|
||||||
|
" ,constellations\n",
|
||||||
|
" ,12)"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 31,
|
||||||
|
"id": "revolutionary-eight",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [
|
||||||
|
{
|
||||||
|
"data": {
|
||||||
|
"text/plain": [
|
||||||
|
"tensor([[[0.0000],\n",
|
||||||
|
" [0.0000]],\n",
|
||||||
|
"\n",
|
||||||
|
" [[0.0000],\n",
|
||||||
|
" [0.0021]],\n",
|
||||||
|
"\n",
|
||||||
|
" [[0.1109],\n",
|
||||||
|
" [0.0835]],\n",
|
||||||
|
"\n",
|
||||||
|
" [[0.0884],\n",
|
||||||
|
" [0.1051]],\n",
|
||||||
|
"\n",
|
||||||
|
" [[0.0000],\n",
|
||||||
|
" [0.0000]]], grad_fn=<ReluBackward0>)"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
"execution_count": 31,
|
||||||
|
"metadata": {},
|
||||||
|
"output_type": "execute_result"
|
||||||
|
}
|
||||||
|
],
|
||||||
|
"source": [
|
||||||
|
"launch_decisions = enn.forward(s.values).choices\n",
|
||||||
|
"launch_decisions"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 12,
|
||||||
|
"id": "abroad-mobile",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"w = torch.tensor([[1.0,0],[0,-0.2]])\n",
|
||||||
|
"ww = torch.tensor([1.0, -0.2])"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 13,
|
||||||
|
"id": "seasonal-companion",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [
|
||||||
|
{
|
||||||
|
"data": {
|
||||||
|
"text/plain": [
|
||||||
|
"torch.Size([5, 1, 2])"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
"execution_count": 13,
|
||||||
|
"metadata": {},
|
||||||
|
"output_type": "execute_result"
|
||||||
|
}
|
||||||
|
],
|
||||||
|
"source": [
|
||||||
|
"stocks.size()"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 16,
|
||||||
|
"id": "jewish-zoning",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"class LinearProfit():\n",
|
||||||
|
" \"\"\"\n",
|
||||||
|
" The simplest type of profit function available.\n",
|
||||||
|
" \"\"\"\n",
|
||||||
|
" def __init__(self, batch_size, constellation_number, discount_factor, benefit_weights, launch_cost, deorbit_cost=0, ):\n",
|
||||||
|
" self.batch_size = batch_size\n",
|
||||||
|
" \n",
|
||||||
|
" \n",
|
||||||
|
" #track which constellation this is.\n",
|
||||||
|
" self.constellation_number = constellation_number\n",
|
||||||
|
" \n",
|
||||||
|
" #get the number of constellations (pull from the benefit weight, in the dimension that counts across constellations)\n",
|
||||||
|
" self.number_of_constellations = benefit_weights.size()[0] -1\n",
|
||||||
|
"\n",
|
||||||
|
" #parameters describing the agent's situation\n",
|
||||||
|
" self.discount_factor = discount_factor\n",
|
||||||
|
" self.benefit_weights = benefit_weights\n",
|
||||||
|
" self.launch_cost = launch_cost\n",
|
||||||
|
" self.deorbit_cost = deorbit_cost\n",
|
||||||
|
"\n",
|
||||||
|
" def __str__(self):\n",
|
||||||
|
" return \"LinearProfit\\n Benefit weights:\\t{}\\n launch cost:\\t{}\\n Deorbit cost:\\t{}\".format(self.benefit_weights, self.launch_cost, self.deorbit_cost)\n",
|
||||||
|
"\n",
|
||||||
|
" \n",
|
||||||
|
" def _period_benefit(self,stocks,debris,launches):\n",
|
||||||
|
" # multiply benefits times stocks\n",
|
||||||
|
" # sum across constellations\n",
|
||||||
|
" # reshape to standard dimensions\n",
|
||||||
|
" # subtract launch costs. \n",
|
||||||
|
" profit = torch.tensordot(self.benefit_weights,stocks, [[0],[1]])[:,self.constellation_number] \\\n",
|
||||||
|
" - (self.launch_cost * launches)[:,self.constellation_number,0]\n",
|
||||||
|
" return profit.view(batch_size,1)\n",
|
||||||
|
"\n",
|
||||||
|
"\n",
|
||||||
|
"\n"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 17,
|
||||||
|
"id": "surgical-diversity",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"def test(stocks,launches):\n",
|
||||||
|
" # multiply benefits times stocks\n",
|
||||||
|
" # sum across constellations\n",
|
||||||
|
" # reshape to standard dimensions\n",
|
||||||
|
" # subtract launch costs. \n",
|
||||||
|
" profit = torch.tensordot(ww,stocks, [[0],[1]])[:,0] - (launch_cost * launch_decisions)[:,0,0]\n",
|
||||||
|
" return profit.view(batch_size,1)"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 18,
|
||||||
|
"id": "western-sixth",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"t = LinearProfit(batch_size #batch_size\n",
|
||||||
|
" ,0 #constellation index\n",
|
||||||
|
" ,0.95 #discount\n",
|
||||||
|
" ,benefit_weight0\n",
|
||||||
|
" ,launch_cost #launch_cost\n",
|
||||||
|
" )"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 19,
|
||||||
|
"id": "conscious-debut",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [
|
||||||
|
{
|
||||||
|
"data": {
|
||||||
|
"text/plain": [
|
||||||
|
"<__main__.LinearProfit at 0x7f0664fad4c0>"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
"execution_count": 19,
|
||||||
|
"metadata": {},
|
||||||
|
"output_type": "execute_result"
|
||||||
|
}
|
||||||
|
],
|
||||||
|
"source": [
|
||||||
|
"t"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 20,
|
||||||
|
"id": "eight-cheat",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [
|
||||||
|
{
|
||||||
|
"data": {
|
||||||
|
"text/plain": [
|
||||||
|
"tensor([[5.8800],\n",
|
||||||
|
" [1.9600],\n",
|
||||||
|
" [6.3054],\n",
|
||||||
|
" [2.4978],\n",
|
||||||
|
" [8.8200]], grad_fn=<ViewBackward>)"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
"execution_count": 20,
|
||||||
|
"metadata": {},
|
||||||
|
"output_type": "execute_result"
|
||||||
|
}
|
||||||
|
],
|
||||||
|
"source": [
|
||||||
|
"t._period_benefit(s.stocks,s.debris,launch_decisions)"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 25,
|
||||||
|
"id": "juvenile-barcelona",
|
||||||
|
"metadata": {
|
||||||
|
"tags": []
|
||||||
|
},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"def f(stocks, debris, launches):\n",
|
||||||
|
" return torch.autograd.functional.jacobian(t._period_benefit\n",
|
||||||
|
" ,(stocks,debris,launches)\n",
|
||||||
|
" ,create_graph=True\n",
|
||||||
|
" )\n",
|
||||||
|
"def ff(stocks, debris, launches):\n",
|
||||||
|
" return torch.autograd.functional.jacobian(f\n",
|
||||||
|
" ,(stocks,debris,launches)\n",
|
||||||
|
" ,create_graph=True\n",
|
||||||
|
" )"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 30,
|
||||||
|
"id": "freelance-publicity",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [
|
||||||
|
{
|
||||||
|
"data": {
|
||||||
|
"text/plain": [
|
||||||
|
"torch.Size([5, 1, 5, 1, 2, 5, 1, 2])"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
"execution_count": 30,
|
||||||
|
"metadata": {},
|
||||||
|
"output_type": "execute_result"
|
||||||
|
}
|
||||||
|
],
|
||||||
|
"source": []
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 55,
|
||||||
|
"id": "vocational-operator",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [
|
||||||
|
{
|
||||||
|
"data": {
|
||||||
|
"text/plain": [
|
||||||
|
"tensor([[[6.0000, 1.0000],\n",
|
||||||
|
" [2.0000, 0.0000],\n",
|
||||||
|
" [7.0000, 6.0000],\n",
|
||||||
|
" [3.0000, 6.0000],\n",
|
||||||
|
" [9.0000, 1.0000]],\n",
|
||||||
|
"\n",
|
||||||
|
" [[4.8000, 0.8000],\n",
|
||||||
|
" [1.6000, 0.0000],\n",
|
||||||
|
" [5.6000, 4.8000],\n",
|
||||||
|
" [2.4000, 4.8000],\n",
|
||||||
|
" [7.2000, 0.8000]]])"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
"execution_count": 55,
|
||||||
|
"metadata": {},
|
||||||
|
"output_type": "execute_result"
|
||||||
|
}
|
||||||
|
],
|
||||||
|
"source": [
|
||||||
|
"torch.tensordot(torch.tensor([[1.0,-0.2],[0,1]]),stocks, [[0],[1]])"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": 48,
|
||||||
|
"id": "nuclear-alberta",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [
|
||||||
|
{
|
||||||
|
"data": {
|
||||||
|
"text/plain": [
|
||||||
|
"torch.Size([5, 1, 2])"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
"execution_count": 48,
|
||||||
|
"metadata": {},
|
||||||
|
"output_type": "execute_result"
|
||||||
|
}
|
||||||
|
],
|
||||||
|
"source": [
|
||||||
|
"stocks.shape"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "immune-machinery",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": []
|
||||||
|
}
|
||||||
|
],
|
||||||
|
"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": "3.8.8"
|
||||||
|
}
|
||||||
|
},
|
||||||
|
"nbformat": 4,
|
||||||
|
"nbformat_minor": 5
|
||||||
|
}
|
||||||
Loading…
Reference in New Issue