daviddoji/pbdl-book
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

diffphys code cleanup

Browse Source
This commit is contained in:
NT 2022-08-29 10:48:43 +02:00
parent 91f2656a0a
commit 82800c94ed
2 changed files with 4 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
diffphys-code-burgers.ipynb
View File

@ -89,9 +89,9 @@
"source": [ "source": [
"## Gradients\n", "## Gradients\n",
"\n", "\n",
"The `record_gradients` function of phiflow triggers the generation of a gradient tape to compute gradients of a simulation via `math.gradients(loss, values)`.\n", "The `math.gradient` operation of phiflow generates a gradient function for a scalar loss, and we use it below to compute gradients of a whole simulation with the chosen number of 32 time steps.\n",
"\n", "\n",
"To use it for the Burgers case we need to specify a loss function: we want the solution at $t=0.5$ to match the reference data. Thus we simply compute an $L^2$ difference between step number 16 and our constraint array as `loss`. Afterwards, we evaluate the gradient of the initial velocity state `velocity` with respect to this loss." "To use it for the Burgers case we need to compute an appropriate loss: we want the solution at $t=0.5$ to match the reference data. Thus we simply compute an $L^2$ difference between step number 16 and our constraint array as `loss`. Afterwards, we evaluate the gradient function of the initial velocity state `velocity` with respect to this loss. Phiflow's `math.gradient` generates a function that returns a gradient for each parameter, and as we only have a single one in form of the velocity here, `grad[0]` represents the gradient for the initial velocity."
] ]
}, },
{ {
@ -609,7 +609,8 @@
"\n", "\n",
"* You can try to adjust the training parameters to further improve the reconstruction.\n", "* You can try to adjust the training parameters to further improve the reconstruction.\n",
"* Activate a different optimizer, and observe the changing (not necessarily improved) convergence behavior.\n", "* Activate a different optimizer, and observe the changing (not necessarily improved) convergence behavior.\n",
"* Vary the number of steps, or the resolution of the simulation and reconstruction.\n" "* Vary the number of steps, or the resolution of the simulation and reconstruction.\n",
"* Try adding `@jit_compile` in a line before `loss_function`. This will incude a one-time compilation cost, but greatly speed up the optimization. \n"
] ]
} }
], ],

10
diffphys-code-sol.ipynb
View File

@ -618,16 +618,6 @@
"outputId": "e38b8b33-7d6f-40e8-ce64-0250c908db7a" "outputId": "e38b8b33-7d6f-40e8-ce64-0250c908db7a"
}, },
"outputs": [ "outputs": [
{
"output_type": "stream",
"name": "stderr",
"text": [
"/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:16: DeprecationWarning: Domain is deprecated and will be removed in a future release. Use a dict instead, e.g. CenteredGrid(values, extrapolation, **domain_dict)\n",
" app.launch_new_instance()\n",
"/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:16: FutureWarning: Domain is deprecated and will be removed in a future release. Use a dict instead, e.g. CenteredGrid(values, extrapolation, **domain_dict)\n",
" app.launch_new_instance()\n"
]
},
{ {
"output_type": "stream", "output_type": "stream",
"name": "stdout", "name": "stdout",