Merge pull request #18 from tum-pbs/develop

FP Precision

Author: thunil

.gitattributes

@@ -0,0 +1 @@
+webglviewer/webglviewer.min.js linguist-vendored

.travis.yml

@@ -1,7 +1,6 @@
 language: python
 
 python:
-  - "2.7"
   - "3.6"
   - "3.7"
   - "3.8"
@@ -17,7 +16,6 @@ install:
   - if [[ $TRAVIS_PYTHON_VERSION != 3.8 ]]; then pip install --quiet tensorflow==1.14.0; fi
   - if [[ $TRAVIS_PYTHON_VERSION == "3.8" ]]; then pip install --quiet https://github.com/ppwwyyxx/tensorflow-wheels/releases/download/v0.2/tensorflow-1.15.0-cp38-cp38-linux_x86_64.whl numba llvmlite pytest; fi
   - pip install torch==1.4.0+cpu -f https://download.pytorch.org/whl/torch_stable.html
-  - if [[ $TRAVIS_PYTHON_VERSION == "2.7" ]]; then pip install future; fi
   - pip install nbformat ipython pylint;
   - pip install .
 

MANIFEST.in

@@ -1,3 +1,4 @@
 include phi/VERSION
+include documentation/Package_Info.md
 recursive-include phi/tf/cuda/build *
 recursive-include webglviewer *

README.md

@@ -7,10 +7,10 @@
 
 ![Gui](documentation/figures/WebInterface.png)
 
-Φ<sub>Flow</sub> is a research-oriented, open-source fluid simulation toolkit.
-It is written mostly in Python and can use both NumPy and TensorFlow for execution.
+Φ<sub>Flow</sub> is a research-oriented, open-source PDE solving toolkit that is fully differentiable.
+It is written mostly in Python and can use both NumPy, TensorFlow and PyTorch for execution.
 
-Having all functionality of a fluid simulation running in TensorFlow opens up the possibility of back-propagating gradients through the simulation as well as running the simulation on GPUs.
+Having all functionality of a fluid simulation running in a deep learning framework opens up the possibility of back-propagating gradients through the simulation as well as running the simulation on GPUs.
 
 ## Features
 
@@ -23,10 +23,11 @@ Having all functionality of a fluid simulation running in TensorFlow opens up th
 ## Publications
 
 - [Learning to Control PDEs with Differentiable Physics](https://ge.in.tum.de/publications/2020-iclr-holl/), *Philipp Holl, Vladlen Koltun, Nils Thuerey*, ICLR 2020.
+- [Solver-in-the-Loop: Learning from Differentiable Physics to Interact with Iterative PDE-Solvers](https://ge.in.tum.de/publications/2020-um-solver-in-the-loop/), *Kiwon Um, Yun Fei, Robert Brand, Philipp Holl, Nils Thuerey*, pre-print 2020.
 
 ## Installation
 
-To install Φ<sub>Flow</sub> with web interface, run:
+To install Φ<sub>Flow</sub> with its web interface, run:
 
 ```bash
 $ pip install phiflow[gui]

demos/fluid_logo.py

@@ -17,7 +17,7 @@
 
 def create_tum_logo():
     for x in range(1, 10, 2):
-        world.add(Obstacle(box[41:83, 15 + x * 7:15 + (x+1) * 7]))
+        world.add(Obstacle(box[41:83, 15 + x * 7:15 + (x + 1) * 7]))
     world.add_all(Obstacle(box[41:48, 43:50]), Obstacle(box[83:90, 15:43]), Obstacle(box[83:90, 50:85]))
 
 

demos/kuramoto_sivashinsky.py

@@ -0,0 +1,11 @@
+from phi.flow import *
+from phi.physics.flame import KuramotoSivashinsky
+
+
+flame1 = CenteredGrid.sample(Noise(scale=5) * 2, Domain([32], PERIODIC, box=box(32)), name='flame1')
+flame2 = (flame1 + Noise(scale=1) * 0.3).at(flame1).copied_with(name='flame2')
+world.add(flame1, KuramotoSivashinsky())
+world.add(flame2, KuramotoSivashinsky())
+
+app = App('Kuramoto-Sivashinsky', 'Chaotic flame simulation', dt=0.1, framerate=10)
+show(app)

demos/manual_fluid_numpy_or_tf.py

@@ -44,7 +44,6 @@
 
     SAVE_IMAGES = True
 
-
     def save_img(array, scale, name, idx=0):
         if len(array.shape) <= 4:
             ima = np.reshape(array[idx], [array.shape[1], array.shape[2]])  # remove channel dimension , 2d
@@ -104,4 +103,3 @@ def save_img(array, scale, name, idx=0):
 
         print("Step SESSION.run %04d done, DENSITY shape %s, means %s %s" %
               (i, FLOW.density.data.shape, np.mean(FLOW.density.data), np.mean(FLOW.velocity.staggered_tensor())))
-

demos/marker.py

@@ -9,10 +9,10 @@
 
 
 def checkerboard(resolution, size=8, offset=2):
-    data = math.zeros([1]+list(resolution)+[1])
+    data = math.zeros([1] + list(resolution) + [1])
     for y in range(size):
         for x in range(size):
-            data[:, y+offset::size*2, x+offset::size*2, :] = 1
+            data[:, y + offset::size * 2, x + offset::size * 2, :] = 1
     return data
 
 

demos/optimize_pressure.py

@@ -22,7 +22,7 @@ def __init__(self):
         with self.model_scope():
             initial_velocity = randn(domain.staggered_grid(0).shape) * 0.2
             optimizable_velocity = variable(initial_velocity)
-        velocity = optimizable_velocity * mask(box[0:62, 0:31])
+        velocity = optimizable_velocity * mask(box[0:62, 0:31]).at(optimizable_velocity)
         velocity = divergence_free(velocity, domain)
         # --- Target ---
         y, x = np.meshgrid(*[np.arange(-0.5, dim + 0.5) for dim in domain.resolution])

demos/rotating_bar.py

@@ -2,13 +2,13 @@
 physics_config.x_first()
 
 
-def update_obstacle(obstacle, dt, angular_velocity=0.05):
+def update_obstacle(obstacle, dt, angular_velocity=0.1):
     geometry = obstacle.geometry.rotated(angular_velocity * dt)
     return obstacle.copied_with(geometry=geometry, angular_velocity=angular_velocity)
 
 
 world.add(Fluid(Domain([128, 128], CLOSED, box=box([100, 100]))), physics=IncompressibleFlow())
-world.add(Obstacle(box[20:80, 48:52], name='bar'), physics=update_obstacle)
+world.add(Obstacle(box[10:90, 48:52], name='bar'), physics=update_obstacle)
 world.step()
 
 app = App('Moving Objects Demo', framerate=10)