Bump version: 0.1.0-rc1 → 0.1.0

- add `use_gpu()`, `use_gpu()` in `__init__.py`

.bumpversion.cfg

@@ -1,5 +1,5 @@
 [bumpversion]
-current_version = 0.0.1
+current_version = 0.1.0
 commit = True
 tag = True
 tag_name = v{new_version}

README.md

@@ -7,10 +7,7 @@
 [![DockerHub](https://img.shields.io/docker/pulls/idrl/idrlnet.svg)](https://hub.docker.com/r/idrl/idrlnet)
 [![CodeFactor](https://www.codefactor.io/repository/github/idrl-lab/idrlnet/badge/master)](https://www.codefactor.io/repository/github/idrl-lab/idrlnet/overview/master)
 
-
-**IDRLnet** is a machine learning library on top of [PyTorch](https://pytorch.org/). Use IDRLnet if you need a machine
-learning library that solves both forward and inverse differential equations via physics-informed neural
-networks (PINN). IDRLnet is a flexible framework inspired by [Nvidia Simnet](https://developer.nvidia.com/simnet>).
+**IDRLnet** is a machine learning library on top of [PyTorch](https://pytorch.org/). Use IDRLnet if you need a machine learning library that solves both forward and inverse differential equations via physics-informed neural networks (PINN). IDRLnet is a flexible framework inspired by [Nvidia Simnet](https://developer.nvidia.com/simnet>).
 
 ## Docs
 
@@ -66,21 +63,27 @@ pip install -e .
 
 IDRLnet supports
 
--  complex domain geometries without mesh generation. Provided geometries include interval, triangle, rectangle, polygon,
-   circle, sphere... Other geometries can be constructed using three boolean operations: union, difference, and
-   intersection;
+-  complex domain geometries without mesh generation. Provided geometries include interval, triangle, rectangle, polygon, circle, sphere... Other geometries can be constructed using three boolean operations: union, difference, and intersection;
+   ![Geometry](https://raw.githubusercontent.com/weipeng0098/picture/master/20210617081809.png)
    
--  sampling in the interior of the defined geometry or on the boundary with given conditions.
+- sampling in the interior of the defined geometry or on the boundary with given conditions.
 
--  enables the user code to be structured. Data sources, operations, constraints are all represented by ``Node``. The graph
-   will be automatically constructed via label symbols of each node. Getting rid of the explicit construction via
-   explicit expressions, users model problems more naturally.
+- enables the user code to be structured. Data sources, operations, constraints are all represented by ``Node``. The graph will be automatically constructed via label symbols of each node. Getting rid of the explicit construction via explicit expressions, users model problems more naturally.
+
+- builds computational graph automatically;
+
+   ![computationDomain](https://raw.githubusercontent.com/weipeng0098/picture/master/20220815142531.png)
+
+-  user-defined callbacks;
+   
+   ![callback](https://raw.githubusercontent.com/weipeng0098/picture/master/20220815142621.png)
    
 -  solving variational minimization problem;
+   <img src="https://raw.githubusercontent.com/weipeng0098/picture/master/20210617082331.gif" alt="miniface" style="zoom:33%;" />
    
--  solving integral differential equation;
+- solving integral differential equation;
 
--  adaptive resampling;
+- adaptive resampling;
 
 -  recover unknown parameters of PDEs from noisy measurement data.
 
@@ -99,11 +102,9 @@ First off, thanks for taking the time to contribute!
 
 -  **Reporting bugs.** To report a bug, simply open an issue in the GitHub "Issues" section.
 
--  **Suggesting enhancements.** To submit an enhancement suggestion for IDRLnet, including completely new features and
-   minor improvements to existing functionality, let us know by opening an issue.
+-  **Suggesting enhancements.** To submit an enhancement suggestion for IDRLnet, including completely new features and minor improvements to existing functionality, let us know by opening an issue.
    
--  **Pull requests.** If you made improvements to IDRLnet, fixed a bug, or had a new example, feel free to send us a
-   pull-request.
+-  **Pull requests.** If you made improvements to IDRLnet, fixed a bug, or had a new example, feel free to send us a pull-request.
    
 -  **Asking questions.** To get help on how to use IDRLnet or its functionalities, you can as well open an issue.
 

docs/conf.py

@@ -22,7 +22,7 @@ copyright = "2021, IDRL"
 author = "IDRL"
 
 # The full version, including alpha/beta/rc tags
-release = "0.0.1"
+release = "0.1.0"
 
 # -- General configuration ---------------------------------------------------
 

docs/user/get_started/5_inverse_wave_equation.md

@@ -3,7 +3,7 @@ Consider the 1d wave equation:
 
 $$
 \begin{equation}
-\frac{\partial^2u}{\partial t^2}=c\frac{\partial^2u}{\partial x^2},
+\frac{\partial^2u}{\partial t^2}=c^2\frac{\partial^2u}{\partial x^2},
 \end{equation}
 $$
 where $c>0$ is unknown and is to be estimated. A group of data pairs $\{x_i, t_i, u_i\}_{i=1,2,\cdot,N}$ is observed.
@@ -11,7 +11,7 @@ Then the problem is formulated as:
 
 $$
 \min_{u,c} \sum_{i=1,2,\cdots,N} \|u(x_i, t_i)-u_i\|^2\\
-s.t. \frac{\partial^2u}{\partial t^2}=c\frac{\partial^2u}{\partial x^2}
+s.t. \frac{\partial^2u}{\partial t^2}=c^2\frac{\partial^2u}{\partial x^2}
 $$
 
 In the context of PINN, $u$ is parameterized to $u_\theta$. 
@@ -20,7 +20,7 @@ The problem above is transformed to the discrete form:
 $$
 \min_{\theta,c} 
  w_1\sum_{i=1,2,\cdots,N} \|u_\theta(x_i, t_i)-u_i\|^2
-+w_2\sum_{i=1,2,\cdots,M}\left|\frac{\partial^2u_\theta(x_i,t_i)}{\partial t^2}-c\frac{\partial^2u_\theta(x_i,t_i)}{\partial x^2}\right|^2.
++w_2\sum_{i=1,2,\cdots,M}\left|\frac{\partial^2u_\theta(x_i,t_i)}{\partial t^2}-c^2\frac{\partial^2u_\theta(x_i,t_i)}{\partial x^2}\right|^2.
 $$
 
 ## Importing External Data

examples/consolidation/Aana.py

@@ -0,0 +1,25 @@
+import idrlnet.shortcut as sc
+import math
+import sympy as sp
+import numpy as np
+import matplotlib.pyplot as plt
+
+order = 10
+s = 0
+h = 1
+cv = 0.6
+x, t = sp.symbols('x t')
+
+for k in range(1, order + 1):
+    s += (-1) ** (k - 1) / (2 * k - 1) * sp.cos((2 * k - 1) * math.pi * x / 2 / h) * sp.exp(
+        -(2 * k - 1) ** 2 * math.pi ** 2 / 4 * cv * t / h / h)
+p_ratio = s * 4 / math.pi
+p_ratio_fn = sc.lambdify_np(p_ratio, ['t', 'x'])
+
+if __name__ == '__main__':
+    t_num = np.linspace(0.01, 1, 100, endpoint=True)
+    z_num = np.linspace(0.01, 1, 100, endpoint=True)
+    tt_num, zz_num = np.meshgrid(t_num, z_num)
+    p_ratio_num = p_ratio_fn(tt_num, zz_num)
+    plt.scatter(tt_num, zz_num, c=p_ratio_num, cmap='jet', vmin=0, vmax=1)
+    plt.show()

examples/consolidation/Boneside_pinn.py

@@ -0,0 +1,124 @@
+import idrlnet.shortcut as sc
+import sympy as sp
+import numpy as np
+import matplotlib.pyplot as plt
+from mpl_toolkits.axes_grid1 import make_axes_locatable
+from Aana import p_ratio_fn
+
+cv = 0.6
+
+x, y = sp.symbols('x y')
+p = sp.Function('p')(x, y)
+geo = sc.Rectangle((0, 0), (1, 1))
+
+
+@sc.datanode
+class Init(sc.SampleDomain):
+    def __init__(self):
+        self.points = geo.sample_boundary(100, sieve=(sp.Eq(y, 0.)))
+        self.constraints = {'p': 1., 'lambda_p': 1 - np.power(self.points['x'], 5)}
+
+    def sampling(self, *args, **kwargs):
+        return self.points, self.constraints
+
+
+@sc.datanode
+class Interior(sc.SampleDomain):
+    def __init__(self):
+        self.points = geo.sample_interior(10000)
+        self.constraints = {'consolidation': np.zeros_like(self.points['x'])}
+
+    def sampling(self, *args, **kwargs):
+        return self.points, self.constraints
+
+
+@sc.datanode
+class UpperBounds(sc.SampleDomain):
+    def __init__(self):
+        self.points = geo.sample_boundary(100, sieve=(sp.Eq(x, 1.)))
+        self.constraints = {'p': 0.,
+                            'lambda_p': np.power(self.points['y'], 0.2)}
+
+    def sampling(self, *args, **kwargs):
+        return self.points, self.constraints
+
+
+@sc.datanode
+class LowerBounds(sc.SampleDomain):
+    def __init__(self):
+        self.points = geo.sample_boundary(100, sieve=(sp.Eq(x, 0.)))
+        self.constraints = {'normal_gradient_p': 0., }
+
+    def sampling(self, *args, **kwargs):
+        return self.points, self.constraints
+
+
+pde = sc.ExpressionNode(name='consolidation', expression=p.diff(y) - cv * p.diff(x, 2))
+grad = sc.NormalGradient(T='p', dim=2, time=False)
+net = sc.get_net_node(inputs=('x', 'y'), outputs=('p',), arch=sc.Arch.mlp, name='net')
+
+s = sc.Solver(sample_domains=(Init(), Interior(), UpperBounds(), LowerBounds()),
+              netnodes=[net],
+              pdes=[pde, grad],
+              max_iter=2000)
+s.solve()
+
+
+@sc.datanode
+class MeshInterior(sc.SampleDomain):
+    def sampling(self, *args, **kwargs):
+        t_num = np.linspace(0.01, 1, 400, endpoint=True)
+        z_num = np.linspace(0.01, 1, 100, endpoint=True)
+        tt_num, zz_num = np.meshgrid(t_num, z_num)
+        points = {'x': zz_num.reshape(-1, 1), 'y': tt_num.reshape(-1, 1)}
+        return points, {}
+
+
+s.sample_domains = (MeshInterior(),)
+points = s.infer_step({'MeshInterior': ['x', 'y', 'p']})
+x_num = points['MeshInterior']['x'].detach().cpu().numpy().ravel()
+y_num = points['MeshInterior']['y'].detach().cpu().numpy().ravel()
+p_num = points['MeshInterior']['p'].detach().cpu().numpy().ravel()
+
+_, ax = plt.subplots(3, 1, figsize=(10, 10))
+
+im = ax[0].scatter(y_num, x_num, c=p_num, vmin=0, vmax=1, cmap='jet')
+ax[0].set_xlim([0, 1.01])
+ax[0].set_ylim([-0.05, 1.05])
+ax[0].set_ylabel('z(m)')
+divider = make_axes_locatable(ax[0])
+cax = divider.append_axes('right', size='2%', pad=0.1)
+plt.colorbar(im, cax=cax)
+ax[0].set_title('Model Prediction: $p_{pred}=p/p_0(z,t)$')
+
+ax[1].set_xlim([0, 1.01])
+ax[1].set_ylim([-0.05, 1.05])
+ax[1].set_ylabel('z(m)')
+ax[1].set_title('Ground Truth: $p_{true}=p/p_0(z,t)$')
+t_num = np.linspace(0.01, 1, 400, endpoint=True)
+z_num = np.linspace(0.01, 1, 100, endpoint=True)
+tt_num, zz_num = np.meshgrid(t_num, z_num)
+
+p_ratio_num = p_ratio_fn(tt_num, zz_num)
+
+im = ax[1].scatter(tt_num, zz_num, c=p_ratio_num, cmap='jet', vmin=0, vmax=1.)
+divider = make_axes_locatable(ax[1])
+cax = divider.append_axes('right', size='2%', pad=0.1)
+plt.colorbar(im, cax=cax)
+
+im = ax[2].scatter(tt_num, zz_num, c=p_num.reshape(100, 400) - p_ratio_num, cmap='bwr', vmin=-1,
+                   vmax=1)
+ax[2].set_xlim([0, 1.01])
+ax[2].set_ylim([-0.05, 1.05])
+ax[2].set_xlabel('t(yr)')
+ax[2].set_ylabel('z(m)')
+ax[2].set_title('Error: $p_{pred}-p_{true}$')
+
+divider = make_axes_locatable(ax[2])
+cax = divider.append_axes('right', size='2%', pad=0.1)
+cbar = plt.colorbar(im, cax=cax)
+
+plt.tight_layout(pad=0.4, w_pad=0.5, h_pad=1.0)
+plt.savefig('test.png')
+plt.show()
+plt.close()

examples/consolidation/readme.md

@@ -0,0 +1,2 @@
+`Aana.py` generate the ground truth via truncated analytical expression.
+`Boneside.py` solves the consolidation equation.

examples/holes_heat_flux/holes_heat_flux.py

@@ -0,0 +1,234 @@
+import idrlnet.shortcut as sc
+import torch
+import matplotlib.pyplot as plt
+import numpy as np
+import sympy as sp
+import abc
+from matplotlib import tri
+
+sc.use_gpu(device=0)
+
+INTERVAL = 10000
+DENSITY = 10000
+
+r1, r2, r3, r4, r5, r6 = sp.symbols('r1 r2 r3 r4 r5 r6')
+
+plate = sc.Tube2D((-1, -0.5), (1, 0.5))
+hole_1 = sc.Circle(center=(-0.6, 0), radius=r1)
+hole_2 = sc.Circle(center=(0., 0.), radius=r2)
+hole_3 = sc.Circle(center=(0.5, -0.5), radius=r3)
+hole_4 = sc.Circle(center=(0.5, 0.5), radius=r4)
+hole_5 = sc.Circle(center=(-0.5, -0.5), radius=r5)
+hole_6 = sc.Circle(center=(-0.5, 0.5), radius=r6)
+geo = plate - hole_1 - hole_2 - hole_3 - hole_4 - hole_5 - hole_6
+
+in_line = sc.Line((-1, -0.5), (-1, 0.5), normal=1)
+out_line = sc.Line((1, -0.5), (1, 0.5), normal=1)
+param_ranges = {r1: (0.05, 0.2),
+                r2: (0.05, 0.2),
+                r3: (0.05, 0.2),
+                r4: (0.05, 0.2),
+                r5: (0.05, 0.2),
+                r6: (0.05, 0.2), }
+
+
+class ReSampleDomain(sc.SampleDomain, metaclass=abc.ABCMeta):
+    """
+    Resampling collocated points every INTERVAL iterations.
+    """
+    count = 0
+    points = sc.Variables()
+    constraints = sc.Variables()
+
+    def sampling(self, *args, **kwargs):
+        if self.count % INTERVAL == 0:
+            self.do_re_sample()
+            sc.logger.info("Resampling...")
+        self.count += 1
+        return self.points, self.constraints
+
+    @abc.abstractmethod
+    def do_re_sample(self):
+        pass
+
+
+@sc.datanode
+class InPlane(ReSampleDomain):
+    def do_re_sample(self):
+        self.points = sc.Variables(
+            in_line.sample_boundary(param_ranges=param_ranges, density=DENSITY,
+                                    low_discrepancy=True)).to_torch_tensor_()
+        self.constraints = {'T': torch.ones_like(self.points['x'])}
+
+
+@sc.datanode
+class OutPlane(ReSampleDomain):
+    def do_re_sample(self):
+        self.points = sc.Variables(
+            out_line.sample_boundary(param_ranges=param_ranges, density=DENSITY,
+                                     low_discrepancy=True)).to_torch_tensor_()
+        self.constraints = {'T': torch.zeros_like(self.points['x'])}
+
+
+@sc.datanode(sigma=10.)
+class Boundary(ReSampleDomain):
+    def do_re_sample(self):
+        self.points = sc.Variables(
+            geo.sample_boundary(param_ranges=param_ranges, density=DENSITY, low_discrepancy=True)).to_torch_tensor_()
+        self.constraints = {'normal_gradient_T': torch.zeros_like(self.points['x'])}
+
+
+@sc.datanode
+class Interior(ReSampleDomain):
+    def do_re_sample(self):
+        self.points = sc.Variables(
+            geo.sample_interior(param_ranges=param_ranges, density=DENSITY, low_discrepancy=True)).to_torch_tensor_()
+        self.constraints = {'diffusion_T': torch.zeros_like(self.points['x'])}
+
+
+net = sc.get_net_node(inputs=('x', 'y', 'r1', 'r2', 'r3', 'r4', 'r5', 'r6'), outputs=('T',), name='net',
+                      arch=sc.Arch.mlp_xl)
+pde = sc.DiffusionNode(T='T', D=1., Q=0, dim=2, time=False)
+grad = sc.NormalGradient('T', dim=2, time=False)
+
+s = sc.Solver(sample_domains=(InPlane(), OutPlane(), Boundary(), Interior()),
+              netnodes=[net],
+              pdes=[pde, grad],
+              max_iter=100000,
+              schedule_config=dict(scheduler='ExponentialLR', gamma=0.99998))
+
+s.solve()
+
+
+# Define inference domains.
+
+@sc.datanode
+class Inference(sc.SampleDomain):
+    def sampling(self, *args, **kwargs):
+        self.points = sc.Variables(
+            geo.sample_interior(param_ranges=param_ranges, density=20000, low_discrepancy=True)).to_torch_tensor_()
+        self.constraints = {'T__x': torch.zeros_like(self.points['x']),
+                            'T__y': torch.zeros_like(self.points['x']), }
+        return self.points, self.constraints
+
+
+@sc.datanode
+class BoundaryInference(sc.SampleDomain):
+    def sampling(self, *args, **kwargs):
+        self.points = sc.Variables(
+            geo.sample_boundary(param_ranges=param_ranges, density=1000, low_discrepancy=True)).to_torch_tensor_()
+        self.constraints = {'T__x': torch.zeros_like(self.points['x']),
+                            'T__y': torch.zeros_like(self.points['x']), }
+        return self.points, self.constraints
+
+
+@sc.datanode
+class InPlane(sc.SampleDomain):
+    def sampling(self, *args, **kwargs):
+        self.points = sc.Variables(
+            in_line.sample_boundary(param_ranges=param_ranges, density=1000,
+                                    low_discrepancy=True)).to_torch_tensor_()
+        self.constraints = {'T__x': torch.zeros_like(self.points['x']),
+                            'T__y': torch.zeros_like(self.points['x']), }
+        return self.points, self.constraints
+
+
+@sc.datanode
+class OutPlane(sc.SampleDomain):
+    def sampling(self, *args, **kwargs):
+        self.points = sc.Variables(
+            out_line.sample_boundary(param_ranges=param_ranges, density=1000,
+                                     low_discrepancy=True)).to_torch_tensor_()
+        self.constraints = {'T__x': torch.zeros_like(self.points['x']),
+                            'T__y': torch.zeros_like(self.points['x']), }
+        return self.points, self.constraints
+
+
+s.sample_domains = (InPlane(), OutPlane(), Inference(), BoundaryInference())
+
+count = [0]
+
+
+def parameter_design(*args):
+    """
+    Do inference and plot the result.
+    """
+
+    param_ranges[r1] = args[0]
+    param_ranges[r2] = args[1]
+    param_ranges[r3] = args[2]
+    param_ranges[r4] = args[3]
+    param_ranges[r5] = args[4]
+    param_ranges[r6] = args[5]
+
+    points = s.infer_step({'Inference': ['x', 'y', 'T__x', 'T__y', ],
+                           'BoundaryInference': ['x', 'y', 'T__x', 'T__y', ],
+                           'InPlane': ['x', 'y', 'T__x', 'T__y', ],
+                           'OutPlane': ['x', 'y', 'T__x', 'T__y', ]})
+
+    plt.figure(figsize=(8, 4))
+    fig = plt.gcf()
+    fig.set_tight_layout(True)
+    ########
+    num_x = points['BoundaryInference']['x'].detach().cpu().numpy().ravel()
+    num_y = points['BoundaryInference']['y'].detach().cpu().numpy().ravel()
+
+    num_T__x = points['BoundaryInference']['T__x'].detach().cpu().numpy().ravel()
+    num_T__y = points['BoundaryInference']['T__y'].detach().cpu().numpy().ravel()
+
+    num_flux = np.sqrt(num_T__x ** 2 + num_T__y ** 2)
+    plt.scatter(x=num_x, y=num_y, c=num_flux, s=3, vmin=0, vmax=0.8, cmap='bwr')
+
+    ########
+    num_x = points['InPlane']['x'].detach().cpu().numpy().ravel()
+    num_y = points['InPlane']['y'].detach().cpu().numpy().ravel()
+
+    num_T__x = points['InPlane']['T__x'].detach().cpu().numpy().ravel()
+    num_T__y = points['InPlane']['T__y'].detach().cpu().numpy().ravel()
+
+    num_flux = np.sqrt(num_T__x ** 2 + num_T__y ** 2)
+    plt.scatter(x=num_x, y=num_y, c=num_flux, s=3, vmin=0, vmax=0.8, cmap='bwr')
+
+    ########
+    num_x = points['OutPlane']['x'].detach().cpu().numpy().ravel()
+    num_y = points['OutPlane']['y'].detach().cpu().numpy().ravel()
+
+    num_T__x = points['OutPlane']['T__x'].detach().cpu().numpy().ravel()
+    num_T__y = points['OutPlane']['T__y'].detach().cpu().numpy().ravel()
+
+    num_flux = np.sqrt(num_T__x ** 2 + num_T__y ** 2)
+    plt.scatter(x=num_x, y=num_y, c=num_flux, s=3, vmin=0, vmax=0.8, cmap='bwr')
+
+    ########
+    num_x = points['Inference']['x'].detach().cpu().numpy().ravel()
+    num_y = points['Inference']['y'].detach().cpu().numpy().ravel()
+
+    num_T__x = points['Inference']['T__x'].detach().cpu().numpy().ravel()
+    num_T__y = points['Inference']['T__y'].detach().cpu().numpy().ravel()
+
+    num_flux = np.sqrt(num_T__x ** 2 + num_T__y ** 2)
+    points['Inference']['T_flux'] = num_flux
+
+    triang = tri.Triangulation(num_x, num_y)
+
+    def apply_mask(triang, alpha=0.4):
+        triangles = triang.triangles
+        xtri = num_x[triangles] - np.roll(num_x[triangles], 1, axis=1)
+        ytri = num_y[triangles] - np.roll(num_y[triangles], 1, axis=1)
+        maxi = np.max(np.sqrt(xtri ** 2 + ytri ** 2), axis=1)
+        triang.set_mask(maxi > alpha)
+
+    apply_mask(triang, alpha=0.04)
+    plt.tricontourf(triang, num_flux, 100, vmin=0, vmax=0.8, cmap='bwr')
+
+    ax = plt.gca()
+    ax.set_facecolor('k')
+    ax.set_xticks([])
+    ax.set_yticks([])
+    ax.set_xlim([-1, 1.])
+    ax.set_ylim([-0.5, 0.5])
+    plt.savefig("holes.png")
+    plt.close()
+
+
+parameter_design(0.14, 0.1, 0.2, 0.09, 0.05, 0.17)

examples/ldc/ldc.py

@@ -0,0 +1,115 @@
+from idrlnet import shortcut as sc
+import sympy as sp
+import torch
+import numpy as np
+from idrlnet.pde_op.equations import NavierStokesNode
+from matplotlib import tri
+import matplotlib.pyplot as plt
+
+x, y = sp.symbols('x y')
+rec = sc.Rectangle((-0.05, -0.05), (0.05, 0.05))
+
+
+@sc.datanode(name='flow_domain')
+class InteriorDomain(sc.SampleDomain):
+    def __init__(self):
+        points = rec.sample_interior(400000, bounds={x: (-0.05, 0.05), y: (-0.05, 0.05)})
+
+        constraints = sc.Variables({'continuity': torch.zeros(len(points['x']), 1),
+                                    'momentum_x': torch.zeros(len(points['x']), 1),
+                                    'momentum_y': torch.zeros(len(points['x']), 1)})
+
+        points['area'] = np.ones_like(points['area']) * 2.5e-6
+
+        constraints['lambda_continuity'] = points['sdf']
+        constraints['lambda_momentum_x'] = points['sdf']
+        constraints['lambda_momentum_y'] = points['sdf']
+        self.points = sc.Variables(points).to_torch_tensor_()
+        self.constraints = sc.Variables(constraints).to_torch_tensor_()
+
+    def sampling(self, *args, **kwargs):
+        return self.points, self.constraints
+
+
+@sc.datanode(name='left_right_down')
+class LeftRightDownBoundaryDomain(sc.SampleDomain):
+    def __init__(self):
+        points = rec.sample_boundary(3333, sieve=(y < 0.05))
+        constraints = sc.Variables({'u': torch.zeros(len(points['x']), 1), 'v': torch.zeros(len(points['x']), 1)})
+        points['area'] = np.ones_like(points['area']) * 1e-4
+
+        self.points = sc.Variables(points).to_torch_tensor_()
+        self.constraints = sc.Variables(constraints).to_torch_tensor_()
+
+    def sampling(self, *args, **kwargs):
+        return self.points, self.constraints
+
+
+@sc.datanode(name='up')
+class UpBoundaryDomain(sc.SampleDomain):
+    def __init__(self):
+        points = rec.sample_boundary(10000, sieve=sp.Eq(y, 0.05))
+        points['area'] = np.ones_like(points['area']) * 1e-4
+        constraints = sc.Variables({'u': torch.ones(len(points['x']), 1), 'v': torch.zeros(len(points['x']), 1)})
+        constraints['lambda_u'] = 1 - 20 * abs(points['x'].copy())
+        self.points = sc.Variables(points).to_torch_tensor_()
+        self.constraints = sc.Variables(constraints).to_torch_tensor_()
+
+    def sampling(self, *args, **kwargs):
+        return self.points, self.constraints
+
+
+torch.autograd.set_detect_anomaly(True)
+net = sc.MLP([2, 100, 100, 100, 100, 3], activation=sc.Activation.tanh, initialization=sc.Initializer.Xavier_uniform,
+             weight_norm=False)
+
+net_u = sc.NetNode(inputs=('x', 'y',), outputs=('u', 'v', 'p'), net=net, name='net_u')
+pde = NavierStokesNode(nu=0.01, rho=1.0, dim=2, time=False)
+
+s = sc.Solver(sample_domains=(InteriorDomain(), LeftRightDownBoundaryDomain(), UpBoundaryDomain()),
+              netnodes=[net_u],
+              pdes=[pde],
+              max_iter=4000,
+              network_dir='./result/tanh_Xavier_uniform',
+              )
+s.solve()
+
+
+def interoir_domain_infer():
+    points = rec.sample_interior(1000000, bounds={x: (-0.05, 0.05), y: (-0.05, 0.05)})
+    constraints = sc.Variables({'continuity': torch.zeros(len(points['x']), 1),
+                                'momentum_x': torch.zeros(len(points['x']), 1),
+                                'momentum_y': torch.zeros(len(points['x']), 1)})
+    return points, constraints
+
+
+data_infer = sc.get_data_node(interoir_domain_infer, name='flow_domain')
+s.sample_domains = [data_infer]
+
+pred = s.infer_step({'flow_domain': ['x', 'y', 'v', 'u', 'p']})
+num_x = pred['flow_domain']['x'].detach().cpu().numpy().ravel()
+num_y = pred['flow_domain']['y'].detach().cpu().numpy().ravel()
+num_u = pred['flow_domain']['u'].detach().cpu().numpy().ravel()
+num_v = pred['flow_domain']['v'].detach().cpu().numpy().ravel()
+num_p = pred['flow_domain']['p'].detach().cpu().numpy().ravel()
+triang_total = tri.Triangulation(num_x, num_y)
+
+triang_total = tri.Triangulation(num_x, num_y)
+u_pre = num_u.flatten()
+
+fig = plt.figure(figsize=(15, 5))
+ax1 = fig.add_subplot(131)
+tcf = ax1.tricontourf(triang_total, num_u, 100, cmap="jet")
+tc_bar = plt.colorbar(tcf)
+ax1.set_title('u')
+
+ax2 = fig.add_subplot(132)
+tcf = ax2.tricontourf(triang_total, num_v, 100, cmap="jet")
+tc_bar = plt.colorbar(tcf)
+ax2.set_title('v')
+
+ax3 = fig.add_subplot(133)
+tcf = ax3.tricontourf(triang_total, num_p, 100, cmap="jet")
+tc_bar = plt.colorbar(tcf)
+ax3.set_title('p')
+plt.show()

idrlnet/__init__.py

@@ -1,16 +1,43 @@
 import torch
+from .header import logger
 
+GPU_AVAILABLE = False
+GPU_ENABLED = False
 # todo more careful check
-GPU_ENABLED = True
 if torch.cuda.is_available():
     try:
         _ = torch.Tensor([0.0, 0.0]).cuda()
-        torch.set_default_tensor_type("torch.cuda.FloatTensor")
-        print("gpu available")
-        GPU_ENABLED = True
+        logger.info("GPU available")
+        GPU_AVAILABLE = True
     except:
-        print("gpu not available")
-        GPU_ENABLED = False
+        logger.info("GPU not available")
+        GPU_AVAILABLE = False
 else:
-    print("gpu not available")
-    GPU_ENABLED = False
+    logger.info("GPU not available")
+    GPU_AVAILABLE = False
+
+
+def use_gpu(device=0):
+    """Use GPU with device `device`.
+
+    Args:
+        device (torch.device or int): selected device.
+    """
+    if GPU_AVAILABLE:
+        try:
+            torch.cuda.set_device(device)
+            torch.set_default_tensor_type("torch.cuda.FloatTensor")
+            logger.info(f"Using GPU device {device}")
+            global GPU_ENABLED
+            GPU_ENABLED = True
+        except:
+            logger.warning("Invalid device ordinal")
+
+
+def use_cpu():
+    """
+    Use CPU.
+    """
+    if GPU_ENABLED:
+        torch.set_default_tensor_type("torch.FloatTensor")
+    logger.info(f"Using CPU")

idrlnet/data.py

@@ -223,7 +223,7 @@ def get_data_nodes(funs: List[Callable], *args, **kwargs) -> Tuple[DataNode]:
 
 
 class SampleDomain(metaclass=abc.ABCMeta):
-    """Template for Callable sampling function."""
+    """The Template for Callable sampling functions."""
 
     @abc.abstractmethod
     def sampling(self, *args, **kwargs):

idrlnet/geo_utils/geo_obj.py

@@ -4,7 +4,7 @@ import math
 from math import pi
 from typing import Union, List, Tuple
 import numpy as np
-from sympy import symbols, Abs, sqrt, Max, Min, cos, sin, log, sign, Heaviside
+from sympy import symbols, Abs, sqrt, Max, Min, cos, sin, log, sign, Heaviside, asin
 from sympy.vector import CoordSys3D
 from .geo import Edge, Geometry1D, Geometry2D, Geometry3D
 
@@ -662,11 +662,21 @@ class Box(Geometry3D):
 
 class Sphere(Geometry3D):
     def __init__(self, center, radius):
-        x, y, z, v_1, v_2, u_1, u_2 = symbols("x y z v_1 v_2 u_1 u_2")
-        ranges = {v_1: (0, 1), v_2: (0, 1), u_1: (0, 1), u_2: (0, 1)}
-        r_1 = sqrt(-log(v_1)) * cos(2 * pi * u_1)
-        r_2 = sqrt(-log(v_1)) * sin(2 * pi * u_1)
-        r_3 = sqrt(-log(v_2)) * cos(2 * pi * u_2)
+        x, y, z, v_1, v_2 = symbols("x y z v_1 v_2")
+        ranges = {v_1: (0, 1), v_2: (0, 1)}
+
+        theta = 2 * pi * v_1
+        r = sqrt(v_2)
+
+        phi = 2 * asin(r)  # latitude
+        r_1 = cos(theta) * sin(phi)
+        r_2 = sin(theta) * sin(phi)
+        r_3 = cos(phi)
+        # x, y, z, v_1, v_2, u_1, u_2 = symbols("x y z v_1 v_2 u_1 u_2")
+        # ranges = {v_1: (0, 1), v_2: (0, 1), u_1: (0, 1), u_2: (0, 1)}
+        # r_1 = sqrt(-log(v_1)) * cos(2 * pi * u_1)
+        # r_2 = sqrt(-log(v_1)) * sin(2 * pi * u_1)
+        # r_3 = sqrt(-log(v_2)) * cos(2 * pi * u_2)
 
         norm = sqrt(r_1 ** 2 + r_2 ** 2 + r_3 ** 2)
         edge_1 = Edge(

idrlnet/shortcut.py

@@ -10,4 +10,4 @@ from idrlnet.callbacks import GradientReceiver
 from idrlnet.receivers import Receiver, Signal
 from idrlnet.variable import Variables, export_var
 from idrlnet.header import logger
-from idrlnet import GPU_ENABLED
+from idrlnet import GPU_AVAILABLE, GPU_ENABLED, use_gpu

requirements.txt

@@ -19,3 +19,4 @@ flask==1.1.2
 requests==2.25.0
 torch>=1.7.1
 networkx==2.5.1
+protobuf~=3.20

setup.py

@@ -21,7 +21,7 @@ def load_requirements(path_dir=here, comment_char="#"):
 
 setuptools.setup(
     name="idrlnet",  # Replace with your own username
-    version="0.0.1",
+    version="0.1.0",
     author="Intelligent Design & Robust Learning lab",
     author_email="weipeng@deepinfar.cn",
     description="IDRLnet",