Minor Update

datasets/generate_test_case_I.py

@@ -0,0 +1,83 @@
+import os
+import sys
+module_path = os.path.abspath(os.path.join('../'))
+module_path = os.path.abspath(os.path.join('../src'))
+if module_path not in sys.path:
+    sys.path.append(module_path)
+
+# sph related imports
+from BasisConvolution.test_case_I.sph import initSimulation, runSimulation
+from BasisConvolution.test_case_I.perlin import generate1DPeriodicNoise
+from BasisConvolution.test_case_I.io import export
+# from BasisConvolution.oneDimensionalSPH.plotting import plotSimulationState, regularPlot
+# neural network rlated imports
+from torch.optim import Adam
+# from rbfConv import *
+# from torch_geometric.loader import DataLoader
+# from trainingHelper import *
+# plotting/UI related imports
+import matplotlib as mpl
+# plt.style.use('dark_background')
+cmap = mpl.colormaps['viridis']
+from tqdm.autonotebook import tqdm
+from IPython.display import display, Latex
+from datetime import datetime
+import h5py
+import torch
+import numpy as np
+
+
+numParticles = 2048 # Modify th number of desird particles here, this value is defined here to ensure that the linar sampling for some pdfs is accurate.
+generator = 'perlin'
+generatorSettings = {'r' : 0.75, 'freq' : 1, 'octaves' : 2, 'seed' : 1234, 'mag' : 0.25, 'offset': 2}
+noise  = generatorSettings['offset'] + generate1DPeriodicNoise(numSamples = numParticles, r = generatorSettings['r'], freq = generatorSettings['freq'], octaves = generatorSettings['octaves'], plot = False, seed = generatorSettings['seed']) * generatorSettings['mag']
+pdf = lambda x : np.interp(x, np.linspace(-1,1,numParticles), noise)
+
+def pdf(x):
+    x = np.array(x)
+    out = np.array(np.ones_like(x))
+    out[x > 0] = x[x>0] + 0
+    out[x < 0] = x[x<0] + 2
+    return out + 1
+
+# simulation parameters
+minDomain = -1 # minimum domain, leave at -1 for the most part
+maxDomain = 1 # maximum domain, leave at 1 for the most part
+# change base area to change initial starting density
+baseArea = 2 / numParticles * 2
+particleRadius = baseArea / 2.0
+# change particle support to make simulation more/less smooth
+particleSupport = particleRadius * 8.
+# SPH parameters
+xsphConstant = 0.0
+diffusionAlpha = 1. # kinematic viscosity coefficient
+diffusionBeta = 2.
+kappa = 10 # EOS kappa term
+restDensity = 1000 # EOS rest density term
+dt = 1e-3 # fixed global timestep
+c0 = 10 # speed of sound used in kinematic viscosity
+
+timesteps = 2048 # timesteps to be simulated
+# display(Latex(f'Estimated kinematic diffusion $\\mu\\approx\\frac{{1}}{{2(d+2)}}\\alpha c_s h = \\frac{{1}}{{2(1+2)}} \\cdot {diffusionAlpha} \\cdot {c0} \\cdot {particleSupport:4.2e} = {1/6 * diffusionAlpha * c0 * particleSupport:4.2e}$'))
+
+device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+dtype = torch.float64 if device == 'cpu' else torch.float32
+# print('Running on ', device, 'using type', dtype)
+
+torch.manual_seed(10957120378)
+seeds = torch.randint(0,2**30, (1,36)).flatten()
+
+seeds = [141077155, 9771992, 521260098, 943800868, 879567972, 843734504, 302569625, 172735587, 549857803, 123272218, 734684074, 926567716, 927288476, 883194790, 497511951, 341412464, 86702325, 31736609, 773229209, 379424249, 577570728, 323111077, 581356742, 712262378, 884632587, 450979520, 15165885, 159763383, 882368412, 783832494, 137881214, 686823772, 433611252, 960745502, 759352143, 37078787]
+
+print(seeds)
+
+
+for seed in tqdm(seeds):
+    generator = 'perlin'
+    generatorSettings = {'r' : 0.75, 'freq' : 1, 'octaves' : 2, 'seed' : int(seed), 'mag' : 0.25, 'offset': 2}
+    noise  = generatorSettings['offset'] + generate1DPeriodicNoise(numSamples = numParticles, r = generatorSettings['r'], freq = generatorSettings['freq'], octaves = generatorSettings['octaves'], plot = False, seed = generatorSettings['seed']) * generatorSettings['mag']
+    pdf = lambda x : np.interp(x, np.linspace(-1,1,numParticles), noise)
+
+    fluidPositions, fluidAreas, fluidVelocities = initSimulation(pdf, numParticles, minDomain, maxDomain, baseArea, particleSupport, dtype, device, plot = False)
+    simulationStates = runSimulation(fluidPositions, fluidAreas, fluidVelocities, timesteps, minDomain, maxDomain, kappa, restDensity, diffusionAlpha, diffusionBeta, c0, xsphConstant, particleSupport, dt)
+    export(simulationStates, numParticles, timesteps, minDomain, maxDomain, kappa, restDensity, diffusionAlpha, diffusionBeta, c0, xsphConstant, particleRadius, baseArea, particleSupport, dt, generator, generatorSettings, folder = '../datasets/test_case_I',nameOverride = True)

src/BasisConvolution/test_case_I/train.py

@@ -44,7 +44,7 @@ def trainModel(particleData, testData, settings, dataSet, trainingFiles, offset,
     windowFn = getWindowFunction(window, norm = windowNorm) if window is not None else None
     model = BasisNetwork(fluidFeatures = fluidFeatures, 
                    layers = layers, 
-                   denseLayer = True, activation = 'ReLU', coordinateMapping = 'cartesian', 
+                   denseLayer = True, activation = 'relu', coordinateMapping = 'cartesian', 
                    dims = [n], windowFn = windowFn, rbfs = [basis], batchSize = 32, ignoreCenter = True, normalized = False).to(device)   
     lr = initialLR
     with torch.no_grad():