Commit b2287f3a authored by ajuria's avatar ajuria
Browse files

Added config files

parent 0d31de09
####################
# SIMULATION #
####################
simClass: Plume
GPU: True
sim_method: convnet # Choose between convnet and CG (as reference)
#Field saving options
save_field: True
save_field_x_ite: 10
save_post_x_ite: 10
#Plot options
plot_field: True
plot_x_ite: 50
#Post-computations options
post_computations: False
out_dir: './output/dir/'
####################
# PHYSICAL FORCES #
####################
Richardson: 0.1
gravity: 1.0
gravity_x: 0
gravity_y: 1
####################
# DISCRETIZATION #
####################
Nx: 128 #[] number of control volumes in x direction
Ny: 128 #[] number of control volumes in y direction
Nt: 1000 #[] number of time steps to simulate
# CFL
CFL: 0.2
####################
# SOLVER IA #
####################
ite_transition: 0
network_params:
load_path: '/path/to/neurasim/trained_networks/lt_nograd_4_16/Unet_lt_nograd_4_16/'
model_name: 'Unet_lt_nograd_4_16'
new_train: 'new'
####################
# NORMALIZATION #
####################
normalization:
normalize: True
scale_factor: 10.0
debug_folder: './results/debug/'
####################
# GEOMETRY #
####################
#Domain
Lx: 128
Ly: 128
#BC
BC_domain_x: OPEN
BC_domain_y: STICKY
#Cilinder
cylinder: False
D: 10
yD: 150
input_rad: 0.145
input_vel: 1.0
\ No newline at end of file
####################
# SIMULATION #
####################
simClass: Plume
GPU: True
sim_method: convnet # Choose between convnet and CG (as reference)
#Field saving options
save_field: True
save_field_x_ite: 10
save_post_x_ite: 10
#Plot options
plot_field: True
plot_x_ite: 50
#Post-computations options
post_computations: False
out_dir: './output/dir/'
####################
# PHYSICAL FORCES #
####################
Richardson: 0.1
gravity: 1.0
gravity_x: 0
gravity_y: 1
####################
# DISCRETIZATION #
####################
Nx: 128 #[] number of control volumes in x direction
Ny: 128 #[] number of control volumes in y direction
Nt: 1000 #[] number of time steps to simulate
# CFL
CFL: 0.2
####################
# SOLVER IA #
####################
ite_transition: 0
network_params:
load_path: '/path/to/neurasim/trained_networks/lt_nograd_4_16/Unet_lt_nograd_4_16/'
model_name: 'Unet_lt_nograd_4_16'
new_train: 'new' # Option to read networks trained with older versions, not to be modified in this scope
####################
# NORMALIZATION #
####################
normalization:
normalize: True
scale_factor: 10.0
debug_folder: './results/debug/'
####################
# GEOMETRY #
####################
#Domain
Lx: 128
Ly: 128
#BC
BC_domain_x: OPEN
BC_domain_y: STICKY
#Cilinder
cylinder: True
D: 20
yD: 80
input_rad: 0.145
input_vel: 1.0
\ No newline at end of file
####################
# SIMULATION #
####################
simClass: VonKarman_rotative
GPU: True
sim_method: convnet # Choose between convnet and CG (as reference)
#Field saving options
save_field: True
save_field_x_ite: 50
save_post_x_ite: 50
#Plot options
plot_field: True
plot_x_ite: 50
#Post-computations options
post_computations: True
out_dir: './output/dir/'
####################
# PHYSICAL FORCES #
####################
Reynolds: 100.0
Alpha: 0.0 # Rotating dimensionless parameter!
####################
# DISCRETIZATION #
####################
Nx: 896 #[] number of control volumes in x direction
Ny: 608 #[] number of control volumes in y direction
Nt: 10000 #[] number of time steps to simulate
# CFL
CFL: 0.2
####################
# SOLVER IA #
####################
ite_transition: 0
network_params:
load_path: '/path/to/neurasim/trained_networks/lt_nograd_4_16/Unet_lt_nograd_4_16/'
model_name: 'Unet_lt_nograd_4_16'
new_train: 'new' # Option to read networks trained with older versions, not to be modified in this scope
####################
# NORMALIZATION #
####################
normalization:
normalize: True
scale_factor: 10.0
debug_folder: './results/debug/'
####################
# GEOMETRY #
####################
#Domain
Lx: 300
Ly: 200
#BC
BC_domain_x: OPEN
BC_domain_y: STICKY
#Cilinder
D: 10
xD: 100
# Configuration file with default parameters.
# Some can be modified through the command line. See help function for training
# script and README.md for more info.
# This table is saved to disk (as pytorch objects) on every epoch
# so that simulations can be paused and restarted.
#=========================================
# MODEL
#=========================================
#=========================================
# DATA
#=========================================
# dataDir : Dataset location
dataDir: "/absolute/path/to/data/datasets/"
# dataset : Dataset name. Folder inside dataDir with training and testing scenes
dataset: "dataset_name"
# numWorkers : number of parallel workers for dataloader. Set to 0 to allow PyTorch
# to automatically manage loading.
numWorkers: 3
# If true, dataset is preprocessed and programs exists.
# Preprocessing is automatic if no previous preproc is detected on current dataset.
preprocOriginalFluidNetDataOnly: false
# shuffleTraining : Shuffles dataset
shuffleTraining: true
#=========================================
# OUTPUT
#=========================================
# modelDir : Output folder for trained model and loss log.
modelDir: "/absolute/path/to/save/your/model/modelname"
# modelFilename : Trained model name
modelFilename: "convModel"
#=========================================
# TRAINING MONITORING
#=========================================
# freqToFile : Epoch frequency for loss output to file/image saving.
freqToFile: 25
# printTraining : Debug options for training.
# Prints or shows validation dataset and compares net
# output to GT.
# Options: save (save figures), show (shows in windows), none
printTraining: "save"
#=========================================
# TRAINING PARAMETER
#=========================================
batchSize: 64
# maxEpochs : Maximum number of epochs
maxEpochs: 1000
# resume : resume training from checkpoint.
resumeTraining: false
modelParam:
# model : options ('FluidNet', 'ScaleNet')
# -FluidNet : uses the architecture found in lib/model.py (based on FluidNet)
# -ScaleNet : uses a multiscale architecture found in lib/multi_scale_net.py
model: "ScaleNet"
# inputChannels : Network inputs. At least one of them must be set to true!
inputChannels:
div: true
pDiv: false
UDiv: false
# lr : learning rate. If using scientific notation, necessary to precise type
# for yaml->python cast.
lr: !!python/float 5e-5
# fooLambda : Weighting for each loss. Set to 0 to disable loss.
# MSE of pressure
pL2Lambda: 0
# MSE of divergence (Ground truth is zero divergence)
divL2Lambda: 1
# Absolute difference of pressure
pL1Lambda: 0
# Absolute difference of divergence
divL1Lambda: 0
# MSE of long term divergence
# If > 0, implements the Long Term divergence concept from FluidNet
divLongTermLambda: 5
# Differentiable long term loss, or ordinary lt (data augmentation)
ltGrad: false
# longTermDivNumSteps : We want to measure what the divergence is after
# a set number of steps for each training and test sample. Set table
# to nil to disable, (or set longTermDivLambda to 0).
longTermDivNumSteps:
- 2
- 4
# longTermDivProbability is the probability that longTermDivNumSteps[0]
# will be taken, otherwise longTermDivNumSteps[1] will be taken with
# probability of 1 - longTermDivProbability.
longTermDivProbability: 0.9
# normalizeInput : if true, normalizes input by max(std(chan), threshold)
normalizeInput: true
# normalizeInputChan : which channel to calculate std
normalizeInputChan: "UDiv"
# normalizeInputThreshold : don't normalize input noise
normalizeInputThreshold: 0.00001
# normalizing scale factor
scale_factor: 10
# Dictionary for normalization
normalization:
normalize: True
scale_factor: 10.0
debug_folder: "/absolute/path/for/debugging"
#=========================================
# PHYSICAL PARAMETERS
#=========================================
# Time step: default simulation timestep.
dt: 0.1
# Resolution of domain (it must match the data coming from the data loader!)
nnx: 128
nny: 128
# ONLY APPLIED IF LONG TERM DIV IS ACTIVATED
# ----------------------------------
# buoyancyScale : Buoyancy forces scale
# gravityScale : Gravity forces scale
# Note: Manta and FluidNet divide gravity forces into "gravity" and "buoyancy"
# They represent the two terms arising from Boussinesq approximation
# rho*g = rho_0*g + delta_rho*g
# (1) (2)
# rho_0 being the average density and delta_rho local difference of density
# w.r.t average density.
# Mantaflow calls (1) gravity and (2) buoyancy and allows for different g's
buoyancyScale: 0
gravityScale: 0
# Gravity vector: Direction of gravity Vector
gravityVec:
x: 0
y: 0
z: 0
# training buoyancy scale : This is the buoyancy to use when adding buoyancy
# to the long term training. It will be applied in a random cardinal direction.
trainBuoyancyScale: 0. #2.0
# training buoyancy probability : This is the probability to add buoyancy when
# long term training.
trainBuoyancyProb: 0. #0.3
# training gravity scale : This is the gravity to use when adding gravity
# to the long term training. It will be applied in a random cardinal direction.
trainGravityScale: 2.0
# training gravity probability : This is the probability to add buoyancy when
# long term training.
trainGravityProb: 0.3
# ------------------------------------
# Introduces a correcting factor in the denisty equation
# from "A splitting method for incompressible flows with variable
# density based on a pressure Poisson equation" (Guermond, Salgado).
# Not really tested... Recommendation is to leave it as false.
correctScalar: false
# operatingDensity : When applying buoyancy, buoyancyScale is multiplied
# by (density(i,j) - operatingDensity)
operatingDensity: 0.0
# viscosity : introduces a viscous term in moment equation.
# Algortihm taken from the book "Fluid Simulation for Computer Graphics" by
# Bridson
viscosity: 0
# timeScaleSigma : Amplitude of time scale perturb during training.
timeScaleSigma: 1
# maccormackStrength : used in semi-lagrangian MacCormack advection
# when LT div is activated. 0.6 is a good value. If ~1, can lead to
# high frequency artifacts.
maccormackStrength: 0.6
# sampleOutsideFluid : if true, allows particles in advection to 'land' inside
# obstacles. In general, we don't want that, so leave it as false to avoid
# possible artifacts.
sampleOutsideFluid: false
#=========================================
# SIMULATION PARAMETERS
#=========================================
sim_phi:
# GPU utilization
GPU: True
# Domain Discretization
Nx: 128 #[] number of control volumes in x direction
Ny: 128 #[] number of control volumes in y direction
Nt: 1 #[] number of time steps to simulate
#Domain
Lx: 128 #[m] or in mm in consistently changed
Ly: 128 #[m]
# CFL
CFL: 0.2
# time
dt: 1
# Choose between network and CG, set to NN for the training
sim_method: 'convnet' # CG or convnet
# Network normalization and debugging
normalization:
normalize: True
scale_factor: 10.0
debug_folder: "/absolute/path/for/debugging"
# Network to load for lt simulations (matches the network been trained)
network_params:
load_path: "/absolute/path/where/model/is/saved"
model_name: 'modelname'
new_train: 'new' # Legacy option to be deleted
# For debugging purposes
in_dir: './'
out_dir: './'
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment