![]() ![]() Once the test data is in place, execute python. make sure to reduce it to 3 when evaluating a model trained by 'runTrainCpu.py'. The model exponent expo is set in the runTest.py script, so e.g. Hence, you either have to generate data in a new directory with theĭataGen.py script from above, or download the test data set via the link below. runTest.py script.īy default, it assumes that the test data samples (with the same file format as the training samples)Īre located in. To compute relative inference errors for a test data set, you can use the. Optionally, you can also save txt files with the loss progression (see saveL1 in the script). If you don't have a working GPU, you can use 'runTrainCpu.py' to train a smaller model on the CPU.Ī sample image will be generated for each epoch in the results_train directory. ![]() Once the script has finished, it will save the trained model as modelG. Validation loss is printed during training, and should decrease significantly. With 10k iterations on the GPU, loading all data that is available in. By default, this will execute a short training run Switch to the directory containing the training scripts, i.e. Reference data data.targets has the channels. To summarize, in the TurDataset class the inputs data.inputs have the channels, while the The last three channels represent the target, containing one pressure and two velocity The first three channels represent the input,Ĭonsisting (in this order) of two fields corresponding to the freestream velocities in x and yĭirection and one field containing a mask of the airfoil geometry asĪ mask. Sample file is 6x128x128 with dimensions: channels, x, y. Output files are saved as compressed numpy arrays. The samples variables to generate more samples with a single call.įor a first test, 100 samples are sufficient, for higher quality models, more You can call this script repeatedly to generate npz files in a newĭirectory called train. Once dataGen.py has finished, you should find 100. This script executes openfoam and runs gmsh for meshing the airfoil profiles. dataGen.py to generate a first set of 100 airfoils. (The latter contains a subset that shouldn't be used for training.) TheĪirfoild database should contain 1498 files afterwards. Will create airfoil_database and airfoil_database_test directories. Download airfoilsĭownload the airfoil profiles by running. In the source directory, then you can continue with the training step. Simply make sure you have data/train and data/test Note that you can skip the next two steps if you download the trainingĭata packages below. (Details can be found on the installation pages of PyTorch and ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |