Skip to content

Latest commit

 

History

History
25 lines (22 loc) · 1.45 KB

readme.md

File metadata and controls

25 lines (22 loc) · 1.45 KB

Data structure

Current utils expects the dataset to be structured as follows. For each sub-set there must be a sub-directory. The minimum required sub-sets are:

  • train: used for computing the POD i.e. finding the reduced basis, as well as training the regression models.
  • validate: used solely for evaluating the performance during the hypertuning of the ANN.
  • test: used for evaluating the perfomance of the best ANN configuration and the other two regression models, as well as the projection error.

Each sub-set must contain:

  • parameters.txt: containing an array of parameter samples of dimension N_set x N_d.
  • truth_i for i=0,..,N_subset: snapshots in binary consisting of numpy.float64 data types (C doubles) using big-endian byteorder. Each snapshot can be loaded into a numpy vector via e.g. np.load('dataset/train/truth_0', np.float64()).byteswap() (see utils.load_snapshot()). Each vector contains all degrees of freedom. E.g. for a steady problem and components u, v, p (such as the skewed lid-driven-cavity) the vector is as follows:

u \
v    node 0
p /
u \
v    node 1
p /
.
.
.
u \
v    node #nodes-1
p /

For time-dependent problems, this pattern is repeated for every time step, starting at the initial condition. This corresponds to a #components*#nodes*(#timesteps+1) vector.