Removed README.md from datadriven tutorial

Author: EvertBunschoten

compressible_flow/NICFD_nozzle/DataDriven/README.md

@@ -0,0 +1,69 @@
+#!/usr/bin/env python
+
+## \file Tensroflow2SU2.py
+#  \brief Example of converting a trained Tensorflow MLP to SU2 .mlp
+#         format.
+#  \author E.C.Bunschoten
+#  \version 7.5.0 "Blackbird"
+#
+# SU2 Project Website: https://su2code.github.io
+#
+# The SU2 Project is maintained by the SU2 Foundation
+# (http://su2foundation.org)
+#
+# Copyright 2012-2022, SU2 Contributors (cf. AUTHORS.md)
+#
+# SU2 is free software; you can redistribute it and/or
+# modify it under the terms of the GNU Lesser General Public
+# License as published by the Free Software Foundation; either
+# version 2.1 of the License, or (at your option) any later version.
+#
+# SU2 is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+# Lesser General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public
+# License along with SU2. If not, see <http://www.gnu.org/licenses/>.
+
+import tensorflow as tf 
+from write_SU2_MLP import *
+
+# Input names for the SU2 MLP. For the CDataDrivenFluid class, these have to be "Density" and "Energy"
+MLP_input_names = ["Density", "Energy"]
+# The order of the input names should correspond to that of the trained model inputs.
+
+# Optional: give minimum and maximum input normalization values. These are required in case the MLP
+# was trained on normalized input data.
+input_min = [2.81038e-1, 1.79956e5]
+input_max = [2.72979e+01, 4.48318e+05]
+
+# Output names for the SU2 MLP. For the CDataDrivenFluid class, these have to include the following:
+MLP_output_names = ["s",        # Entropy
+                    "dsde_rho", # First entropy derivative w.r.t static energy
+                    "dsdrho_e", # First entropy derivative w.r.t density
+                    "d2sde2",   # Second entropy derivative w.r.t static energy
+                    "d2sdedrho",# Second entropy derivative w.r.t static energy and density
+                    "d2sdrho2"] # Second entropy derivative w.r.t density
+# The order of the output names should correspond to that of the trained model outputs.
+
+# Optional: give minimum and maximum output normalization values. These are required in case the MLP
+# was trained on normalized training data.
+output_min = [ 5.74729e+03,  1.66861e-03, -1.05632e+03, -2.15459e-08,
+       -4.20494e-06,  3.96917e-01]
+
+output_max = [ 7.77852e+03,  4.00000e-03, -1.07148e+01, -3.57370e-09,
+       -4.93767e-07,  3.75786e+03]
+
+# Saved Tensorflow model file name, saved with the "model.save" function after training.
+Tensorflow_model_file = "MLP_Air"
+
+# SU2 MLP output file name (wo extension)
+SU2_MLP_output_file = "MLP_Air_SU2"
+
+# Load the MLP trained through Tensorflow
+Tensorflow_model = tf.keras.models.load_model(Tensorflow_model_file)
+
+# Write SU2 MLP input file
+write_SU2_MLP(SU2_MLP_output_file, MLP_input_names, MLP_output_names, Tensorflow_model, input_min, input_max, output_min, output_max)
+

compressible_flow/NICFD_nozzle/DataDriven/Tensorflow2SU2.py

@@ -0,0 +1,137 @@
+#!/usr/bin/env python
+
+## \file write_SU2_MLP.py
+#  \brief Python script for translating a trained Tensorflow model 
+#         to an SU2 MLP input file.
+#  \author E.C.Bunschoten
+#  \version 7.5.0 "Blackbird"
+#
+# SU2 Project Website: https://su2code.github.io
+#
+# The SU2 Project is maintained by the SU2 Foundation
+# (http://su2foundation.org)
+#
+# Copyright 2012-2022, SU2 Contributors (cf. AUTHORS.md)
+#
+# SU2 is free software; you can redistribute it and/or
+# modify it under the terms of the GNU Lesser General Public
+# License as published by the Free Software Foundation; either
+# version 2.1 of the License, or (at your option) any later version.
+#
+# SU2 is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+# Lesser General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public
+# License along with SU2. If not, see <http://www.gnu.org/licenses/>.
+
+
+def write_SU2_MLP(file_out, input_names, output_names, model, input_min=[], input_max=[], output_min=[], output_max=[]):
+    # This function writes the MLP to a format which can be read by the SU2 MLP import tool
+    # Inputs:
+    # - file_out: output file name without extension
+    # - input_names: list of strings with the variable names of the MLP input(s)
+    # - output names: list of strings with the variable names of the MLP output(s)
+    # - model: tensorflow.keras.model; the trained model
+    # - input_min: lower normalization values for the input
+    # - input_max: upper normalization values for the input
+    # - output_min: lower normalization values for the output
+    # - output_max: upper normalization values for the output
+
+    # MLP config
+    model_config = model.get_config()
+
+    # Number of input variables in the model
+    n_inputs = model_config['layers'][0]['config']['batch_input_shape'][1]
+    # Number of output variables in the model
+    n_outputs = model_config['layers'][-1]['config']['units']
+
+    # Checking if number of provided input and output names are equal to those in the model
+    if not n_inputs == len(input_names):
+        raise Exception("Number of provided input names unequal to the number of inputs in the model")
+    if not n_outputs == len(output_names):
+        raise Exception("Number of provided output names unequal to the number of outputs in the model")
+
+    if len(input_max) != len(input_min):
+        raise Exception("Upper and lower input normalizations should have the same length")
+    if len(output_max) != len(output_min):
+        raise Exception("Upper and lower output normalizations should have the same length")
+
+    if len(input_max) > 0 and len(input_min) != n_inputs:
+        raise Exception("Input normalization not provided for all inputs")
+
+    if len(output_max) > 0 and len(output_min) != n_outputs:
+        raise Exception("Output normalization not provided for all outputs")
+
+
+    # Creating output file
+    fid = open(file_out+'.mlp', 'w+')
+    fid.write("<header>\n\n")
+    n_layers = len(model_config['layers'])
+
+    # Writing number of neurons per layer
+    fid.write('[number of layers]\n%i\n\n' % n_layers)
+    fid.write('[neurons per layer]\n')
+    activation_functions = []
+
+    for iLayer in range(n_layers-1):
+        layer_class = model_config['layers'][iLayer]['class_name']
+        if layer_class == 'InputLayer':
+            # In case of the input layer, the input shape is written instead of the number of units
+            activation_functions.append('linear')
+            n_neurons = model_config['layers'][iLayer]['config']['batch_input_shape'][1]
+        else:
+            activation_functions.append(model_config['layers'][iLayer]['config']['activation'])
+            n_neurons = model_config['layers'][iLayer]['config']['units']
+
+        fid.write('%i\n' % n_neurons)
+    fid.write('%i\n' % n_outputs)
+
+    activation_functions.append('linear')
+
+    # Writing the activation function for each layer
+    fid.write('\n[activation function]\n')
+    for iLayer in range(n_layers):
+        fid.write(activation_functions[iLayer] + '\n')
+
+    # Writing the input and output names
+    fid.write('\n[input names]\n')
+    for input in input_names:
+        fid.write(input + '\n')
+    
+    if len(input_min) > 0:
+        fid.write('\n[input normalization]\n')
+        for i in range(len(input_names)):
+            fid.write('%+.16e\t%+.16e\n' % (input_min[i], input_max[i]))
+    
+    fid.write('\n[output names]\n')
+    for output in output_names:
+        fid.write(output+'\n')
+    
+    if len(output_min) > 0:
+        fid.write('\n[output normalization]\n')
+        for i in range(len(output_names)):
+            fid.write('%+.16e\t%+.16e\n' % (output_min[i], output_max[i]))
+
+    fid.write("\n</header>\n")
+    # Writing the weights of each layer
+    fid.write('\n[weights per layer]\n')
+    for layer in model.layers:
+        fid.write('<layer>\n')
+        weights = layer.get_weights()[0]
+        for row in weights:
+            fid.write("\t".join(f'{w:+.16e}' for w in row) + "\n")
+        fid.write('</layer>\n')
+    
+    # Writing the biases of each layer
+    fid.write('\n[biases per layer]\n')
+    
+    # Input layer biases are set to zero
+    fid.write('%+.16e\t%+.16e\t%+.16e\n' % (0.0, 0.0, 0.0))
+
+    for layer in model.layers:
+        biases = layer.get_weights()[1]
+        fid.write("\t".join([f'{b:+.16e}' for b in biases]) + "\n")
+
+    fid.close()