Implemented legacy version for active learning

Author: AHartmaier

examples/Active_Learning/qbc_svc.py

@@ -244,18 +244,20 @@ def hard_test_cases(a, b):
 sig = sunit * x1[:, None]
 print('Calculated {} yield stresses.'.format(nsamples_init))
 # train SVC with yield stress data generated from Hill flow rules
-C = 6.0
-gamma = 2.0
-Fe = 0.5
-Ce = 0.99
-Nseq = 20
+C = 2.0
+gamma = 2.5
+Ce = 0.95
+Fe = 0.1
+Nseq = 25
 vlevel = 0
-cvals = [5., 6., 8.]
-gvals = [1., 2., 3.]
+cvals = [4, 6, 8, 10]  # [4, 6, 8, 10, 12, 15]
+gvals = [0.6, 1, 2, 2.5, 3]  # [0.6, 0.8, 1, 2, 2.5, 3]
 
 mat_ml = FE.Material(name='ML-Hill')  # define material
-mat_ml.train_SVC(C=C, gamma=gamma, Fe=Fe, Ce=Ce, Nseq=Nseq, sdata=sig,
-                 gridsearch=True, cvals=[1., 2., 4.], gvals=[0.5, 1., 1.5], verbose=vlevel)
+mat_ml.dev_only = True
+mat_ml.train_SVC(C=C, gamma=gamma, Fe=Fe, Ce=Ce, Nseq=Nseq, sdata=sig, extend=False,
+                 gridsearch=True, cvals=cvals, gvals=gvals,
+                 al_legacy=True, verbose=vlevel)
 plot_yield_locus(mat_ml=mat_ml, mat_h=mat_h, niter=0)
 np.savetxt('DATA_sig_iter_0.txt', sig)
 np.savetxt('DATA_sunit_iter_0.txt', sunit)
@@ -270,9 +272,11 @@ def hard_test_cases(a, b):
     for j in range(nmembers):
         idx = np.random.choice(np.arange(sig.shape[0]), int(sig.shape[0] * subset_percentage), replace=False)
         mat_ml = FE.Material(name='ML-Hill_{}'.format(j))
+        mat_ml.dev_only = True
         mat_ml.train_SVC(C=C, gamma=gamma, Fe=Fe, Ce=Ce, Nseq=Nseq,
-                         sdata=sig[idx, :],
-                         gridsearch=True, cvals=cvals, gvals=gvals, verbose=vlevel)
+                         sdata=sig[idx, :], extend=False,
+                         gridsearch=True, cvals=cvals, gvals=gvals,
+                         al_legacy=True, verbose=vlevel)
         hyp_C_list.append(mat_ml.C_yf)
         hyp_g_list.append(mat_ml.gam_yf)
         committee.append(mat_ml)
@@ -294,9 +298,11 @@ def hard_test_cases(a, b):
     sunit = np.vstack([sunit, sunit_new])
 
 mat_ml = FE.Material(name='ML-Hill')  # define material
+mat_ml.dev_only = True
 mat_ml.train_SVC(C=C, gamma=gamma, Fe=Fe, Ce=Ce, Nseq=Nseq,
-                 sdata=sig,
-                 gridsearch=True, cvals=cvals, gvals=gvals, verbose=vlevel)
+                 sdata=sig, extend=False,
+                 gridsearch=True, cvals=cvals, gvals=gvals,
+                 al_legacy=True, verbose=vlevel)
 
 # Create ML model with conventional training approach
 Ntot = nsamples_init + nsamples_to_generate

src/pylabfea/material.py

@@ -1342,7 +1342,7 @@ def train_SVC(self, C=10, gamma=4, Nlc=36, Nseq=25, fs=0.3, extend=False,
                   mat_ref=None, sdata=None, plot=False, fontsize=16,
                   gridsearch=False, cvals=None, gvals=None, Fe=0.1, Ce=0.99, scaler=None, pca=None,
                   train_index=None, test_index=None, verbose=1, metric='acc', pca_dim=10,
-                  reversal=None):
+                  reversal=None, al_legacy=False):
         """Train SVC for all yield functions of the microstructures provided
         in msparam and for flow stresses to capture work hardening. In first
         step, the training data for each set is generated by creating stresses
@@ -1852,7 +1852,8 @@ def test_data_generation(self, C=10, gamma=4, Nlc=36, Nseq=25, fs=0.3, extend=Fa
         return xt, yt
 
     def create_sig_data(self, N=None, mat_ref=None, sdata=None, Nseq=2, sflow=None,
-                        offs=0.01, extend=False, rand=False, Fe=0.1, Ce=0.99):
+                        offs=0.01, extend=False, rand=False, Fe=0.1, Ce=0.99,
+                        al_legacy=False):
         """Function to create consistent data sets on the deviatoric stress plane
         for training or testing of ML yield function. Either the number "N" of raw data points, i.e. load angles, to be 
         generated and a reference material "mat_ref" has to be provided, or a list of raw data points "sdata" with 
@@ -1945,6 +1946,9 @@ def create_sig_data(self, N=None, mat_ref=None, sdata=None, Nseq=2, sflow=None,
             elastic_seq = np.linspace(Fe, Ce, Nseq)
             plastic_seq = np.linspace(2. - Ce, 2. - Fe, Nseq)
             seq = np.append(elastic_seq, plastic_seq)
+        if al_legacy:
+            seq = np.linspace(offs, 0.95, Nseq)  # old version ???
+            seq = np.append(seq, np.linspace(1.05, 2., Nseq))
         if extend:
             # add training points in plastic regime to avoid fallback of SVC decision fct. to zero
             seq = np.append(seq, np.array([2.4, 3., 4., 5.]))