#Change Data_Base_Updated_Final_Rotated_Train.json to Database_Random…

….json

examples/active_learning/QBC_SVC.py

@@ -165,7 +165,7 @@ def plot_variances(var_list):
     plt.ylabel('Variance')
     plt.title('Variance vs Iteration')
     plt.grid()
-    plt.savefig('variances_vs_iterations.png', dpi = 300)
+    plt.savefig('variances_vs_iterations_weight=999.png', dpi = 300)
     plt.close()
 
 def save_hard_test_cases(a , b, num_tests=5):
@@ -180,7 +180,7 @@ def save_hard_test_cases(a , b, num_tests=5):
 
 nmembers=5  # Number of committee members
  # Number of initial samples - can be chosen by the user
-nsamples_to_generate=70 #  Number of iterations
+nsamples_to_generate=30 #  Number of iterations
 sampling_scheme='max_disagreement'  # max disagreement for yf-predictions, for classifiers generally possible: vote_entropy, consensus_entropy or maximum_disagreement, cf. https://modal-python.readthedocs.io/en/latest/content/query_strategies/Disagreement-sampling.html#disagreement-sampling
 subset_percentage=0.8
 subset_assignment='random'
@@ -194,11 +194,12 @@ def save_hard_test_cases(a , b, num_tests=5):
 mat_h.elasticity(E = E, nu = nu)
 mat_h.plasticity(sy = sy, hill = hill)
 mat_h.calc_properties(eps = 0.0013, sigeps = True)
-c = 8
-d = 22
-N = c+d
+c = 200
+d = 99
+N = 200
 nsamples_init = N
-sunit= FE.load_cases(number_3d=c, number_6d=d)
+# sunit= FE.load_cases(number_3d=c, number_6d=d)
+sunit = creator_rnd(200,8)
 np.savetxt('Test_Cases.txt', sunit)
 # create set of unit stresses and
 print('Created {0} unit stresses (6d Voigt tensor).'.format(N))

examples/active_learning/QBC_SVC_Batch.py

@@ -216,7 +216,7 @@ def plot_variances(var_list):
     plt.ylabel('Variance')
     plt.title('Variance vs Iteration')
     plt.grid()
-    plt.savefig('variances_vs_iterations.png', dpi = 300)
+    plt.savefig('variances_vs_iterations_weight=999.png', dpi = 300)
     plt.close()
 
 def save_hard_test_cases(a , b, num_tests=5):
@@ -231,7 +231,7 @@ def save_hard_test_cases(a , b, num_tests=5):
 
 nmembers=5  # Number of committee members
 # Number of initial samples - can be chosen by the user
-nsamples_to_generate=70 #  Number of iterations
+nsamples_to_generate=50 #  Number of iterations
 sampling_scheme='max_disagreement'  # max disagreement for yf-predictions, for classifiers generally possible: vote_entropy, consensus_entropy or maximum_disagreement, cf. https://modal-python.readthedocs.io/en/latest/content/query_strategies/Disagreement-sampling.html#disagreement-sampling
 subset_percentage=0.8
 subset_assignment='random'
@@ -245,15 +245,16 @@ def save_hard_test_cases(a , b, num_tests=5):
 mat_h.elasticity(E = E, nu = nu)
 mat_h.plasticity(sy = sy, hill = hill)
 mat_h.calc_properties(eps = 0.0013, sigeps = True)
-c = 8
-d = 22
+c = 1
+d = 99
 N = c+d
 nsamples_init = N
 sunit_random= FE.load_cases(number_3d=c, number_6d=d)
 sunit = apply_repulsion(sunit_random, k=5, iterations=60, learning_rate=0.01)
 np.savetxt('Test_Cases.txt', sunit)
 final_knn_distances = calculate_knn_distances(sunit, k =5)
 average_distance = np.mean(final_knn_distances)
+print('Average distance to 5th nearest neighbor: {0:.4f}'.format(average_distance))
 # create set of unit stresses and
 print('Created {0} unit stresses (6d Voigt tensor).'.format(N))
 x1=fsolve(find_yloc, np.ones(N) * mat_h.sy, args = (sunit, mat_h), xtol = 1.e-5)
@@ -274,7 +275,7 @@ def save_hard_test_cases(a , b, num_tests=5):
         else:
             raise NotImplementedError('chosen subset assignment not implemented')
         mat_ml=FE.Material(name = 'ML-Hill_{}'.format(j))
-        mat_ml.train_SVC(C = C, gamma = gamma, sdata = sig[idx, :], gridsearch = True)
+        mat_ml.train_SVC(C = C, gamma = gamma, sdata = sig[idx, :], gridsearch = False)
         committee.append(mat_ml)
 
     # Search for next unit vector to query
@@ -285,7 +286,7 @@ def save_hard_test_cases(a , b, num_tests=5):
         res=differential_evolution(
             eval_max_disagreement,
             bounds,
-            args = (committee, sunit_new_list, average_distance, 999, 'input_comparison', mat_h, sig),
+            args = (committee, sunit_new_list, average_distance, 99, 'input_comparison', mat_h, sig),
             popsize = 90,
             polish = True,
             updating = 'immediate'
@@ -303,6 +304,7 @@ def save_hard_test_cases(a , b, num_tests=5):
         variance=res.fun
         final_knn_distances=calculate_knn_distances(sunit, k = 5)
         average_distance=np.mean(final_knn_distances)
+        print('Average distance to 5th nearest neighbor: {0:.4f}'.format(average_distance))
 
     if i == nsamples_to_generate - 1:
         np.savetxt('DATA_sig_iter_{}.txt'.format(i + 1), sig)
@@ -311,7 +313,7 @@ def save_hard_test_cases(a , b, num_tests=5):
     C=2
     gamma=2.5
     mat_ml=FE.Material(name = 'ML-Hill')  # define material
-    mat_ml.train_SVC(C = C, gamma = gamma, sdata = sig, gridsearch = True)
+    mat_ml.train_SVC(C = C, gamma = gamma, sdata = sig, gridsearch = False)
     # stress strain curves
     print("Calculating properties of ML material, this might take a while ...")
     mat_ml.elasticity(E = E, nu = nu)

examples/data_6d/Data_6D.py

@@ -19,7 +19,7 @@ def rgb_to_hex(rgb):
 
 
 # Import Data
-db = FE.Data("Data_Base_Updated_Goss.json", wh_data=True) #"Data_Base_Updated_Final_Rotated_Train.JSON"
+db = FE.Data("Data_Base_Updated_Final_Rotated_Train.JSON", wh_data=True) #"Data_Base_Updated_Final_Rotated_Train.JSON"
 #db = FE.Data("Data_Base_UpdatedE-07.json", Work_Hardening=False)
 mat_ref = FE.Material(name="reference")  # define reference material, J2 plasticity, linear w.h.
 mat_ref.elasticity(E=db.mat_data['E_av'], nu=db.mat_data['nu_av'])
@@ -36,10 +36,10 @@ def rgb_to_hex(rgb):
 print(f'Training successful.\nNumber of support vectors: {len(mat_ml.svm_yf.support_vectors_)}')
 
 # # Testing
-# sig_tot, epl_tot, yf_ref = CTD.Create_Test_Sig(Json="Data_Base_Updated_Final_Rotated_Test.json")
-# yf_ml = mat_ml.calc_yf(sig_tot, epl_tot, pred=False)
-# Results = CTD.training_score(yf_ref, yf_ml)
-# print(Results)
+sig_tot, epl_tot, yf_ref = CTD.Create_Test_Sig(Json="Data_Base_Updated_Final_Rotated_Train.JSON")
+yf_ml = mat_ml.calc_yf(sig_tot, epl_tot, pred=False)
+Results = CTD.training_score(yf_ref, yf_ml)
+print(Results)
 #Plot Hardening levels over a meshed space
 #Plot initial and final hardening level of trained ML yield function together with data points
 ngrid = 100
@@ -57,8 +57,8 @@ def rgb_to_hex(rgb):
 Z2 = mat_ml.calc_yf(sig=Cart_hh_6D, epl=normalized_grad_hh * 0.005, pred=False)
 Z3 = mat_ml.calc_yf(sig=Cart_hh_6D, epl=normalized_grad_hh * 0.01, pred=False)
 Z4 = mat_ml.calc_yf(sig=Cart_hh_6D, epl=normalized_grad_hh * 0.015, pred=False)
-Z5 = mat_ml.calc_yf(sig=Cart_hh_6D, epl=normalized_grad_hh * 0.02, pred=False)
-Z6 = mat_ml.calc_yf(sig=Cart_hh_6D, epl=normalized_grad_hh * 0.025, pred=False)
+Z5 = mat_ml.calc_yf(sig=Cart_hh_6D, epl=normalized_grad_hh * 0.018, pred=False)
+# Z6 = mat_ml.calc_yf(sig=Cart_hh_6D, epl=normalized_grad_hh * 0.025, pred=False)
 
 colors_hex = ['#550000', '#990000', '#bb0000', '#cc3333', '#ee3333', '#ff5050']
 fig = plt.figure(figsize=(4.2, 4.2))
@@ -69,18 +69,18 @@ def rgb_to_hex(rgb):
 line3 = mat_ml.plot_data(Z3, ax, xx, yy, c=colors_hex[2])
 line4 = mat_ml.plot_data(Z4, ax, xx, yy, c=colors_hex[3])
 line5 = mat_ml.plot_data(Z5, ax, xx, yy, c=colors_hex[4])
-line6 = mat_ml.plot_data(Z6, ax, xx, yy, c=colors_hex[5])
+# line6 = mat_ml.plot_data(Z6, ax, xx, yy, c=colors_hex[5])
 fig.savefig('Hardening_Levels.png', dpi=300)
 handle1 = Line2D([], [], color=colors_hex[0], label='Equivalent Plastic Strain : 0 ')
 handle2 = Line2D([], [], color=colors_hex[1], label='Equivalent Plastic Strain : 0.5% ')
 handle3 = Line2D([], [], color=colors_hex[2], label='Equivalent Plastic Strain : 1% ')
 handle4 = Line2D([], [], color=colors_hex[3], label='Equivalent Plastic Strain : 1.5% ')
 handle5 = Line2D([], [], color=colors_hex[4], label='Equivalent Plastic Strain : 1.8% ')
-handle6 = Line2D([], [], color=colors_hex[5], label='Equivalent Plastic Strain : 2.5% ')
+# handle6 = Line2D([], [], color=colors_hex[5], label='Equivalent Plastic Strain : 2.5% ')
 fig_leg = plt.figure(figsize=(4, 4))
 ax_leg = fig_leg.add_subplot(111)
 ax_leg.axis('off')
-ax_leg.legend(handles=[handle1, handle2, handle3, handle4, handle5, handle6], loc="center")
+ax_leg.legend(handles=[handle1, handle2, handle3, handle4, handle5], loc="center")
 fig_leg.savefig('Legend.png', dpi=300)
 plt.show()
 #
@@ -104,7 +104,7 @@ def rgb_to_hex(rgb):
 
 # Reconstruct Stress-Strain Curve
 Keys = list(db.Data_Visualization.keys())
-Key = "Us_A1B2C2D1E1F2_4092b_5e411_Tx_Rnd"  # random.choice(Keys) # Select Data from database randomly
+Key = "Us_A2B2C1D0E0F0_592bb_0abb1_Tx_Gs"  # random.choice(Keys) # Us_A2B2C2D2E1F1_09ad0_0abb1_Tx_Gs
 print("Selected Key is: {}".format(Key))
 Stresses = db.Data_Visualization[Key]["Stress"]
 Eq_Stresses = db.Data_Visualization[Key]["Eq_Stress"]
@@ -163,9 +163,9 @@ def rgb_to_hex(rgb):
 # get stress data
 peeq_dat = FE.eps_eq(db.mat_data['plastic_strain'])
 # ind0 = np.nonzero(peeq_dat < 0.0002)[0]
-ind0 =np.nonzero(np.logical_and(peeq_dat > 0.00018, peeq_dat < 0.00022))[0]
+ind0 =np.nonzero(np.logical_and(peeq_dat > 0.00019, peeq_dat < 0.00021))[0]
 sig_d0 = FE.s_cyl(db.mat_data['flow_stress'][ind0, :], mat_ml)
-ind1 = np.nonzero(np.logical_and(peeq_dat > 0.0248, peeq_dat < 0.0252))[0]
+ind1 = np.nonzero(np.logical_and(peeq_dat > 0.0149, peeq_dat < 0.0151))[0] #0.0248, 0.0252
 sig_d1 = FE.s_cyl(db.mat_data['flow_stress'][ind1, :], mat_ml)
 # calculate ML flow stresses
 ngrid = 100
@@ -193,7 +193,7 @@ def rgb_to_hex(rgb):
 fig.savefig('ML+ScatterData.png', dpi=300)
 plt.show()
 handle1 = mlines.Line2D([], [], color="#550000", label='Equivalent Plastic Strain : 0 ')
-handle2 = mlines.Line2D([], [], color="#ff3333", label='Equivalent Plastic Strain : 2.5% ')
+handle2 = mlines.Line2D([], [], color="#ff3333", label='Equivalent Plastic Strain : 1.5% ')
 fig_leg = plt.figure(figsize=(4, 4))
 ax_leg = fig_leg.add_subplot(111)
 ax_leg.axis('off')

examples/data_6d/Data_6D_DOE.py

@@ -24,7 +24,7 @@ def rgb_to_hex(rgb):
     return '#{:02x}{:02x}{:02x}'.format(int(rgb[0]*255), int(rgb[1]*255), int(rgb[2]*255))
 
 #Training
-db = FE.Data("Data_Base_Updated_Final_Rotated_Train.json", wh_data=True)
+db = FE.Data("Database_Random.json", wh_data=True)
 mat_ref = FE.Material(name="reference") # define reference material, J2 plasticity, linear w.h.
 mat_ref.elasticity(E=db.mat_data['E_av'], nu=db.mat_data['nu_av'])             # identic elastic properties as mat1
 mat_ref.plasticity(sy=db.mat_data['sy_av'], khard= 4.5e3)        # same yield strength as mat1 and mat2, high w.h. coefficient 4.5e3)

examples/data_6d/Data_6D_new.py

@@ -25,7 +25,7 @@ def find_yloc(x, sunit, epl, mat):
 
 
 # Import Data
-db = FE.Data("Data_Base_Updated_Final_Rotated_Train.json", wh_data=True)
+db = FE.Data("Database_Random.json", wh_data=True)
 # db = FE.Data("Data_Base_UpdatedE-07.json", Work_Hardening=False)
 # define reference material, J2 plasticity, linear w.h.
 mat_ref = FE.Material(name="reference")

examples/data_6d/Stress_Strain_Calculation.py

@@ -9,7 +9,7 @@
 import pylabfea as FE
 
 # Import Data
-db = FE.Data("Data_Base_Updated_Final_Rotated_Train.json", wh_data=True)
+db = FE.Data("Database_Random.json", wh_data=True)
 
 # db.plot_yield_locus(db=db, mat_data=db.mat_data, active='flow_stress')
 print(f'Successfully imported data for {db.mat_data["Nlc"]} load cases')

src/pylabfea/data.py

@@ -89,7 +89,7 @@ def __init__(self, source, path_data='./',
                  name='Dataset', mat_name="Simulanium",
                  sdim=6,
                  epl_crit=2.e-3,
-                 epl_start=1.e-3, epl_max=0.02,
+                 epl_start=1.e-3, epl_max=0.01,
                  plot=False,
                  wh_data=True):
         if sdim!=3 and sdim!=6: