def plot_Hot_point_wise_err(point_wise_err,
size_vec2mat=20,
actName=None,
seedNo=1000,
outPath=None):
# 逐点误差分布热力图
square_err_color2sin = np.reshape(point_wise_err,
(size_vec2mat, size_vec2mat))
plt.figure(figsize=(10, 8))
ax = plt.gca()
plt.imshow(square_err_color2sin,
interpolation='nearest',
cmap=cm.coolwarm,
origin='lower')
plt.colorbar(shrink=0.85)
plt.xticks(())
plt.yticks(())
# # plt.title('point-wise error', fontsize=14)
# if actName == 'sReLU':
# fntmp = '%s/%spErr_%s' % (outPath, seedNo, 'sReLU')
# elif actName == 's2ReLU':
# fntmp = '%s/%spErr_%s' % (outPath, seedNo, 's2ReLU')
# elif actName == 'ReLU':
# fntmp = '%s/%spErr_%s' % (outPath, seedNo, 'ReLU')
fntmp = '%s/%spErr_%s' % (outPath, seedNo, actName)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plot_Hot_solution2test(solu2test,
size_vec2mat=20,
actName=None,
seedNo=1000,
outPath=None):
solu2color = np.reshape(solu2test, (size_vec2mat, size_vec2mat))
plt.figure()
ax = plt.gca()
plt.imshow(solu2color,
interpolation='nearest',
cmap=cm.coolwarm,
origin='lower')
plt.colorbar(shrink=0.9)
plt.xticks(())
plt.yticks(())
# # plt.title('exact solution', fontsize=14)
# if actName == 'Utrue':
# fntmp = '%s/%s%s' % (outPath, seedNo, 'Utrue2test')
# elif actName == 'sReLU':
# fntmp = '%s/%s_%s' % (outPath, seedNo, 'UsReLU2test')
# elif actName == 's2ReLU':
# fntmp = '%s/%s_%s' % (outPath, seedNo, 'Us2ReLU2test')
# elif actName == 'ReLU':
# fntmp = '%s/%s_%s' % (outPath, seedNo, 'UReLU2test')
if str.lower(actName) == 'utrue':
fntmp = '%s/%s%s' % (outPath, seedNo, 'Utrue2test')
else:
fntmp = '%s/%s_%s' % (outPath, seedNo, actName)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plot_Hot_solution2test(solu2test, size_vec2mat=20, actName=None, seedNo=1000, outPath=None):
solu2color = np.reshape(solu2test, (size_vec2mat, size_vec2mat))
plt.figure()
ax = plt.gca()
plt.imshow(solu2color, interpolation='nearest', cmap=cm.coolwarm, origin='lower')
plt.colorbar(shrink=0.9)
plt.xticks(())
plt.yticks(())
# plt.title('exact solution', fontsize=14)
if str.lower(actName) == 'utrue':
fntmp = '%s/%s%s' % (outPath, seedNo, 'Utrue2test')
elif str.lower(actName) == 'srelu':
fntmp = '%s/%s_%s' % (outPath, seedNo, 'UsReLU2test')
elif str.lower(actName) == 's2relu':
fntmp = '%s/%s_%s' % (outPath, seedNo, 'Us2ReLU2test')
elif str.lower(actName) == 'relu':
fntmp = '%s/%s_%s' % (outPath, seedNo, 'UReLU2test')
elif str.lower(actName) == 'tanh':
fntmp = '%s/%s_%s' % (outPath, seedNo, 'Utanh2test')
elif str.lower(actName) == 'sintanh':
fntmp = '%s/%s_%s' % (outPath, seedNo, 'Ustanh2test')
elif str.lower(actName) == 'singauss':
fntmp = '%s/%s_%s' % (outPath, seedNo, 'Usgauss2test')
elif str.lower(actName) == 'gauss':
fntmp = '%s/%s_%s' % (outPath, seedNo, 'Ugauss2test')
elif str.lower(actName) == 'mexican':
fntmp = '%s/%s_%s' % (outPath, seedNo, 'Umexican2test')
elif str.lower(actName) == 'modify_mexican':
fntmp = '%s/%s_%s' % (outPath, seedNo, 'Ummexican2test')
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plot_scatter_solutions2test(solu1_test, solu2_test, test_batch, actName1=None, actName2=None, seedNo=1000,
outPath=None):
dim2test_batch = 2
if 2 == dim2test_batch:
test_x_bach = np.reshape(test_batch[:, 0], newshape=[-1, 1])
test_y_bach = np.reshape(test_batch[:, 1], newshape=[-1, 1])
# 绘制解的3D散点图(真解和预测解)
fig = plt.figure(figsize=(10, 10))
ax = Axes3D(fig)
ax.scatter(test_x_bach, test_y_bach, solu1_test, c='b', label=actName1)
ax.scatter(test_x_bach, test_y_bach, solu2_test, c='b', label=actName2)
# 绘制图例
ax.legend(loc='best')
# 添加坐标轴(顺序是X，Y, Z)
ax.set_xlabel('X', fontdict={'size': 15, 'color': 'red'})
ax.set_ylabel('Y', fontdict={'size': 15, 'color': 'red'})
ax.set_zlabel('u', fontdict={'size': 15, 'color': 'red'})
# plt.title('solution', fontsize=15)
fntmp = '%s/%ssolus_%s' % (outPath, seedNo, actName2)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
else:
return
def plot_Hot_point_wise_err(point_wise_err, size_vec2mat=20, actName=None, seedNo=1000, outPath=None):
# 逐点误差分布热力图
square_err_color2sin = np.reshape(point_wise_err, (size_vec2mat, size_vec2mat))
plt.figure(figsize=(10, 8))
ax = plt.gca()
plt.imshow(square_err_color2sin, interpolation='nearest', cmap=cm.coolwarm, origin='lower')
plt.colorbar(shrink=0.85)
plt.xticks(())
plt.yticks(())
# plt.title('point-wise error', fontsize=14)
if str.lower(actName) == 'srelu':
fntmp = '%s/%spErr_%s' % (outPath, seedNo, 'sReLU')
elif str.lower(actName) == 's2relu':
fntmp = '%s/%spErr_%s' % (outPath, seedNo, 's2ReLU')
elif str.lower(actName) == 'relu':
fntmp = '%s/%spErr_%s' % (outPath, seedNo, 'ReLU')
elif str.lower(actName) == 'tanh':
fntmp = '%s/%spErr_%s' % (outPath, seedNo, 'tanh')
elif str.lower(actName) == 'sin':
fntmp = '%s/%spErr_%s' % (outPath, seedNo, 'sin')
elif str.lower(actName) == 'sintanh':
fntmp = '%s/%spErr_%s' % (outPath, seedNo, 'stanh')
elif str.lower(actName) == 'singauss':
fntmp = '%s/%spErr_%s' % (outPath, seedNo, 'sgauss')
elif str.lower(actName) == 'gauss':
fntmp = '%s/%spErr_%s' % (outPath, seedNo, 'gauss')
elif str.lower(actName) == 'mexican':
fntmp = '%s/%spErr_%s' % (outPath, seedNo, 'mexican')
elif str.lower(actName) == 'modify_mexican':
fntmp = '%s/%spErr_%s' % (outPath, seedNo, 'mmexican')
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plot_3Trainlosses_1act_func(data2loss_1,
data2loss_2,
data2loss_3,
lossName1=None,
lossName2=None,
lossName3=None,
lossType=None,
seedNo=1000,
outPath=None,
xaxis_scale=False,
yaxis_scale=False):
plt.figure()
ax = plt.gca()
plt.plot(data2loss_1, 'b-.', label=lossName1)
plt.plot(data2loss_2, 'r:', label=lossName2)
plt.plot(data2loss_3, 'c*', label=lossName3)
plt.xlabel('epoch', fontsize=14)
plt.ylabel('loss', fontsize=14)
plt.legend(fontsize=13)
if xaxis_scale:
ax.set_yscale('log')
if yaxis_scale:
ax.set_yscale('log')
# plt.title('loss_it', fontsize=15)
fntmp = '%s/%s%s' % (outPath, seedNo, lossType)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plotTrain_MSE_REL_1act_func(data2mse,
data2rel,
actName=None,
seedNo=1000,
outPath=None,
xaxis_scale=False,
yaxis_scale=False):
plt.figure()
ax = plt.gca()
plt.plot(data2mse, 'r-.', label='MSE')
plt.plot(data2rel, 'b:', label='REL')
if xaxis_scale:
ax.set_yscale('log')
if yaxis_scale:
ax.set_yscale('log')
plt.xlabel('epoch', fontsize=18)
plt.ylabel('error', fontsize=18)
plt.legend(fontsize=18)
# # plt.title('training error', fontsize=15)
# if actName == 'sReLU':
# fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 'sReLU')
# elif actName == 's2ReLU':
# fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 's2ReLU')
# elif actName == 'ReLU':
# fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 'ReLU')
fntmp = '%s/%strainErr_%s' % (outPath, seedNo, actName)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plotTest_MSEs_RELs_3act_funcs(mse2data1,
mse2data2,
mse2data3,
rel2data1,
rel2data2,
rel2data3,
actName1=None,
actName2=None,
actName3=None,
seedNo=1000,
outPath=None,
xaxis_scale=False,
yaxis_scale=False):
plt.figure(figsize=(10, 8))
ax = plt.gca()
plt.plot(mse2data1, 'g-.', label=str('MSE-' + actName1))
plt.plot(rel2data1, 'b:', label=str('REL-' + actName1))
plt.plot(mse2data2, 'm--.', label=str('MSE-' + actName2))
plt.plot(rel2data2, 'c-*', label=str('REL-' + actName2))
plt.plot(mse2data3, color='k', marker='v', label=str('MSE-' + actName3))
plt.plot(rel2data3, color='gold', marker='x', label=str('REL-' + actName3))
if xaxis_scale:
ax.set_yscale('log')
if yaxis_scale:
ax.set_yscale('log')
plt.xlabel('epoch', fontsize=14)
plt.ylabel('error', fontsize=14)
ax.legend(loc='center', bbox_to_anchor=(0.485, 1.055), ncol=3, fontsize=15)
# plt.legend(fontsize=11)
# plt.title(' train error', fontsize=15)
fntmp = '%s/%stest_Errs' % (outPath, seedNo)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plot_Test_MSE_REL_3Types(mse2s2ReLU,
mse2sReLU,
mse2ReLU,
rel2s2ReLU,
rel2sReLU,
rel2ReLU,
epoches=100,
seedNo=1000,
outPath=None):
# fig2mse_test = plt.figure(figsize=(10, 8), dpi=98)
fig2mse_test = plt.figure(figsize=(9, 6.5), dpi=98)
ax = plt.gca()
ax.plot(epoches, mse2s2ReLU, 'g-.', label='MSE-s2ReLU')
ax.plot(epoches, rel2s2ReLU, 'b:', label='REL-s2ReLU')
ax.plot(epoches, mse2sReLU, 'm--', label='MSE-sReLU')
ax.plot(epoches, rel2sReLU, 'c-*', label='REL-sReLU')
ax.plot(epoches, mse2ReLU, color='k', marker='v', label='MSE-ReLU')
ax.plot(epoches, rel2ReLU, color='gold', marker='x', label='REL-ReLU')
plt.xlabel('epoch/1000', fontsize=14)
plt.ylabel('error', fontsize=14)
ax.set_yscale('log')
ax.legend(loc='center', bbox_to_anchor=(0.49, 1.06), ncol=3, fontsize=12)
# plt.legend(fontsize=11)
# plt.title('testing error ', fontsize=15)
fntmp = '%s/%stest_error' % (outPath, seedNo)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plot_Test_MSE_REL_2ActFuncs(data_mse1,
data_rel1,
data_mse2,
data_rel2,
epoches,
actName1=None,
actName2=None,
seedNo=1000,
outPath=None,
xaxis_scale=False,
yaxis_scale=False):
# fig2mse_test = plt.figure(figsize=(10, 8), dpi=98)
fig2mse_test = plt.figure(figsize=(9, 6.5), dpi=98)
ax = plt.gca()
ax.plot(epoches, data_mse1, 'g-.', label=str('MSE-' + actName1))
ax.plot(epoches, data_rel1, 'b:', label=str('REL' + actName1))
ax.plot(epoches, data_mse2, 'm--', label=str('MSE' + actName2))
ax.plot(epoches, data_rel2, 'c-*', label=str('REL' + actName2))
plt.xlabel('epoch/1000', fontsize=14)
plt.ylabel('error', fontsize=14)
if xaxis_scale:
ax.set_yscale('log')
if yaxis_scale:
ax.set_yscale('log')
ax.legend(loc='center', bbox_to_anchor=(0.49, 1.06), ncol=3, fontsize=12)
# plt.legend(fontsize=11)
# plt.title('testing error ', fontsize=15)
fntmp = '%s/%stest_error' % (outPath, seedNo)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plotTrain_MSE_REL_1act_func(data2mse, data2rel, actName=None, seedNo=1000, outPath=None, xaxis_scale=False,
yaxis_scale=False):
plt.figure()
ax = plt.gca()
plt.plot(data2mse, 'r-.', label='MSE')
plt.plot(data2rel, 'b:', label='REL')
if xaxis_scale:
ax.set_yscale('log')
if yaxis_scale:
ax.set_yscale('log')
plt.xlabel('epoch', fontsize=18)
plt.ylabel('error', fontsize=18)
plt.legend(fontsize=18)
# plt.title('training error', fontsize=15)
if str.lower(actName) == 'srelu':
fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 'sReLU')
elif str.lower(actName) == 'sin':
fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 'sin')
elif str.lower(actName) == 'sinaddcos':
fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 'sincos')
elif str.lower(actName) == 's2relu':
fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 's2ReLU')
elif str.lower(actName) == 's3relu':
fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 's3ReLU')
elif str.lower(actName) == 'csrelu':
fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 'CsReLU')
elif str.lower(actName) == 'relu':
fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 'ReLU')
elif str.lower(actName) == 'elu':
fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 'elu')
elif str.lower(actName) == 'gelu':
fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 'gelu')
elif str.lower(actName) == 'mgelu':
fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 'mgelu')
elif str.lower(actName) == 'tanh':
fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 'tanh')
elif str.lower(actName) == 'sintanh':
fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 'sintanh')
elif str.lower(actName) == 'singauss':
fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 'singauss')
elif str.lower(actName) == 'gauss':
fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 'gauss')
elif str.lower(actName) == 'mish':
fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 'mish')
elif str.lower(actName) == 'gcu':
fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 'gcu')
elif str.lower(actName) == 'mexican':
fntmp = '%s/%strainErr_%s' % (outPath, seedNo, 'mexican')
elif str.lower(actName) == 'modify_mexican':
fntmp = '%s/%s_%s' % (outPath, seedNo, 'Ummexican2test')
elif str.lower(actName) == 'sin_modify_mexican':
fntmp = '%s/%s_%s' % (outPath, seedNo, 'Usm_mexican2test')
else:
fntmp = '%s/%s_%s' % (outPath, seedNo, actName)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plotTrain_REL_1act_func(data2rel, relType=None, seedNo=1000, outPath=None, xaxis_scale=False, yaxis_scale=False):
plt.figure()
ax = plt.gca()
plt.plot(data2rel, 'b-.', label=relType)
plt.xlabel('epoch', fontsize=14)
plt.ylabel('error', fontsize=14)
plt.legend(fontsize=13)
if xaxis_scale:
ax.set_yscale('log')
if yaxis_scale:
ax.set_yscale('log')
# plt.title('training mse', fontsize=13)
fntmp = '%s/%strain_mse' % (outPath, seedNo)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plotTrain_loss_1act_func(data2loss, lossType=None, seedNo=1000, outPath=None, xaxis_scale=False, yaxis_scale=False):
plt.figure()
ax = plt.gca()
plt.plot(data2loss, 'b-.', label=lossType)
plt.xlabel('epoch', fontsize=14)
plt.ylabel(lossType, fontsize=14)
plt.legend(fontsize=18)
if xaxis_scale:
ax.set_yscale('log')
if yaxis_scale:
ax.set_yscale('log')
# plt.title('loss_it', fontsize=15)
fntmp = '%s/%s%s' % (outPath, seedNo, lossType)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plotTest_MSE_REL(data2mse, data2rel, epoches, actName=None, seedNo=1000, outPath=None, xaxis_scale=False, yaxis_scale=False):
plt.figure()
ax = plt.gca()
plt.plot(epoches, data2mse, 'r-.', label='MSE')
plt.plot(epoches, data2rel, 'b:', label='REL')
plt.xlabel('epoch/1000', fontsize=18)
# plt.ylabel('L2error', fontsize=18)
if xaxis_scale:
ax.set_yscale('log')
if yaxis_scale:
ax.set_yscale('log')
plt.legend(fontsize=18)
plt.title('testing error ', fontsize=15)
fntmp = '%s/%stestERR_%s' % (outPath, seedNo, actName)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plot_2TestRELs(data2rel1, data2rel2, relType1=None, relType2=None, epoches=1000, seedNo=1000, outPath=None,
xaxis_scale=False, yaxis_scale=False):
plt.figure()
ax = plt.gca()
plt.plot(epoches, data2rel1, 'r-.', label=relType1)
plt.plot(epoches, data2rel2, 'b:', label=relType2)
plt.xlabel('epoch/1000', fontsize=18)
# plt.ylabel('L2error', fontsize=18)
if xaxis_scale:
ax.set_yscale('log')
if yaxis_scale:
ax.set_yscale('log')
plt.legend(fontsize=18)
plt.title('testing mse ', fontsize=15)
fntmp = '%s/%stest_rel' % (outPath, seedNo)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plotTrain_MSEs_2act_funcs(data2mse1, data2mse2, mseName1=None, mseName2=None, seedNo=1000, outPath=None,
xaxis_scale=False, yaxis_scale=False):
plt.figure()
ax = plt.gca()
plt.plot(data2mse1, 'b-.', label=mseName1)
plt.plot(data2mse2, 'r:', label=mseName2)
plt.xlabel('epoch', fontsize=14)
plt.ylabel('error', fontsize=14)
plt.legend(fontsize=13)
if xaxis_scale:
ax.set_yscale('log')
if yaxis_scale:
ax.set_yscale('log')
# plt.title('training mse', fontsize=13)
fntmp = '%s/%strain_mses' % (outPath, seedNo)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plot_2scatter_solus2test(exact_solu, predict_solu, coord_points=None, batch_size=100, seedNo=1000, outPath=None,
actName=None):
# fig11 = plt.figure(figsize=(9, 6.5))
# plt.scatter(coord_points, exact_solu, s=batch_size, c='b', marker='.', label='Exact')
# plt.scatter(coord_points, predict_solu, s=batch_size, c='r', marker='1', label='UNN')
# plt.legend(loc='right', bbox_to_anchor=(0.9, 1.05), ncol=4, fontsize=12)
# plt.xlabel('x', fontsize=14)
# plt.ylabel('y', fontsize=14)
fig11 = plt.figure(figsize=(9, 6.5))
ax = plt.gca()
ax.scatter(coord_points, exact_solu, s=batch_size, c='b', marker='.', label='Exact')
ax.scatter(coord_points, predict_solu, s=batch_size, c='r', marker='1', label='UNN2'+actName)
ax.legend(loc='right', bbox_to_anchor=(0.9, 1.05), ncol=4, fontsize=12)
ax.set_xlabel('x', fontsize=14)
ax.set_ylabel('solus', fontsize=14)
fntmp = '%s/%ssolu2test' % (outPath, seedNo)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plot_testSolu2convid(solu_arr,
name2solu=None,
coord_points2test=None,
seedNo=1000,
outPath=None,
xaxis_scale=False,
yaxis_scale=False):
plt.figure(figsize=(10, 8))
ax = plt.gca()
plt.plot(coord_points2test, solu_arr, 'r-.', label=str(name2solu))
plt.xlabel('day', fontsize=18)
plt.ylabel(str(name2solu), fontsize=18)
if xaxis_scale:
ax.set_yscale('log')
if yaxis_scale:
ax.set_yscale('log')
plt.legend(fontsize=18)
plt.title(str(name2solu), fontsize=15)
fntmp = '%s/%s' % (outPath, str(name2solu))
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plotTrain_losses(loss2act1,
loss2act2,
loss2act3,
lossType=None,
actName1=None,
actName2=None,
actName3=None,
seedNo=1000,
outPath=None):
if 'loss_it' == lossType:
plt.figure()
ax = plt.gca()
plt.plot(loss2act1, 'b-.', label='actName1')
plt.plot(loss2act2, 'r:', label='actName2')
plt.plot(loss2act3, 'c-*', label='actName3')
plt.xlabel('epoch', fontsize=14)
plt.ylabel('loss_it', fontsize=14)
plt.legend(fontsize=13)
# plt.title('loss_it', fontsize=15)
fntmp = '%s/%s%s' % (outPath, seedNo, lossType)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plotTrain_losses(loss2s2ReLU,
loss2sReLU,
loss2ReLU,
lossType=None,
seedNo=1000,
outPath=None):
if 'loss_it' == lossType:
plt.figure()
ax = plt.gca()
plt.plot(loss2s2ReLU, 'b-.', label='s2ReLU')
plt.plot(loss2sReLU, 'r:', label='sReLU')
plt.plot(loss2ReLU, 'c-*', label='ReLU')
plt.xlabel('epoch', fontsize=14)
plt.ylabel('loss_it', fontsize=14)
plt.legend(fontsize=13)
# plt.title('loss_it', fontsize=15)
fntmp = '%s/%sloss_it' % (outPath, seedNo)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
elif 'loss_bd' == lossType:
plt.figure()
ax = plt.gca()
plt.plot(loss2s2ReLU, 'b-.', label='s2ReLU')
plt.plot(loss2sReLU, 'r:', label='sReLU')
plt.plot(loss2ReLU, 'c-*', label='ReLU')
ax.set_yscale('log')
plt.xlabel('epoch', fontsize=14)
plt.ylabel('loss_bd', fontsize=14)
plt.legend(fontsize=13)
# plt.title('loss_bd', fontsize=15)
fntmp = '%s/%sloss_bd' % (outPath, seedNo)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
elif 'loss' == lossType:
plt.figure()
ax = plt.gca()
plt.plot(loss2s2ReLU, 'b-.', label='s2ReLU')
plt.plot(loss2sReLU, 'r:', label='sReLU')
plt.plot(loss2ReLU, 'c-*', label='ReLU')
plt.xlabel('epoch', fontsize=14)
plt.ylabel('loss', fontsize=14)
plt.legend(fontsize=13)
# plt.title('loss', fontsize=15)
fntmp = '%s/%sloss' % (outPath, seedNo)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
def plot_3solutions2test(exact_solu2test, s2ReLU_solu2test, sReLU_solu2test, ReLU_solu2test,
coord_points2test=None, batch_size2test=1000, seedNo=1000, outPath=None, subfig_type=1):
# 嵌入绘制局部放大图的坐标系
if subfig_type == 1:
subgfig = plt.figure(figsize=(10, 8), dpi=98)
ax = plt.gca() # fig, ax = plt.subplots(a,b)用来控制子图个数：a为行数，b为列数。
ax.plot(coord_points2test, exact_solu2test, 'b-.', label='exact')
ax.plot(coord_points2test, s2ReLU_solu2test, 'g:', label='s2ReLU')
ax.plot(coord_points2test, sReLU_solu2test, 'm--', label='sReLU')
ax.plot(coord_points2test, ReLU_solu2test, 'c-', label='ReLU')
ax.legend(loc='right', bbox_to_anchor=(0.85, 1.03), ncol=4, fontsize=12)
ax.set_xlabel('epoch', fontsize=14)
axins = inset_axes(ax, width="50%", height="40%", loc=8, bbox_to_anchor=(0.2, 0.2, 0.5, 0.5),
bbox_transform=ax.transAxes)
# 在子坐标系中绘制原始数据
axins.plot(coord_points2test, exact_solu2test, color='b', linestyle='-.')
axins.plot(coord_points2test, s2ReLU_solu2test, color='g', linestyle=':')
axins.plot(coord_points2test, sReLU_solu2test, color='m', linestyle='--')
axins.plot(coord_points2test, ReLU_solu2test, color='c', linestyle='-')
axins.set_xticks([])
axins.set_yticks([])
# 设置放大区间
zone_left = int(0.4 * batch_size2test)
zone_right = int(0.4 * batch_size2test) + 150
# 坐标轴的扩展比例（根据实际数据调整）
x_ratio = 0.075 # x轴显示范围的扩展比例
y_ratio = 0.04 # y轴显示范围的扩展比例
# X轴的显示范围
xlim0 = coord_points2test[zone_left] - (coord_points2test[zone_right] - coord_points2test[zone_left]) * x_ratio
xlim1 = coord_points2test[zone_right] + (coord_points2test[zone_right] - coord_points2test[zone_left]) * x_ratio
# Y轴的显示范围
y = np.hstack((exact_solu2test[zone_left:zone_right], s2ReLU_solu2test[zone_left:zone_right]))
ylim0 = np.min(y) - (np.max(y) - np.min(y)) * y_ratio
ylim1 = np.max(y) + (np.max(y) - np.min(y)) * y_ratio
# 调整子坐标系的显示范围
axins.set_xlim(xlim0, xlim1)
axins.set_ylim(ylim0, ylim1)
# 建立父坐标系与子坐标系的连接线
# loc1 loc2: 坐标系的四个角
# 1 (右上) 2 (左上) 3(左下) 4(右下)
mark_inset(ax, axins, loc1=3, loc2=1, fc="none", ec='k', lw=1)
fntmp = '%s/%ssolu2test' % (outPath, seedNo)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
elif subfig_type == 2:
plt.figure(figsize=(16, 10), dpi=98)
ax = plt.gca()
p1 = plt.subplot(121) # 1行2列，第一个图
p2 = plt.subplot(122) # 1行2列，第二个图
p1.plot(coord_points2test, exact_solu2test, color='b', linestyle='-.', label='true')
p1.plot(coord_points2test, s2ReLU_solu2test, color='g', linestyle=':', label='predict')
ax.legend(fontsize=10)
p2.plot(coord_points2test, exact_solu2test, color='b', linestyle='-.', label='true')
p2.plot(coord_points2test, s2ReLU_solu2test, color='g', linestyle=':', label='predict')
p2.axis([0.35, 0.65, 0.2, 0.27])
# plot the box of
tx0 = 0.35
tx1 = 0.65
ty0 = 0.2
ty1 = 0.27
sx = [tx0, tx1, tx1, tx0, tx0]
sy = [ty0, ty0, ty1, ty1, ty0]
p1.plot(sx, sy, "purple")
# plot patch lines
xy = (0.64, 0.265)
xy2 = (0.36, 0.265)
con = ConnectionPatch(xyA=xy2, xyB=xy, coordsA="data", coordsB="data", axesA=p2, axesB=p1)
p2.add_artist(con)
xy = (0.64, 0.21)
xy2 = (0.36, 0.205)
con = ConnectionPatch(xyA=xy2, xyB=xy, coordsA="data", coordsB="data",
axesA=p2, axesB=p1)
p2.add_artist(con)
fntmp = '%s/%ssolu2test' % (outPath, seedNo)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
else:
# fig11 = plt.figure(figsize=(10, 8))
fig11 = plt.figure(figsize=(9, 6.5))
ax = plt.gca()
ax.plot(coord_points2test, exact_solu2test, 'b-.', label='exact')
ax.plot(coord_points2test, s2ReLU_solu2test, 'g:', label='s2ReLU')
ax.plot(coord_points2test, sReLU_solu2test, 'm--', label='sReLU')
ax.plot(coord_points2test, ReLU_solu2test, 'c-', label='ReLU')
# box = ax.get_position()
# ax.set_position([box.x0, box.y0, box.width, box.height * 0.8])
ax.legend(loc='right', bbox_to_anchor=(0.9, 1.05), ncol=4, fontsize=12)
ax.set_xlabel('x', fontsize=14)
ax.set_ylabel('u', fontsize=14)
fntmp = '%s/%ssolu2test' % (outPath, seedNo)
DNN_tools.mySaveFig(plt, fntmp, ax=ax, isax=1, iseps=0)
