def train(totpath, fotpath, workPath, tSampleNamePart, tModelNamePart):
X, Y, s2n = getData2(totpath, fotpath, workPath, tSampleNamePart)
print(X.shape)
print(Y.shape)
print(s2n.shape)
N_data = X.shape[0]
N_train = int(N_data * 0.9)
print("train: %d, test: %d" % (N_train, N_data - N_train))
X_train, Y_train, s2n_train = X[:N_train], Y[:N_train], s2n[:N_train]
X_test, Y_test, s2n_test = X[N_train:], Y[N_train:], s2n[N_train:]
model = createModel()
#optimizer = keras.optimizers.SGD(lr=0.0001, decay=1e-6, momentum=0.9, nesterov=True)
optimizer = keras.optimizers.Adam(lr=0.00001,
beta_1=0.9,
beta_2=0.999,
decay=0.0)
model.compile(loss='mean_squared_error', optimizer=optimizer)
model.fit(X_train,
Y_train,
batch_size=128,
epochs=100,
validation_split=0.2)
modelName = "model_RealFOT_%s.h5" % (tModelNamePart)
model.save("%s/%s" % (workPath, modelName))
def test3(totpath, fotpath, workPath, tSampleNamePart, tModelNamePart1,
tModelNamePart2):
X, Y, s2n = getData2(totpath, fotpath, workPath, tSampleNamePart)
print(X.shape)
print(Y.shape)
print(s2n.shape)
N_data = X.shape[0]
N_train = int(N_data * 0.9)
print("train: %d, test: %d" % (N_train, N_data - N_train))
X_test, Y_test, s2n_test = X[N_train:], Y[N_train:], s2n[N_train:]
from keras.models import load_model
modelName1 = "model_RealFOT_%s.h5" % (tModelNamePart1)
model = load_model("%s/%s" % (workPath, modelName1))
Y_pred = model.predict(X_test)
pbb_threshold = 0.5
pred_labels = np.array((Y_pred[:, 1] > pbb_threshold), dtype="int")
print("Correctly classified %d out of %d" %
((pred_labels == Y_test[:, 1].astype(int)).sum(), Y_test.shape[0]))
print("accuracy = %f" %
(1. *
(pred_labels == Y_test[:, 1].astype(int)).sum() / Y_test.shape[0]))
TIdx = Y_test[:, 1] == 1
FIdx = Y_test[:, 1] == 0
T_pred_rst = pred_labels[TIdx]
F_pred_rst = pred_labels[FIdx]
print(T_pred_rst.shape)
print(F_pred_rst.shape)
TP = ((T_pred_rst == 1).astype(int)).sum()
TN = ((F_pred_rst == 0).astype(int)).sum()
FP = ((F_pred_rst == 1).astype(int)).sum()
FN = ((T_pred_rst == 0).astype(int)).sum()
print("total=%d,TP=%d,TN=%d,FP=%d,FN=%d" %
(Y_test.shape[0], TP, TN, FP, FN))
accuracy = (TP + TN) * 1.0 / (TP + FN + TN + FP)
precision = (TP) * 1.0 / (TP + FP)
recall = (TP) * 1.0 / (TP + FN)
f1_score = 2.0 * (precision * recall) / (precision + recall)
print("accuracy=%f%%" % (accuracy * 100))
print("precision=%f%%" % (precision * 100))
print("recall=%f%%" % (recall * 100))
print("f1_score=%f%%" % (f1_score * 100))
modelName1 = "model_RealFOT_%s.h5" % (tModelNamePart2)
model = load_model("%s/%s" % (workPath, modelName1))
Y_pred = model.predict(X_test)
pbb_threshold = 0.5
pred_labels = np.array((Y_pred[:, 1] > pbb_threshold), dtype="int")
print("Correctly classified %d out of %d" %
((pred_labels == Y_test[:, 1].astype(int)).sum(), Y_test.shape[0]))
print("accuracy = %f" %
(1. *
(pred_labels == Y_test[:, 1].astype(int)).sum() / Y_test.shape[0]))
TIdx = Y_test[:, 1] == 1
FIdx = Y_test[:, 1] == 0
T_pred_rst = pred_labels[TIdx]
F_pred_rst = pred_labels[FIdx]
print(T_pred_rst.shape)
print(F_pred_rst.shape)
TP = ((T_pred_rst == 1).astype(int)).sum()
TN = ((F_pred_rst == 0).astype(int)).sum()
FP = ((F_pred_rst == 1).astype(int)).sum()
FN = ((T_pred_rst == 0).astype(int)).sum()
print("total=%d,TP=%d,TN=%d,FP=%d,FN=%d" %
(Y_test.shape[0], TP, TN, FP, FN))
accuracy = (TP + TN) * 1.0 / (TP + FN + TN + FP)
precision = (TP) * 1.0 / (TP + FP)
recall = (TP) * 1.0 / (TP + FN)
f1_score = 2.0 * (precision * recall) / (precision + recall)
print("accuracy=%f%%" % (accuracy * 100))
print("precision=%f%%" % (precision * 100))
print("recall=%f%%" % (recall * 100))
print("f1_score=%f%%" % (f1_score * 100))
def test(totpath, fotpath, workPath, tSampleNamePart, tModelNamePart):
X, Y, s2n = getData2(totpath, fotpath, workPath, tSampleNamePart)
print(X.shape)
print(Y.shape)
print(s2n.shape)
N_data = X.shape[0]
N_train = int(N_data * 0.9)
print("train: %d, test: %d" % (N_train, N_data - N_train))
X_train, Y_train, s2n_train = X[:N_train], Y[:N_train], s2n[:N_train]
X_test, Y_test, s2n_test = X[N_train:], Y[N_train:], s2n[N_train:]
from keras.models import load_model
modelName = "model_RealFOT_%s.h5" % (tModelNamePart)
model = load_model("%s/%s" % (workPath, modelName))
Y_pred = model.predict(X_test)
pbb_threshold = 0.5
pred_labels = np.array((Y_pred[:, 1] > pbb_threshold), dtype="int")
print("Correctly classified %d out of %d" %
((pred_labels == Y_test[:, 1].astype(int)).sum(), Y_test.shape[0]))
print("accuracy = %f" %
(1. *
(pred_labels == Y_test[:, 1].astype(int)).sum() / Y_test.shape[0]))
TIdx = Y_test[:, 1] == 1
FIdx = Y_test[:, 1] == 0
T_pred_rst = pred_labels[TIdx]
F_pred_rst = pred_labels[FIdx]
print(T_pred_rst.shape)
print(F_pred_rst.shape)
TP = ((T_pred_rst == 1).astype(int)).sum()
TN = ((F_pred_rst == 0).astype(int)).sum()
FP = ((F_pred_rst == 1).astype(int)).sum()
FN = ((T_pred_rst == 0).astype(int)).sum()
print("total=%d,TP=%d,TN=%d,FP=%d,FN=%d" %
(Y_test.shape[0], TP, TN, FP, FN))
accuracy = (TP + TN) * 1.0 / (TP + FN + TN + FP)
precision = (TP) * 1.0 / (TP + FP)
recall = (TP) * 1.0 / (TP + FN)
f1_score = 2.0 * (precision * recall) / (precision + recall)
print("accuracy=%f%%" % (accuracy * 100))
print("precision=%f%%" % (precision * 100))
print("recall=%f%%" % (recall * 100))
print("f1_score=%f%%" % (f1_score * 100))
falseImg = X_test[pred_labels != Y_test[:, 1]]
falseLabel = Y_test[pred_labels != Y_test[:, 1]]
falsePred = Y_pred[pred_labels != Y_test[:, 1]]
falseS2n = s2n_test[pred_labels != Y_test[:, 1]]
tnum1 = 0
tnum2 = 0
for i in range(falseImg.shape[0]):
if falseLabel[i, 1] == 1:
tnum1 = tnum1 + 1
else:
tnum2 = tnum2 + 1
print("total %d, miss classified %d" %
(Y_test.shape[0], falseImg.shape[0]))
print("True classified as False %d" % (tnum1))
print("False classified as True %d" % (tnum2))
print("\n\n***********************")
print("image of True classified as False")
for i in range(falseImg.shape[0]):
if falseLabel[i, 1] == 1:
objWidz = zscale_image(falseImg[i][0])
tmpWidz = zscale_image(falseImg[i][1])
resiWidz = zscale_image(falseImg[i][2])
if objWidz.shape[0] == 0:
objWidz = falseImg[i][0]
if tmpWidz.shape[0] == 0:
tmpWidz = falseImg[i][1]
if resiWidz.shape[0] == 0:
resiWidz = falseImg[i][2]
plt.clf()
fig, axes = plt.subplots(1, 3, figsize=(6, 2))
axes.flat[0].imshow(objWidz, interpolation="nearest", cmap='gray')
axes.flat[1].imshow(tmpWidz, interpolation="nearest", cmap='gray')
axes.flat[2].imshow(resiWidz, interpolation="nearest", cmap='gray')
axes.flat[2].set_title("predicted pbb = " +
str(np.round(falsePred[i][1], 2)) +
", label = " + str(falseLabel[i, 1]) +
", s2n = " + str(falseS2n[i]))
plt.show()
print("\n\n***********************")
print("image of False classified as True")
for i in range(falseImg.shape[0]):
if falseLabel[i, 1] == 0:
objWidz = zscale_image(falseImg[i][0])
tmpWidz = zscale_image(falseImg[i][1])
resiWidz = zscale_image(falseImg[i][2])
if objWidz.shape[0] == 0:
objWidz = falseImg[i][0]
if tmpWidz.shape[0] == 0:
tmpWidz = falseImg[i][1]
if resiWidz.shape[0] == 0:
resiWidz = falseImg[i][2]
plt.clf()
fig, axes = plt.subplots(1, 3, figsize=(6, 2))
axes.flat[0].imshow(objWidz, interpolation="nearest", cmap='gray')
axes.flat[1].imshow(tmpWidz, interpolation="nearest", cmap='gray')
axes.flat[2].imshow(resiWidz, interpolation="nearest", cmap='gray')
axes.flat[2].set_title("predicted pbb = " +
str(np.round(falsePred[i][1], 2)) +
", label = " + str(falseLabel[i, 1]) +
", s2n = " + str(falseS2n[i]))
plt.show()
trueImg = X_test[pred_labels == Y_test[:, 1]]
trueLabel = Y_test[pred_labels == Y_test[:, 1]]
truePred = Y_pred[pred_labels == Y_test[:, 1]]
trueS2n = s2n_test[pred_labels == Y_test[:, 1]]
showNum = 50
tnum = 0
print("\n\n***********************")
print("image of True classified as True")
for i in range(trueImg.shape[0]):
if trueLabel[i, 1] == 1:
objWidz = zscale_image(trueImg[i][0])
tmpWidz = zscale_image(trueImg[i][1])
resiWidz = zscale_image(trueImg[i][2])
if objWidz.shape[0] == 0:
objWidz = falseImg[i][0]
if tmpWidz.shape[0] == 0:
tmpWidz = falseImg[i][1]
if resiWidz.shape[0] == 0:
resiWidz = falseImg[i][2]
plt.clf()
fig, axes = plt.subplots(1, 3, figsize=(6, 2))
axes.flat[0].imshow(objWidz, interpolation="nearest", cmap='gray')
axes.flat[1].imshow(tmpWidz, interpolation="nearest", cmap='gray')
axes.flat[2].imshow(resiWidz, interpolation="nearest", cmap='gray')
axes.flat[2].set_title("predicted pbb = " +
str(np.round(truePred[i][1], 2)) +
", label = " + str(trueLabel[i, 1]) +
", s2n = " + str(trueS2n[i]))
plt.show()
tnum = tnum + 1
if tnum > showNum:
break
tnum = 0
print("\n\n***********************")
print("image of False classified as False")
for i in range(trueImg.shape[0]):
if trueLabel[i, 1] == 0:
objWidz = zscale_image(trueImg[i][0])
tmpWidz = zscale_image(trueImg[i][1])
resiWidz = zscale_image(trueImg[i][2])
if objWidz.shape[0] == 0:
objWidz = falseImg[i][0]
if tmpWidz.shape[0] == 0:
tmpWidz = falseImg[i][1]
if resiWidz.shape[0] == 0:
resiWidz = falseImg[i][2]
plt.clf()
fig, axes = plt.subplots(1, 3, figsize=(6, 2))
axes.flat[0].imshow(objWidz, interpolation="nearest", cmap='gray')
axes.flat[1].imshow(tmpWidz, interpolation="nearest", cmap='gray')
axes.flat[2].imshow(resiWidz, interpolation="nearest", cmap='gray')
axes.flat[2].set_title("predicted pbb = " +
str(np.round(truePred[i][1], 2)) +
", label = " + str(trueLabel[i, 1]) +
", s2n = " + str(trueS2n[i]))
plt.show()
tnum = tnum + 1
if tnum > showNum:
break
def doAll():
fotpath = "/home/xy/Downloads/myresource/deep_data2/simot/multi_scale_20190120/all"
totpath = "/home/xy/Downloads/myresource/deep_data2/simot/rest_data_20190120"
realDataPath = "/home/xy/Downloads/myresource/deep_data2/simot/multi_scale_20190120/20190116tot"
#dateStr = datetime.strftime(datetime.now(), "%Y%m%d")
dateStr = '20190122'
workPath = "/home/xy/Downloads/myresource/deep_data2/simot/train_%s" % (
dateStr)
if not os.path.exists(workPath):
os.system("mkdir %s" % (workPath))
print("work path is %s" % (workPath))
tSampleNamePart = "64_fot10w_%s" % (dateStr)
X, Y, s2n = getData2(totpath, fotpath, workPath, tSampleNamePart)
#X,Y,s2n = getRealData(realDataPath, workPath, tSampleNamePart)
magerr = s2n[Y[:, 1] == 1]
magerr2 = s2n[Y[:, 1] == 0]
print(s2n.shape[0])
print(magerr.shape[0])
print(magerr2.shape[0])
from matplotlib.ticker import MultipleLocator, FormatStrFormatter
xmajorLocator = MultipleLocator(10)
xminorLocator = MultipleLocator(5)
plt.figure(figsize=(6, 3))
ax = plt.subplot(111)
plt.hist(magerr,
bins=20,
range=(0, 100),
normed=False,
weights=None,
cumulative=False,
bottom=None,
histtype=u'bar',
align=u'left',
orientation=u'vertical',
rwidth=0.6,
log=False,
color='lightgreen',
label=None,
stacked=False,
hold=None)
ax.xaxis.set_major_locator(xmajorLocator)
ax.xaxis.set_minor_locator(xminorLocator)
ax.set_xlim(0, 100)
ax.set_ylabel("number of star")
ax.set_xlabel('s2n of star')
plt.subplots_adjust(left=0.15, right=0.95, top=0.95, bottom=0.15)
#plt.grid()
#plt.show()
plt.savefig('s2nAll.png')
xmajorLocator = MultipleLocator(1)
plt.figure(figsize=(5, 3))
ax = plt.subplot(111)
plt.hist(magerr,
bins=20,
range=(0, 20),
normed=False,
weights=None,
cumulative=False,
bottom=None,
histtype=u'bar',
align=u'left',
orientation=u'vertical',
rwidth=0.6,
log=False,
color='darkorange',
label=None,
stacked=False,
hold=None)
ax.xaxis.set_major_locator(xmajorLocator)
ax.set_xlim(4, 20)
#plt.grid()
#plt.show()
plt.savefig('s2nPart.png')
