def test_all_detections_class(fn):
folder_name = 'test'
test_detections = preprocessing.all_detections(folder_name)
view_name = fn.split('/')[-1].split('.feat')[0]
with open(fn) as f:
lines = f.readlines()
test_detections.update(lines, view_name)
labels = test_detections.get_all_tp_fp()
print(' ')
def test_save_to_txt_class(fn):
out_dir = '//lyta/tomodev/DeepLearning/Playground_stage3/output_pick1mark/score_txt/'
folder_name = 'test'
test_detections = preprocessing.all_detections(folder_name)
view_name = fn.split('/')[-1].split('_feat.txt')[0]
with open(fn) as f:
lines = f.readlines()
test_detections.update(lines, view_name)
labels = test_detections.get_all_tp_fp()
postprocessing.pick_one_mark_one_group(test_detections)
postprocessing.save_to_txt(test_detections, out_dir)
def run_all_test_weighted_logit_save_to_txt():
out_dir = '//lyta/tomodev/DeepLearning/Playground_stage3/output_pick1mark/score_txt/weighted_logit/'
for fn in glob.glob('C:/experiments/temp_feature_vector/test/*.txt'):
# hard code to make the program run through
folder_name = 'test'
test_detections = preprocessing.all_detections(folder_name)
view_name = fn.split('\\')[-1].split('_feat.txt')[0]
with open(fn) as f:
lines = f.readlines()
test_detections.update(lines, view_name)
postprocessing.pick_one_mark_one_group_use_logit(test_detections)
postprocessing.save_to_txt(test_detections, out_dir)
def run():
# get training set
os.chdir(test_dir)
folder_name = train_dir.split('/')[-2]
# initiate train detections
train_detections = preprocessing.all_detections(folder_name)
for file in glob.glob('*.txt'):
with open(file) as f:
lines = f.readlines()
view_name = file.split('_feat.txt')[0]
# adding detections from one view to this instance
train_detections.update(lines, view_name)
# get testing set
os.chdir(test_dir)
folder_name = test_dir.split('/')[-2]
# initiate train detections
test_detections = preprocessing.all_detections(folder_name)
for file in glob.glob('*.txt'):
with open(file) as f:
lines = f.readlines()
view_name = file.split('_feat.txt')[0]
# adding detections from one view to this instance
test_detections.update(lines, view_name)
# get train_x and train_y
train_y = np.array([
train_detections.groups[i].group_tp_fp
for i in range(train_detections.get_total_group_count())
])
train_x = lstm_processing.form_lstm_features_v3(train_detections)
train_x = lstm_processing.repeat_and_pad(train_x, maxlen)
train_x = sequence.pad_sequences(train_x, maxlen=maxlen, dtype='float')
train_x = np.array(train_x)
del train_detections
# get test_x and test_y
test_y = np.array([
test_detections.groups[i].group_tp_fp
for i in range(test_detections.get_total_group_count())
])
test_x = lstm_processing.form_lstm_features_v3(test_detections)
test_x = lstm_processing.repeat_and_pad(test_x, maxlen)
test_x = sequence.pad_sequences(test_x, maxlen=maxlen, dtype='float')
test_x = np.array(test_x)
del test_detections
model = Sequential()
model.add(Bidirectional(LSTM(128), input_shape=(maxlen, 4)))
model.add(Dropout(0.5))
# model.add(Bidirectional(LSTM(512, return_sequences=True)))
# model.add(Dropout(0.5))
# model.add(Bidirectional(LSTM(128)))
# model.add(Dropout(0.5))
model.add(Dense(1, activation='sigmoid'))
# try using different optimizers and different optimizer configs
model.compile('adam', 'binary_crossentropy', metrics=['accuracy'])
print('Train...')
model.fit(train_x,
train_y,
batch_size=batch_size,
epochs=20,
validation_data=[test_x, test_y])
# predictions on train set
predictions_train = model.predict(train_x)
loss_train = model.evaluate(train_x, train_y, verbose=1)
print('This is the training set result: \n')
# Compute AUC
fpr, tpr, thresholds = roc_curve(train_y, predictions_train, pos_label=1)
az_train = auc(fpr, tpr)
print('AUC for train set is ' + str(az_train))
# predictions on test set
predictions_test = model.predict(test_x)
loss_test = model.evaluate(test_x, test_y, verbose=1)
print('This is the training set result: \n')
# Compute AUC
fpr, tpr, thresholds = roc_curve(test_y, predictions_test, pos_label=1)
az_test = auc(fpr, tpr)
print('AUC for testing set is ' + str(az_test))
# # record the predictions into the class
# train_detections.record_group_lstm_scores(predictions_train[:,0])
# test_detections.record_group_lstm_scores(predictions_test[:,0])
# save model
model.save_weights(out_dir + 'model.h5')
# Now test on each view and dump txt for scoring
for fn in glob.glob('C:/experiments/temp_feature_vector/test/*.txt'):
folder_name = 'test'
test_detections_one_view = preprocessing.all_detections(folder_name)
view_name = fn.split('\\')[-1].split('_feat.txt')[0]
with open(fn) as f:
lines = f.readlines()
test_detections_one_view.update(lines, view_name)
# Get the feature values
test_x = lstm_processing.form_lstm_features_v3(
test_detections_one_view)
test_x = lstm_processing.repeat_and_pad(test_x, maxlen)
test_x = sequence.pad_sequences(test_x, maxlen=maxlen, dtype='float')
test_x = np.array(test_x)
# predict
predictions_test_one_view = model.predict(test_x)
# record the predictions
test_detections_one_view.record_group_lstm_scores(
predictions_test_one_view[:, 0])
lstm_processing.save_to_txt_no_pick_one_mark(test_detections_one_view,
out_dir)
def run():
# get training set
os.chdir(train_dir)
folder_name = train_dir.split('/')[-2]
# initiate train detections
train_detections = preprocessing.all_detections(folder_name)
for file in glob.glob('*.txt')[0:300]:
with open(file) as f:
lines = f.readlines()
view_name = file.split('_feat.txt')[0]
# adding detections from one view to this instance
train_detections.update(lines, view_name)
# get testing set
os.chdir(test_dir)
folder_name = test_dir.split('/')[-2]
# initiate train detections
test_detections = preprocessing.all_detections(folder_name)
for file in glob.glob('*.txt')[0:1]:
with open(file) as f:
lines = f.readlines()
view_name = file.split('_feat.txt')[0]
# adding detections from one view to this instance
test_detections.update(lines, view_name)
# get train_x and train_y
train_y = np.array(train_detections.get_all_tp_fp())
train_x = np.array(train_detections.get_all_dl_feat_1536())
del train_detections
tsne = TSNE(n_components=n_components,
init='random',
random_state=0,
perplexity=perplexity)
train_x = tsne.fit_transform(train_x)
# get test_x and test_y
test_y = np.array(test_detections.get_all_tp_fp())
test_x = np.array(test_detections.get_all_dl_feat_1536())
del test_detections
test_x = tsne.fit_transform(test_x)
# scale the features
scaler = StandardScaler()
scaler.fit(train_x)
train_x = scaler.transform(train_x)
test_x = scaler.transform(test_x)
# start the training
mlp = MLPClassifier(hidden_layer_sizes=(30, 30, 30))
mlp.fit(train_x, train_y)
# predictions on train set
predictions_train = mlp.predict(train_x)
prob_train = mlp.predict_proba(train_x)[:, 1]
print('This is the training set result: \n')
print(confusion_matrix(train_y, predictions_train))
print(classification_report(train_y, predictions_train))
# Compute AUC
fpr, tpr, thresholds = roc_curve(train_y, prob_train, pos_label=1)
az_train = auc(fpr, tpr)
print('AUC for train set is ' + str(az_train))
# predictions on test set
predictions_test = mlp.predict(test_x)
prob_test = mlp.predict_proba(test_x)[:, 1]
print('This is the testing set result: \n')
print(confusion_matrix(test_y, predictions_test))
print(classification_report(test_y, predictions_test))
# Compute AUC
fpr, tpr, thresholds = roc_curve(test_y, prob_test, pos_label=1)
az_test = auc(fpr, tpr)
print('AUC for testing set is ' + str(az_test))
# record the predictions into the class
# train_detections.record_scores(prob_train)
# test_detections.record_scores(prob_test)
# Now test on each view and dump txt for scoring
for fn in glob.glob('C:/experiments/temp_feature_vector/test/*.txt'):
folder_name = 'test'
test_detections_one_view = preprocessing.all_detections(folder_name)
view_name = fn.split('\\')[-1].split('_feat.txt')[0]
with open(fn) as f:
lines = f.readlines()
test_detections_one_view.update(lines, view_name)
# Get the feature values
test_x = np.array(test_detections_one_view.get_all_dl_feat_1536())
test_x = tsne.fit_transform(test_x)
test_x = scaler.transform(test_x)
predictions_test_one_view = mlp.predict_proba(test_x)[:, 1]
# record the predictions
test_detections_one_view.record_scores(predictions_test_one_view)
postprocessing.pick_one_mark_one_group_use_prediction(
test_detections_one_view)
postprocessing.save_to_txt(test_detections_one_view, out_dir)
def run():
# get training set
os.chdir(train_dir)
folder_name = train_dir.split('/')[-2]
# initiate train detections
train_detections = preprocessing.all_detections(folder_name)
for file in glob.glob('*.txt')[0:10]:
with open(file) as f:
lines = f.readlines()
view_name = file.split('_feat.txt')[0]
# adding detections from one view to this instance
train_detections.update(lines, view_name)
# get testing set
os.chdir(test_dir)
folder_name = test_dir.split('/')[-2]
# initiate train detections
test_detections = preprocessing.all_detections(folder_name)
for file in glob.glob('*.txt')[0:1]:
with open(file) as f:
lines = f.readlines()
view_name = file.split('_feat.txt')[0]
# adding detections from one view to this instance
test_detections.update(lines, view_name)
# get train_x and train_y
y = np.array(train_detections.get_all_tp_fp())
X = np.array(train_detections.get_all_dl_feat_1536())
del train_detections
n_components = 2
(fig, subplots) = plt.subplots(3, 5, figsize=(15, 8))
perplexities = [5, 30, 50, 100, 200]
n_iters = [500, 1000, 5000]
red = y == 0
green = y == 1
ax = subplots[0][0]
ax.scatter(X[red, 0], X[red, 1], c="r")
ax.scatter(X[green, 0], X[green, 1], c="g")
ax.xaxis.set_major_formatter(NullFormatter())
ax.yaxis.set_major_formatter(NullFormatter())
plt.axis('tight')
for i, n_iter in enumerate(n_iters):
for j, perplexity in enumerate(perplexities):
ax = subplots[i][j]
t0 = time()
tsne = manifold.TSNE(n_components=n_components,
init='random',
random_state=0,
perplexity=perplexity,
n_iter=n_iter)
Y = tsne.fit_transform(X)
t1 = time()
print("perplexity=%d in %.2g sec" % (perplexity, t1 - t0))
ax.set_title("Perplexity=%d, iteration =%d" % (perplexity, n_iter))
ax.scatter(Y[red, 0], Y[red, 1], c="r")
ax.scatter(Y[green, 0], Y[green, 1], c="g")
ax.xaxis.set_major_formatter(NullFormatter())
ax.yaxis.set_major_formatter(NullFormatter())
ax.axis('tight')
plt.show()
print(' ')
def run():
# get training set
os.chdir(train_dir)
folder_name = train_dir.split('/')[-2]
# initiate train detections
train_detections = preprocessing.all_detections(folder_name)
for file in glob.glob('*.feat'):
with open(file) as f:
lines = f.readlines()
view_name = file.split('.feat')[0]
# adding detections from one view to this instance
train_detections.update(lines, view_name)#, update_tp_group_only = True)
# get testing set
os.chdir(test_dir)
folder_name = test_dir.split('/')[-2]
# initiate train detections
test_detections = preprocessing.all_detections(folder_name)
for file in glob.glob('*.feat'):
with open(file) as f:
lines = f.readlines()
view_name = file.split('.feat')[0]
# adding detections from one view to this instance
test_detections.update(lines, view_name)
# get train_x and train_y
train_y = np.array(train_detections.get_all_tp_fp())
train_x = np.array([train_detections.get_all_od(), train_detections.get_all_dl(), train_detections.get_all_logit(), train_detections.get_all_area()]).transpose()
# get test_x and test_y
test_y = np.array(test_detections.get_all_tp_fp())
test_x = np.array([test_detections.get_all_od(), test_detections.get_all_dl(), test_detections.get_all_logit(), test_detections.get_all_area()]).transpose()
# scale the features
scaler = StandardScaler()
scaler.fit(train_x)
train_x = scaler.transform(train_x)
test_x = scaler.transform(test_x)
# start the training
mlp = MLPClassifier(hidden_layer_sizes=(8,4))
mlp.fit(train_x, train_y)
# save the model
joblib.dump(mlp, out_dir + 'pick_one_mark_nn.pkl')
joblib.dump(scaler, out_dir + 'pick_one_mark_scaler.pkl')
# predictions on train set
predictions_train = mlp.predict(train_x)
prob_train = mlp.predict_proba(train_x)[:,1]
print('This is the training set result: \n')
print(confusion_matrix(train_y, predictions_train))
print(classification_report(train_y, predictions_train))
# Compute AUC
fpr, tpr, thresholds = roc_curve(train_y, prob_train, pos_label=1)
az_train = auc(fpr, tpr)
print('AUC for train set is ' + str(az_train))
# predictions on test set
predictions_test = mlp.predict(test_x)
prob_test = mlp.predict_proba(test_x)[:,1]
print('This is the testing set result: \n')
print(confusion_matrix(test_y, predictions_test))
print(classification_report(test_y, predictions_test))
# Compute AUC
fpr, tpr, thresholds = roc_curve(test_y, prob_test, pos_label=1)
az_test = auc(fpr, tpr)
print('AUC for testing set is ' + str(az_test))
# record the predictions into the class
train_detections.record_scores(prob_train)
test_detections.record_scores(prob_test)
# Now test on each view and dump txt for scoring
for fn in glob.glob(test_dir + '*.feat'):
folder_name = 'test'
view_name = fn.split('\\')[-1].split('.feat')[0]
test_detections_one_view = preprocessing.all_detections_one_view(folder_name, view_name)
with open(fn) as f:
lines = f.readlines()
test_detections_one_view.update(lines, view_name)
# Get the feature values
test_x = np.array([test_detections_one_view.get_all_od(), test_detections_one_view.get_all_dl(), test_detections_one_view.get_all_logit()]).transpose()
test_x = scaler.transform(test_x)
predictions_test_one_view = mlp.predict_proba(test_x)[:,1]
# record the predictions
test_detections_one_view.record_scores(predictions_test_one_view)
postprocessing.pick_one_mark_one_group_use_prediction(test_detections_one_view)
postprocessing.save_to_txt(test_detections_one_view, out_dir, view_name)