def run_ablation_svm(X, y, class_names, quality): """ Runs ablation tests on svm model (i.e. runs the SVM on only a subset of features to determine which features are more important than others) Args: X: input examples y: labels of examples class_names: list of actual signs quality: flag that determines the quality of the data being passed in Returns: None """ feature_array = preprocessing.get_feature_list(quality) feature_remove_list = [] for feature in feature_array: feature_remove_list.append(feature) ablated_X1 = preprocessing.get_ablated_matrix(X, quality, [feature]) if len(feature_remove_list) > 1 and len(feature_remove_list) < len(feature_array): ablated_X2 = preprocessing.get_ablated_matrix(X, quality, feature_remove_list) #create splits X_train, y_train, X_test, y_test = preprocessing.create_train_test_split(ablated_X1, y, test_size=0.3, shuffle=True) flattened_Xtrain = preprocessing.flatten_matrix(X_train) flattened_Xtest = preprocessing.flatten_matrix(X_test) print "fitting svm on feature set without feature %s..." % feature # fit svm model svm_model = svm.SVC(kernel="linear", decision_function_shape='ovr') svm_model.fit(flattened_Xtrain, y_train) y_predict_train = svm_model.predict(flattened_Xtrain) y_predict = svm_model.predict(flattened_Xtest) print "running analyses on feature set without feature %s..." % feature analysis.run_analyses(y_predict_train, y_train, y_predict, y_test, class_names, True) #true ablation tests if len(feature_remove_list) > 1 and len(feature_remove_list) < len(feature_array): X_train, y_train, X_test, y_test = preprocessing.create_train_test_split(ablated_X2, y, test_size=0.3, shuffle=False) flattened_Xtrain = preprocessing.flatten_matrix(X_train) flattened_Xtest = preprocessing.flatten_matrix(X_test) print "fitting svm on feature set without the features %s..." % ", ".join(feature_remove_list) # fit svm model svm_model = svm.SVC(kernel="linear", decision_function_shape='ovr') svm_model.fit(flattened_Xtrain, y_train) y_predict_train = svm_model.predict(flattened_Xtrain) y_predict = svm_model.predict(flattened_Xtest) print "running analyses on feature set without the features %s..." % ", ".join(feature_remove_list) analysis.run_analyses(y_predict_train, y_train, y_predict, y_test, class_names, True)
def run_logreg(quality='high'):
"""
Runs a simple logistic regression model; first fits the model
on the training data (70 percent of the total data) and tests on
the rest of the data.
Args:
none
Returns:
none
"""
data = io.load_data(quality=quality)
X, y, class_names = preprocessing.create_data_tensor(data)
X_train, y_train, X_test, y_test = preprocessing.create_train_test_split(
X, y, test_size=0.3, shuffle=True)
# flatten data
flattened_Xtrain = preprocessing.flatten_matrix(X_train)
flattened_Xtest = preprocessing.flatten_matrix(X_test)
# fit logistic regression model
logreg_model = linear_model.LogisticRegression(multi_class='ovr')
logreg_model.fit(flattened_Xtrain, y_train)
y_predict_train = logreg_model.predict(flattened_Xtrain)
y_predict = logreg_model.predict(flattened_Xtest)
# print metrics and confusion plot
analysis.run_analyses(y_predict_train, y_train, y_predict, y_test,
class_names)
def run_svm(quality="high", ablation=False, concat=True):
"""
Runs a simple SVM model with a linear kernel; first fits the model
on the training data (70 percent of the total data) and tests on
the rest of the data.
Args:
None
Returns:
None
"""
data = io.load_data(quality=quality)
X, y, class_names = preprocessing.create_data_tensor(data)
if ablation:
run_ablation_svm(X, y, class_names, quality)
return
X_train, y_train, X_test, y_test = preprocessing.create_train_test_split(X, y, test_size=0.3, shuffle=False)
flattened_Xtrain = preprocessing.flatten_matrix(X_train)
flattened_Xtest = preprocessing.flatten_matrix(X_test)
if concat:
X_train_400 = np.load('../data/seq_mining_features/k_2-w_2/X_train-400.npy')
X_test_400 = np.load('../data/seq_mining_features/k_2-w_2/X_test-400.npy')
y_train_400 = np.load('../data/seq_mining_features/k_2-w_2/y_train-400.npy')
y_test_400 = np.load('../data/seq_mining_features/k_2-w_2/y_test-400.npy')
print X_train_400.shape, y_train_400.shape
print flattened_Xtrain.shape, y_train.shape
print '-----------'
print X_test_400.shape, y_test_400.shape
print flattened_Xtest.shape, y_test.shape
flattened_Xtrain_concatenated = np.hstack((flattened_Xtrain, X_train_400[:, 0:75]))
flattened_Xtest_concatenated = np.hstack((flattened_Xtest, X_test_400[:, 0:75]))
print '-----------'
print flattened_Xtrain.shape, y_train.shape
print flattened_Xtest.shape, y_test.shape
C = 0.01
# fit svm model
svm_model = svm.SVC(kernel="linear", C=C, decision_function_shape='ovr')
svm_model.fit(flattened_Xtrain, y_train)
y_predict_train = svm_model.predict(flattened_Xtrain)
y_predict = svm_model.predict(flattened_Xtest)
analysis.run_analyses(y_predict_train, y_train, y_predict, y_test, class_names, ablation=False, confusion=False)
print '-----------==================-----------'
svm_model = svm.SVC(kernel="linear", C=C, decision_function_shape='ovr')
svm_model.fit(flattened_Xtrain_concatenated, y_train)
y_predict_train = svm_model.predict(flattened_Xtrain_concatenated)
y_predict = svm_model.predict(flattened_Xtest_concatenated)
analysis.run_analyses(y_predict_train, y_train, y_predict, y_test, class_names, ablation=False, confusion=False)
print '-----------==================-----------'
def main():
data = io.load_data(quality="low")
X, y, class_names = preprocessing.create_data_tensor(data)
X_train, y_train, X_test, y_test = preprocessing.create_train_test_split(
X, y)
X = preprocessing.scale_spatially(X)
# for i in class_names:
# # print i
# # print class_names[i]
# # print np.where(y==i)
# # print np.where(y==i)[0]
# # print np.where(y==i)[0].size
# print class_names[i], i, np.where(y==i)[0].size
# analysis.plot_signals_two_column(data[class_names[y[0]]][0][:, 0:3],
# X[0, 0:3, :].T,
# ['Raw X', 'Raw Y', 'Raw Z'],
# ['Resampled X', 'Resampled Y', 'Resampled Z'])
shop_idx = np.where(y == 3)[0]
shop_idx = shop_idx[0:6]
# print X[shop_idx, 0, :].shape
NUM = 3
C1 = 0
C2 = 3
d1 = X[np.where(y == C1)[0][0:NUM], 2, :].T
d2 = X[np.where(y == C2)[0][0:NUM], 2, :].T
d1p = np.roll(d1, -1, 0) - d1
# d1p = d1p[0:d1.shape[0]-1]
d1p[-1, :] = d1p[-2, :]
d2p = np.roll(d2, -1, 0) - d2
d2p[-1, :] = d2p[-2, :]
labels1 = [str(class_names[C1]) + ' ' + str(i) for i in xrange(NUM)] + [
str(class_names[C1]) + '\' ' + str(i) for i in xrange(NUM)
]
labels2 = [str(class_names[C2]) + ' ' + str(i) for i in xrange(NUM)] + [
str(class_names[C2]) + '\' ' + str(i) for i in xrange(NUM)
]
print d1.shape, d1p.shape
print d2.shape, d2p.shape
print np.concatenate((d1, d1p), 1).shape
# analysis.plot_signals_two_column(np.concatenate((d1, d1p), 1),
# np.concatenate((d2, d2p), 1),
# labels1,
# labels2)
print class_names[4]
d2 = X[np.where(y == 4)[0][0:2], 2, :].T
analysis.plot_signals(d2, labels2)
def run_nn(quality='low'):
"""
Runs a simple neural network model; first fits the model
on the training data (70 percent of the total data) and tests on
the rest of the data.
Args:
none
Returns:
none
"""
data = io.load_data(quality=quality)
X, y, class_names = preprocessing.create_data_tensor(data)
X_train, y_train, X_test, y_test = preprocessing.create_train_test_split(
X, y, test_size=0.3, shuffle=False)
y_train_one_hot = np.zeros((y_train.shape[0], len(class_names)))
for i in range(y_train.shape[0]):
y_train_one_hot[i, y_train[i]] = 1
y_test_one_hot = np.zeros((y_test.shape[0], len(class_names)))
for i in range(y_test.shape[0]):
y_test_one_hot[i, y_test[i]] = 1
# flatten data
flattened_Xtrain = preprocessing.flatten_matrix(X_train)
flattened_Xtest = preprocessing.flatten_matrix(X_test)
# fit neural network model
HIDDEN_LAYER_SIZE = y_train_one_hot.shape[1]
# HIDDEN_LAYER_SIZE = y_train_one_hot.shape[1]*2
############ FAILED ##################
# HIDDEN_LAYER_SIZE = X_train.shape[1]
#
######################################
nn_model = Sequential()
####################################### MODELS #######################################
####################################### MODEL1 #######################################
# Training Error: 0.517945411562
# Testing Error: 0.778446115288
# Epochs: 200
#
nn_model.add(
Dense(HIDDEN_LAYER_SIZE,
input_dim=flattened_Xtrain.shape[1],
init='uniform',
activation="tanh"))
nn_model.add(Dropout(0.2))
nn_model.add(
Dense(y_train_one_hot.shape[1], init='uniform', activation="tanh"))
N_EPOCHS = 200
####################################### MODEL1 #######################################
####################################### MODEL2 #######################################
# Training Error: 0.
# Testing Error: 0.
#
# nn_model.add(Dense(HIDDEN_LAYER_SIZE, input_dim=flattened_Xtrain.shape[1], init='uniform', activation="tanh"))
# nn_model.add(Dropout(0.2))
# nn_model.add(Dense(HIDDEN_LAYER_SIZE*10, activation="tanh", init='uniform'))
# nn_model.add(Dropout(0.3))
# nn_model.add(Dense(HIDDEN_LAYER_SIZE*2, activation="tanh", init='uniform'))
# nn_model.add(Dense(y_train_one_hot.shape[1], init='uniform', activation="tanh"))
####################################### MODEL2 #######################################
sgd = SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True)
nn_model.compile(loss='mean_squared_error', optimizer=sgd)
nn_model.fit(flattened_Xtrain, y_train_one_hot, nb_epoch=N_EPOCHS)
y_predict_train = nn_model.predict_classes(flattened_Xtrain)
y_predict = nn_model.predict_classes(flattened_Xtest)
# y_predict_one_hot = nn_model.predict(flattened_Xtest)
# print metrics and confusion plot
analysis.run_analyses(y_predict_train, y_train, y_predict, y_test,
class_names)
def seg_mining(use_all_signs):
# Loading data
data = io.load_data(quality="low")
X, y, class_names = preprocessing.create_data_tensor(data)
X_train, y_train, X_test, y_test = preprocessing.create_train_test_split(
X, y, test_size=0.3, shuffle=False)
if not use_all_signs:
# TODO: change not break
X_train = preprocessing.scale_spatially(
X_train)[:NUM_SIGNS * EXAMPLES_PER_SIGN, :NUM_SIGNALS, :]
y_train = y_train[:NUM_SIGNS * EXAMPLES_PER_SIGN]
X_test = preprocessing.scale_spatially(
X_test)[:NUM_SIGNS * (70 - EXAMPLES_PER_SIGN), :NUM_SIGNALS, :]
y_test = y_test[:NUM_SIGNS * (70 - EXAMPLES_PER_SIGN)]
else:
X_train = preprocessing.scale_spatially(X_train)
X_test = preprocessing.scale_spatially(X_test)
# print X.shape
# sys.exit()
# Computing fake slopes
dX_train = np.roll(X_train, -1, 2) - X_train
dX_train[:, :, -1] = dX_train[:, :, -2]
dX_test = np.roll(X_test, -1, 2) - X_test
dX_test[:, :, -1] = dX_test[:, :, -2]
combined_trends, combined_trends_interval_start, combined_trends_interval_end = create_combined_trends(
dX_train)
combined_trends_test, combined_trends_interval_start_test, combined_trends_interval_end_test = create_combined_trends(
dX_test)
# Computing trends
# binary_I_idx_train = (INCREASING_THRESHOLD >= dX_train) & (dX_train > STEADY_THRESHOLD)
# I_idx_train = np.where(binary_I_idx_train == 1)
# binary_VI_idx_train = (VERY_INCREASING_THRESHOLD >= dX_train) & (dX_train > INCREASING_THRESHOLD)
# VI_idx_train = np.where(binary_VI_idx_train == 1)
# WI_idx_train = np.where(dX_train > VERY_INCREASING_THRESHOLD)
# binary_D_idx_train = (-1*INCREASING_THRESHOLD <= dX_train) & (dX_train < -1*STEADY_THRESHOLD)
# D_idx_train = np.where(binary_D_idx_train == 1)
# binary_VD_idx_train = (-1*VERY_INCREASING_THRESHOLD <= dX_train) & (dX_train < -1*INCREASING_THRESHOLD)
# VD_idx_train = np.where(binary_VD_idx_train == 1)
# WD_idx_train = np.where(dX_train < -1*VERY_INCREASING_THRESHOLD)
# S_idx_train = np.where(np.abs(dX_train) <= STEADY_THRESHOLD)
# trends = np.zeros(dX_train.shape, dtype=np.int8)
# trends[I_idx_train] = get_trend_idx_train('I')
# trends[VI_idx_train] = get_trend_idx_train('VI')
# trends[WI_idx_train] = get_trend_idx_train('WI')
# trends[S_idx_train] = get_trend_idx_train('S')
# trends[D_idx_train] = get_trend_idx_train('D')
# trends[VD_idx_train] = get_trend_idx_train('VD')
# trends[WD_idx_train] = get_trend_idx_train('WD')
# I_idx_train = np.where(dX_train > STEADY_THRESHOLD)
# S_idx_train = np.where(np.abs(dX_train) <= STEADY_THRESHOLD)
# D_idx_train = np.where(dX_train < -1*STEADY_THRESHOLD)
# trends = np.zeros(dX_train.shape, dtype=np.int8)
# trends[I_idx_train] = get_trend_idx('I')
# trends[S_idx_train] = get_trend_idx('S')
# trends[D_idx_train] = get_trend_idx('D')
# # print dX_train.shape, np.prod(dX_train.shape)
# # print I_idx_train[0].shape, S_idx_train[0].shape, D_idx_train[0].shape
# # print np.where(dX_train > STEADY_THRESHOLD)
# # print np.where(trends == 0)[0].shape, np.where(trends == 1)[0].shape, np.where(trends == 2)[0].shape
# # Combine trends
# # Intervals are inclusive in nature
# combined_trends = np.ones(trends.shape, dtype=np.int8) * get_trend_idx('useless')
# combined_trends_interval_start = np.ones(trends.shape, dtype=np.int8) * get_trend_idx('useless')
# combined_trends_interval_end = np.ones(trends.shape, dtype=np.int8) * get_trend_idx('useless')
# for e in xrange(combined_trends.shape[0]):
# for f in xrange(combined_trends.shape[1]):
# combined_trend_idx = 0
# combined_trends[e,f,0] = trends[e,f,0]
# combined_trends_interval_start[e,f,0] = 0
# for t in xrange(1, combined_trends.shape[2]):
# if trends[e,f,t] != trends[e,f,t-1]:
# # Set End time for previous trend
# combined_trends_interval_end[e,f,combined_trend_idx] = t
# # Start next trend
# combined_trend_idx += 1
# combined_trends[e,f,combined_trend_idx] = trends[e,f,t]
# combined_trends_interval_start[e,f,combined_trend_idx] = t
# print trends[5, 3, :]
# print combined_trends[5, 3, :]
# print combined_trends_interval_start[5, 3, :]
# print combined_trends_interval_end[5, 3, :]
# NUM_SIGN = 3
# NUM_EXAMPLES = 70
# idx = NUM_SIGN*NUM_EXAMPLES
# print trends[np.where(y==3)[0][0],3,:]
# print combined_trends[np.where(y==3)[0][0],3,:]
# print combined_trends_interval_start[np.where(y==3)[0][0],3,:]
# print combined_trends_interval_end[np.where(y==3)[0][0],3,:]
# support(combined_trends, None, 10)
patterns = [None] * NUM_SIGNS
pattern_counts = Counter()
all_pattern_supports = [None] * NUM_SIGNS
for i in xrange(NUM_SIGNS):
print ''
print 'For sign %d...' % i
print 'Generating one patterns.....'
one_patterns = generate_one_pattern(combined_trends)
# support((combined_trends, combined_trends_interval_start, combined_trends_interval_end),
# 'S:1-o;S:2-b;D:1-o;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1-b;D:1',
# w=5, minsup=20)
# sys.exit()
k = 2
total_time = 0
patterns[i] = [one_patterns]
all_pattern_supports[i] = [None]
while True and k <= K:
# generate_k_patterns
print 'Generating patterns...'
generation_start_time = time.time()
total_time = 0
new_patterns, total_time = generate_k_patterns(
patterns[i][k - 2], k, total_time)
generation_end_time = time.time()
print "k:", k, " -> New patterns:", len(new_patterns)
print "Total time for generation: ", (generation_end_time -
generation_start_time)
print "Total time for comparison: ", (total_time)
if len(new_patterns) == 0: break
# prune_patterns
print 'Pruning patterns...'
combined_trends_sign = combined_trends[i *
EXAMPLES_PER_SIGN:(i + 1) *
EXAMPLES_PER_SIGN, :, :]
combined_trends_interval_start_sign = combined_trends_interval_start[
i * EXAMPLES_PER_SIGN:(i + 1) * EXAMPLES_PER_SIGN, :, :]
combined_trends_interval_end_sign = combined_trends_interval_end[
i * EXAMPLES_PER_SIGN:(i + 1) * EXAMPLES_PER_SIGN, :, :]
pruned_patterns, pattern_supports = prune_patterns(
(combined_trends_sign, combined_trends_interval_start_sign,
combined_trends_interval_end_sign), new_patterns, MIN_SUPPORT,
WINDOW_SIZE)
all_pattern_supports[i].append(pattern_supports)
for pattern in pattern_supports:
pattern_counts[pattern] += 1
print "Pruned Patterns:", len(pruned_patterns)
# increment k
k += 1
# If no k patterns, break
if len(pruned_patterns) == 0:
break
#print pruned_patterns.keys()[0]
patterns[i].append(pruned_patterns)
# patterns.append(new_patterns)
# print "pruned patterns: "
# print patterns[i]
# print all_pattern_supports[i]
ranked_patterns = chi_square(patterns, pattern_counts,
all_pattern_supports)[:NUM_PATTERN_FEATURES]
# cut some ranked_patterns out
X_train_new = construct_feature_vectors(ranked_patterns, \
(combined_trends, combined_trends_interval_start, combined_trends_interval_end), EXAMPLES_PER_SIGN)
X_test_new = construct_feature_vectors(ranked_patterns, \
(combined_trends_test, combined_trends_interval_start_test, combined_trends_interval_end_test), 70 - EXAMPLES_PER_SIGN)
# print ranked_patterns[:NUM_PATTERN_FEATURES]
# print len(ranked_patterns)
# print X_train_new
# print y_train
svm_model = svm.SVC(C=.1, kernel="linear", decision_function_shape='ovr')
svm_model.fit(X_train_new, y_train)
y_predict_train = svm_model.predict(X_train_new)
y_predict = svm_model.predict(X_test_new)
analysis.run_analyses(y_predict_train, y_train, y_predict, y_test,
class_names)
def seg_mining(use_all_signs):
# Loading data
data = io.load_data(quality="low")
X, y, class_names = preprocessing.create_data_tensor(data)
X_train, y_train, X_test, y_test = preprocessing.create_train_test_split(X, y, test_size=0.3, shuffle=False)
# for class_idx in np.unique(y):
# print class_names[class_idx], np.where(class_idx == y)[0].shape[0]
# import sys
# sys.exit(1)
if not use_all_signs:
# TODO: change not break
X_train = preprocessing.scale_spatially(X_train)[:NUM_SIGNS * EXAMPLES_PER_SIGN,:NUM_SIGNALS,:]
y_train = y_train[:NUM_SIGNS * EXAMPLES_PER_SIGN]
X_test = preprocessing.scale_spatially(X_test)[:NUM_SIGNS * (70 - EXAMPLES_PER_SIGN),:NUM_SIGNALS,:]
y_test = y_test[:NUM_SIGNS * (70 - EXAMPLES_PER_SIGN)]
else:
X_train = preprocessing.scale_spatially(X_train)
X_test = preprocessing.scale_spatially(X_test)
# Computing fake slopes
dX_train = X_train
dX_test = X_test
# dX_train = np.roll(X_train, -1, 2) - X_train
# dX_train[:, :, -1] = dX_train[:, :, -2]
# dX_test = np.roll(X_test, -1, 2) - X_test
# dX_test[:, :, -1] = dX_test[:, :, -2]
combined_trends, combined_trends_interval_start, combined_trends_interval_end = sq.create_combined_trends_discritized(dX_train)
combined_trends_test, combined_trends_interval_start_test, combined_trends_interval_end_test = sq.create_combined_trends_discritized(dX_test)
patterns = [None] * NUM_SIGNS
pattern_counts = Counter()
all_pattern_supports = [None] * NUM_SIGNS
threads = []
for i in xrange(NUM_SIGNS):
process_sign(i, combined_trends, combined_trends_interval_start, combined_trends_interval_end, patterns, pattern_counts, all_pattern_supports)
# thread = Thread(target=process_sign, args=(i, combined_trends, combined_trends_interval_start, combined_trends_interval_end, patterns, pattern_counts, all_pattern_supports))
# thread.start()
# threads.append(thread)
# for thread in threads:
# thread.join()
# process_sign(i, combined_trends, combined_trends_interval_start, combined_trends_interval_end, patterns, pattern_counts, all_pattern_supports)
# for pattern_supports in all_pattern_supports[i]:
# if pattern_supports is not None:
# for pattern in pattern_supports:
# pattern_counts[pattern] += 1
# cut some ranked_patterns out
print 'Classes',len(patterns), len(all_pattern_supports)
print 'Iterations',len(patterns[1]), len(all_pattern_supports[1])
# print 'patterns in iteration 1 ',len(patterns[1][0]), len(all_pattern_supports[1][0])
ranked_patterns_all = sq.chi_square(patterns, pattern_counts, all_pattern_supports)
# NUM_PATTERN_FEATURES_ARRAY = [75, 100, 200, 300, 400, 500, 750, 1000, 1500]
# NUM_PATTERN_FEATURES_ARRAY = [400, 500, 750, 1000, 1500, 2000, 2500]
NUM_PATTERN_FEATURES_ARRAY = [NUM_PATTERN_FEATURES]
for num_pattern_features in NUM_PATTERN_FEATURES_ARRAY:
print '---------------------=================---------------------'
ranked_patterns = ranked_patterns_all[:num_pattern_features]
print len(ranked_patterns_all),'->',len(ranked_patterns)
X_train_new = sq.construct_feature_vectors(ranked_patterns, \
(combined_trends, combined_trends_interval_start, combined_trends_interval_end), EXAMPLES_PER_SIGN)
X_test_new = sq.construct_feature_vectors(ranked_patterns, \
(combined_trends_test, combined_trends_interval_start_test, combined_trends_interval_end_test), 70 - EXAMPLES_PER_SIGN)
svm_model = svm.SVC(C=PENALTY, kernel="linear", decision_function_shape='ovr')
svm_model.fit(X_train_new, y_train)
y_predict_train = svm_model.predict(X_train_new)
y_predict = svm_model.predict(X_test_new)
# np.save('../data/seq_mining_features/X_train-%d.npy'%num_pattern_features, X_train_new)
# np.save('../data/seq_mining_features/X_test-%d.npy'%num_pattern_features, X_test_new)
# np.save('../data/seq_mining_features/y_train-%d.npy'%num_pattern_features, y_train)
# np.save('../data/seq_mining_features/y_test-%d.npy'%num_pattern_features, y_test)
analysis.run_analyses(y_predict_train, y_train, y_predict, y_test, class_names)
print 'STEADY_THRESHOLD',STEADY_THRESHOLD
print 'WINDOW_SIZE',WINDOW_SIZE
print 'MIN_SUPPORT',MIN_SUPPORT
print 'K',K
print 'NUM_PATTERN_FEATURES',num_pattern_features
print 'NUM_SIGNS',NUM_SIGNS
print 'NUM_SIGNALS',NUM_SIGNALS
print 'EXAMPLES_PER_SIGN',EXAMPLES_PER_SIGN
print 'PENALTY',PENALTY
print '---------------------=================---------------------'
def run_nn(quality='low'):
"""
Runs a simple neural network model; first fits the model
on the training data (70 percent of the total data) and tests on
the rest of the data.
Args:
none
Returns:
none
"""
data = io.load_data(quality=quality)
X, y, class_names = preprocessing.create_data_tensor(data)
X_train, y_train, X_test, y_test = preprocessing.create_train_test_split(
X, y, test_size=0.3, shuffle=False)
y_train_one_hot = np.zeros((y_train.shape[0], len(class_names)))
for i in range(y_train.shape[0]):
y_train_one_hot[i, y_train[i]] = 1
y_test_one_hot = np.zeros((y_test.shape[0], len(class_names)))
for i in range(y_test.shape[0]):
y_test_one_hot[i, y_test[i]] = 1
# flip last two axes
# tensor: samples x features x time
# tensor: samples x time x features
X_train = np.swapaxes(X_train, 1, 2)
X_test = np.swapaxes(X_test, 1, 2)
print X_train.shape
print y_train_one_hot.shape
lstm_model = Sequential()
############################## MODEL 1 ##############################
# Average Precision: 0.109463986213
# Average Recall: 0.0912280701754
# Average F1: 0.0664461045921
# Training Error: 0.852201933405
# Testing Error: 0.908771929825
# HIDDEN_LAYER = 300
# lstm_model.add(LSTM(output_dim=HIDDEN_LAYER,
# input_dim=X_train.shape[2],
# activation='tanh',
# inner_activation='hard_sigmoid',
# return_sequences=False))
# lstm_model.add(Dense(y_train_one_hot.shape[1], activation='tanh'))
# lstm_model.compile(loss='mean_squared_error', optimizer='rmsprop')
# lstm_model.fit(X_train, y_train_one_hot, batch_size=16, nb_epoch=10)
############################## MODEL 1 ##############################
#
############################### MODEL 2 ##############################
# Average Precision: 0.109463986213
# Average Recall: 0.0912280701754
# Average F1: 0.0664461045921
# Training Error: 0.852201933405
# Testing Error: 0.908771929825
HIDDEN_LAYER = 300
lstm_model.add(
LSTM(output_dim=HIDDEN_LAYER,
input_dim=X_train.shape[2],
activation='tanh',
inner_activation='hard_sigmoid',
return_sequences=False))
lstm_model.add(
LSTM(HIDDEN_LAYER,
activation='tanh',
inner_activation='hard_sigmoid',
return_sequences=False))
lstm_model.add(Dense(y_train_one_hot.shape[1], activation='tanh'))
lstm_model.compile(loss='mean_squared_error', optimizer='rmsprop')
lstm_model.fit(X_train, y_train_one_hot, batch_size=16, nb_epoch=10)
############################## MODEL 2 ##############################
# lstm_model.add(Dropout(0.5))
# lstm_model.add(Dense(95))
# lstm_model.add(Activation('softmax'))
# score = model.evaluate(X_test, Y_test, batch_size=16)
y_predict_train = lstm_model.predict_classes(X_train)
y_predict = lstm_model.predict_classes(X_test)
# y_predict_one_hot = nn_model.predict(flattened_Xtest)
# print y_predict
# print y_predict_one_hot
# print metrics and confusion plot
analysis.run_analyses(y_predict_train, y_train, y_predict, y_test,
class_names, False, False)
print y_predict