def bpic(lstmsize, lstmdropout, lstmoptim):
lstmsize = lstmsize[0] * 10
lstmdropout = lstmdropout[0]
lstmoptim = lstmoptim[0]
lstmnepochs = 50
lstmbatchsize = 64
nfolds = 5
# Load the dataset
#static_train = np.load('/storage/hpc_anna/GMiC/Data/BPIChallenge/f1/preprocessed/train_static.npy')
dynamic_train = np.load('/storage/hpc_anna/GMiC/Data/BPIChallenge/f1/preprocessed/train_dynamic.npy')
#static_val = np.load('/storage/hpc_anna/GMiC/Data/BPIChallenge/f1/preprocessed/test_static.npy')
dynamic_val = np.load('/storage/hpc_anna/GMiC/Data/BPIChallenge/f1/preprocessed/test_dynamic.npy')
labels_train = np.load('/storage/hpc_anna/GMiC/Data/BPIChallenge/f1/preprocessed/train_labels.npy')
labels_val = np.load('/storage/hpc_anna/GMiC/Data/BPIChallenge/f1/preprocessed/test_labels.npy')
# Merge train and test
#static_all = np.concatenate((static_train, static_val), axis=0)
dynamic_all = np.concatenate((dynamic_train, dynamic_val), axis=0)
labels_all = np.concatenate((labels_train, labels_val), axis=0)
nsamples = dynamic_all.shape[0]
# k-fold cross validation to obtain accuracy
val_idx_list = np.array_split(range(nsamples), nfolds)
scores = []
for fid, val_idx in enumerate(val_idx_list):
train_idx = list(set(range(nsamples)) - set(val_idx))
# LSTM on dynamic features (8)
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs, lstmbatchsize)
model_pos, model_neg = lstmcl.train(dynamic_all[train_idx], labels_all[train_idx])
scores.append(lstmcl.test(model_pos, model_neg, dynamic_all[val_idx], labels_all[val_idx]))
print 'Result: %.4f' % np.mean(scores)
return -np.mean(scores)
def function(lstmsize, lstmdropout, lstmoptim):
lstmsize = lstmsize[0] * 10
lstmdropout = lstmdropout[0]
lstmoptim = lstmoptim[0]
lstmnepochs = 50
lstmbatchsize = 64
nfolds = 5
print("Reading data...")
dynamic_train = np.load('/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/train_data.npy')
labels_train = np.load('/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/train_labels.npy')
nsamples = dynamic_train.shape[0]
# k-fold cross validation to obtain accuracy
val_idx_list = np.array_split(range(nsamples), nfolds)
scores = []
for fid, val_idx in enumerate(val_idx_list):
train_idx = list(set(range(nsamples)) - set(val_idx))
print "Current fold is %d / %d" % (fid + 1, nfolds)
# LSTM on dynamic features (8)
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs, lstmbatchsize)
model_pos, model_neg = lstmcl.train(dynamic_train[train_idx], labels_train[train_idx])
scores.append(lstmcl.test(model_pos, model_neg, dynamic_train[val_idx], labels_train[val_idx]))
print 'Result: %.4f' % np.mean(scores)
return -np.mean(scores)
def function(lstmsize, lstmdropout, lstmoptim):
lstmsize = lstmsize[0] * 10
lstmdropout = lstmdropout[0]
lstmoptim = lstmoptim[0]
lstmnepochs = 50
lstmbatchsize = 64
nfolds = 5
print("Reading data...")
dynamic_train = np.load(
'/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/train_data.npy')
labels_train = np.load(
'/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/train_labels.npy')
nsamples = dynamic_train.shape[0]
# k-fold cross validation to obtain accuracy
val_idx_list = np.array_split(range(nsamples), nfolds)
scores = []
for fid, val_idx in enumerate(val_idx_list):
train_idx = list(set(range(nsamples)) - set(val_idx))
print "Current fold is %d / %d" % (fid + 1, nfolds)
# LSTM on dynamic features (8)
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs,
lstmbatchsize)
model_pos, model_neg = lstmcl.train(dynamic_train[train_idx],
labels_train[train_idx])
scores.append(
lstmcl.test(model_pos, model_neg, dynamic_train[val_idx],
labels_train[val_idx]))
print 'Result: %.4f' % np.mean(scores)
return -np.mean(scores)
def modelopt(lstmsize, lstmdropout, lstmoptim, nestimators):
lstmsize = lstmsize[0] * 10
lstmdropout = lstmdropout[0]
lstmoptim = lstmoptim[0]
lstmnepochs = 50
lstmbatchsize = 64
nestimators = nestimators[0] * 100
nfolds = 5
print lstmsize, lstmdropout, lstmoptim, nestimators
# Load the dataset
static_train = np.load("/storage/hpc_anna/GMiC/Data/ECoGmixed/fourier/train_data.npy")
dynamic_train = np.load("/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/train_data.npy")
static_val = np.load("/storage/hpc_anna/GMiC/Data/ECoGmixed/fourier/test_data.npy")
dynamic_val = np.load("/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/test_data.npy")
labels_train = np.load("/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/train_labels.npy")
labels_val = np.load("/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/test_labels.npy")
# Merge train and test
static_all = np.concatenate((static_train, static_val), axis=0)
dynamic_all = np.concatenate((dynamic_train, dynamic_val), axis=0)
labels_all = np.concatenate((labels_train, labels_val), axis=0)
nsamples = static_all.shape[0]
# prepare where to store the ratios
ratios_all_lstm = np.empty(len(labels_all))
# split indices into folds
enrich_idx_list = np.array_split(range(nsamples), nfolds)
# run CV
for fid, enrich_idx in enumerate(enrich_idx_list):
train_idx = list(set(range(nsamples)) - set(enrich_idx))
# extract predictions using LSTM on dynamic
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs, lstmbatchsize)
model_pos, model_neg = lstmcl.train(dynamic_all[train_idx], labels_all[train_idx])
ratios_all_lstm[enrich_idx] = lstmcl.pos_neg_ratios(model_pos, model_neg, dynamic_all[enrich_idx])
# dataset for hybrid learning
enriched_by_lstm = np.concatenate((static_all, np.matrix(ratios_all_lstm).T), axis=1)
# (2.) k-fold cross validation to obtain accuracy
val_idx_list = np.array_split(range(nsamples), nfolds)
scores = []
for fid, val_idx in enumerate(val_idx_list):
train_idx = list(set(range(nsamples)) - set(val_idx))
# Hybrid on features enriched by HMM (3)
rf = RandomForestClassifier(n_estimators=nestimators)
rf.fit(enriched_by_lstm[train_idx], labels_all[train_idx])
scores.append(rf.score(enriched_by_lstm[val_idx], labels_all[val_idx]))
print "Result: %.4f" % np.mean(scores)
return -np.mean(scores)
def bpic(lstmsize, lstmdropout, lstmoptim):
lstmsize = lstmsize[0] * 10
lstmdropout = lstmdropout[0]
lstmoptim = lstmoptim[0]
lstmnepochs = 50
lstmbatchsize = 64
nfolds = 5
# Load the dataset
#static_train = np.load('/storage/hpc_anna/GMiC/Data/BPIChallenge/f1/preprocessed/train_static.npy')
dynamic_train = np.load(
'/storage/hpc_anna/GMiC/Data/BPIChallenge/f1/preprocessed/train_dynamic.npy'
)
#static_val = np.load('/storage/hpc_anna/GMiC/Data/BPIChallenge/f1/preprocessed/test_static.npy')
dynamic_val = np.load(
'/storage/hpc_anna/GMiC/Data/BPIChallenge/f1/preprocessed/test_dynamic.npy'
)
labels_train = np.load(
'/storage/hpc_anna/GMiC/Data/BPIChallenge/f1/preprocessed/train_labels.npy'
)
labels_val = np.load(
'/storage/hpc_anna/GMiC/Data/BPIChallenge/f1/preprocessed/test_labels.npy'
)
# Merge train and test
#static_all = np.concatenate((static_train, static_val), axis=0)
dynamic_all = np.concatenate((dynamic_train, dynamic_val), axis=0)
labels_all = np.concatenate((labels_train, labels_val), axis=0)
nsamples = dynamic_all.shape[0]
# k-fold cross validation to obtain accuracy
val_idx_list = np.array_split(range(nsamples), nfolds)
scores = []
for fid, val_idx in enumerate(val_idx_list):
train_idx = list(set(range(nsamples)) - set(val_idx))
# LSTM on dynamic features (8)
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs,
lstmbatchsize)
model_pos, model_neg = lstmcl.train(dynamic_all[train_idx],
labels_all[train_idx])
scores.append(
lstmcl.test(model_pos, model_neg, dynamic_all[val_idx],
labels_all[val_idx]))
print 'Result: %.4f' % np.mean(scores)
return -np.mean(scores)
train_half = train_nsamples / 2 trainA_static = train_static[:train_half] trainB_static = train_static[train_half:] trainA_dynamic = train_dynamic[:train_half] trainB_dynamic = train_dynamic[train_half:] trainA_labels = train_labels[:train_half] trainB_labels = train_labels[train_half:] # # Train enrichment models on trainA # print 'Training enrichment models...' # train LSTM activations extractor lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs, lstmbatchsize) model_pos, model_neg = lstmcl.train(trainA_dynamic, trainA_labels) trainB_activations_pos = lstmcl.activations(model_pos, trainB_dynamic) trainB_activations_neg = lstmcl.activations(model_neg, trainB_dynamic) trainB_activations = np.concatenate((trainB_activations_pos[:, 499, :], trainB_activations_neg[:, 499, :]), axis=1) test_activations_pos = lstmcl.activations(model_pos, test_dynamic) test_activations_neg = lstmcl.activations(model_neg, test_dynamic) test_activations = np.concatenate((test_activations_pos[:, 499, :], test_activations_neg[:, 499, :]), axis=1) # # Prepare combined datasets for the future experiments # # datasets for hybrid learning
# Train enrichment models on trainA
#
print 'Training enrichment models...'
# extract predictions using RF on static
rf = RandomForestClassifier(n_estimators=nestimators)
rf.fit(trainA_static, trainA_labels)
predictions_trainB_rf = rf.predict_log_proba(trainB_static)
predictions_trainB_rf[predictions_trainB_rf == -inf] = np.min(
predictions_trainB_rf[predictions_trainB_rf != -inf])
predictions_test_rf = rf.predict_log_proba(test_static)
predictions_test_rf[predictions_test_rf == -inf] = np.min(
predictions_test_rf[predictions_test_rf != -inf])
# extract predictions using LSTM on dynamic
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs,
lstmbatchsize)
model_pos, model_neg = lstmcl.train(trainA_dynamic, trainA_labels)
mse_pos, mse_neg = lstmcl.predict_mse(model_pos, model_neg, trainB_dynamic)
predictions_trainB_lstm = np.vstack((mse_pos, mse_neg)).T
ratios_trainB_lstm = lstmcl.pos_neg_ratios(model_pos, model_neg,
trainB_dynamic)
mse_pos, mse_neg = lstmcl.predict_mse(model_pos, model_neg, test_dynamic)
predictions_test_lstm = np.vstack((mse_pos, mse_neg)).T
ratios_test_lstm = lstmcl.pos_neg_ratios(model_pos, model_neg, test_dynamic)
#
# Prepare combined datasets for the future experiments
#
# datasets for ensemble learning
trainB_predictions_combined_rf_lstm = np.concatenate(
labels_all = np.concatenate((labels_train, labels_val), axis=0)
nsamples = static_all.shape[0]
# prepare where to store the ratios
activations_all = np.empty((len(labels_all), lstmsize * 2))
# split indices into folds
enrich_idx_list = np.array_split(range(nsamples), nfolds)
# CV for feature enrichment
for fid, enrich_idx in enumerate(enrich_idx_list):
print "Current fold is %d / %d" % (fid + 1, nfolds)
train_idx = list(set(range(nsamples)) - set(enrich_idx))
# train models for enrichment
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs, lstmbatchsize)
model_pos, model_neg = lstmcl.train(dynamic_all[train_idx], labels_all[train_idx])
# extract activations
activations_pos = lstmcl.activations(model_pos, dynamic_all[enrich_idx])
activations_neg = lstmcl.activations(model_neg, dynamic_all[enrich_idx])
activations_all[enrich_idx] = np.concatenate((activations_pos[:, -1, :], activations_neg[:, -1, :]), axis=1)
# dataset for hybrid learning
enriched_by_lstm = np.concatenate((static_all, activations_all), axis=1)
print static_all.shape
print enriched_by_lstm.shape
# CV for accuracy estimation
val_idx_list = np.array_split(range(nsamples), nfolds)
scores = []
rf = RandomForestClassifier(n_estimators=nestimators)
rf.fit(static_all[train_idx], labels_all[train_idx])
predictions_all_rf[predict_idx] = rf.predict_log_proba(static_all[predict_idx])
predictions_all_rf[predictions_all_rf == -inf] = np.min(predictions_all_rf[predictions_all_rf != -inf])
# extract predictions using HMM on dynamic
print " Extracting predictions on dynamic data with HMM..."
hmmcl = HMMClassifier()
model_pos, model_neg = hmmcl.train(nhmmstates, nhmmiter, hmmcovtype,
dynamic_all[train_idx], labels_all[train_idx])
predictions_all_hmm[predict_idx] = hmmcl.predict_log_proba(model_pos, model_neg, dynamic_all[predict_idx])
ratios_all_hmm[predict_idx] = hmmcl.pos_neg_ratios(model_pos, model_neg, dynamic_all[predict_idx])
# extract predictions using LSTM on dynamic
print " Extracting predictions on dynamic data with LSTM..."
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs)
model_pos, model_neg = lstmcl.train(dynamic_all[train_idx], labels_all[train_idx])
mse_pos, mse_neg = lstmcl.predict_mse(model_pos, model_neg, dynamic_all[predict_idx])
predictions_all_lstm[predict_idx] = np.vstack((mse_pos, mse_neg)).T
ratios_all_lstm[predict_idx] = lstmcl.pos_neg_ratios(model_pos, model_neg, dynamic_all[predict_idx])
#
# Prepare combined datasets for the future experiments
#
# datasets for ensemble learning
predictions_combined_rf_hmm = np.concatenate((predictions_all_rf, predictions_all_hmm), axis=1)
predictions_combined_rf_lstm = np.concatenate((predictions_all_rf, predictions_all_lstm), axis=1)
# datasets for hybrid learning
# general parameters
lstm_nepochs = 20
# load the dataset
print "Loading the dataset.."
static_train = np.load("/storage/hpc_anna/GMiC/Data/ECoGmixed/fourier/train_data.npy")
dynamic_train = np.load("/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/train_data.npy")
static_val = np.load("/storage/hpc_anna/GMiC/Data/ECoGmixed/fourier/test_data.npy")
dynamic_val = np.load("/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/test_data.npy")
labels_train = np.load("/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/train_labels.npy")
labels_val = np.load("/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/test_labels.npy")
# transform static features into "fake" sequences
dynamized_static_train = np.zeros((static_train.shape[0], static_train.shape[1], dynamic_train.shape[2]))
for i in range(static_train.shape[0]):
dynamized_static_train[i, :, :] = np.tile(static_train[i, :], (dynamic_train.shape[2], 1)).T
dynamized_static_val = np.zeros((static_val.shape[0], static_val.shape[1], dynamic_val.shape[2]))
for i in range(static_val.shape[0]):
dynamized_static_val[i, :, :] = np.tile(static_val[i, :], (dynamic_val.shape[2], 1)).T
# meld dynamized static and dynamic features together
all_train = np.concatenate((dynamized_static_train, dynamic_train), axis=1)
all_val = np.concatenate((dynamized_static_val, dynamic_val), axis=1)
# dynamic data with LSTM
lstmcl = LSTMClassifier(2000, 0.5, "adagrad", lstm_nepochs)
model_pos, model_neg = lstmcl.train(all_train, labels_train)
print "LSTM with dynamized static and dynamic features on validation set: %.4f" % lstmcl.test(
model_pos, model_neg, all_val, labels_val
)
print " Extracting predictions on static data with RF..."
rf = RandomForestClassifier(n_estimators=nestimators)
rf.fit(static_all[train_idx], labels_all[train_idx])
predictions_all_rf[predict_idx] = rf.predict_log_proba(static_all[predict_idx])
predictions_all_rf[predictions_all_rf == -inf] = np.min(predictions_all_rf[predictions_all_rf != -inf])
# extract predictions using HMM on dynamic
#print " Extracting predictions on dynamic data with HMM..."
#hmmcl = HMMClassifier()
#model_pos, model_neg = hmmcl.train(nhmmstates, nhmmiter, hmmcovtype, dynamic_all[train_idx], labels_all[train_idx])
#predictions_all_hmm[predict_idx] = hmmcl.predict_log_proba(model_pos, model_neg, dynamic_all[predict_idx])
#ratios_all_hmm[predict_idx] = hmmcl.pos_neg_ratios(model_pos, model_neg, dynamic_all[predict_idx])
# extract predictions using LSTM on dynamic
print " Extracting predictions on dynamic data with LSTM..."
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs, lstmbatchsize)
model_pos, model_neg = lstmcl.train(dynamic_all[train_idx], labels_all[train_idx])
mse_pos, mse_neg = lstmcl.predict_mse(model_pos, model_neg, dynamic_all[predict_idx])
predictions_all_lstm[predict_idx] = np.vstack((mse_pos, mse_neg)).T
ratios_all_lstm[predict_idx] = lstmcl.pos_neg_ratios(model_pos, model_neg, dynamic_all[predict_idx])
#
# Prepare combined datasets for the future experiments
#
# datasets for ensemble learning
#predictions_combined_rf_hmm = np.concatenate((predictions_all_rf, ratios_all_hmm.reshape((ratios_all_hmm.shape[0], 1))), axis=1)
predictions_combined_rf_lstm = np.concatenate((predictions_all_rf, ratios_all_lstm.reshape((ratios_all_lstm.shape[0], 1))), axis=1)
# datasets for hybrid learning
dynamic_train = np.load(
'/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/train_data.npy')
static_val = np.load(
'/storage/hpc_anna/GMiC/Data/ECoGmixed/fourier/test_data.npy')
dynamic_val = np.load(
'/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/test_data.npy')
labels_train = np.load(
'/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/train_labels.npy')
labels_val = np.load(
'/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/test_labels.npy')
nsamples = dynamic_train.shape[0]
# split the data into training and test
train_idx = np.random.choice(range(0, nsamples),
size=np.round(nsamples * 0.7, 0),
replace=False)
test_idx = list(set(range(0, nsamples)) - set(train_idx))
# train the model and report performance
print 'Training the model...'
lstmcl = LSTMClassifier(lstmsize,
lstmdropout,
lstmoptim,
lstmnepochs,
lstmbatch,
validation_split=0.3)
model_pos, model_neg = lstmcl.train(dynamic_train[train_idx],
labels_train[train_idx])
print 'Generative LSTM classifier on dynamic features: %.4f' % lstmcl.test(
model_pos, model_neg, dynamic_train[test_idx], labels_train[test_idx])
(len(labels_all), g_lstmsize * 2)) # LSTM size in generative case
activations_discriminative = np.empty(
(len(labels_all), 100)) # FC layer in disciminative case
predict_idx_list = np.array_split(range(nsamples), nfolds)
for fid, predict_idx in enumerate(predict_idx_list):
print "Current fold is %d" % fid
train_idx = list(set(range(nsamples)) - set(predict_idx))
#
# Generative LSTM
#
print " Extracting ratios and activations from generative LSTM..."
# train the models
lstmcl = LSTMClassifier(g_lstmsize, g_lstmdropout, g_lstmoptim,
g_lstmnepochs, g_lstmbatch)
model_pos, model_neg = lstmcl.train(dynamic_all[train_idx],
labels_all[train_idx])
# extract ratios
mse_pos, mse_neg = lstmcl.predict_mse(model_pos, model_neg,
dynamic_all[predict_idx])
ratios_generative[predict_idx] = lstmcl.pos_neg_ratios(
model_pos, model_neg, dynamic_all[predict_idx])
# extract activations
activations_pos = lstmcl.activations(model_pos, dynamic_all[predict_idx])
activations_neg = lstmcl.activations(model_neg, dynamic_all[predict_idx])
activations_generative[predict_idx] = np.concatenate(
(activations_pos[:, -1, :], activations_neg[:, -1, :]), axis=1)
# dynamic data with HMM hmmcl = HMMClassifier() model_pos, model_neg = hmmcl.train(3, 10, dynamic_train, labels_train) print "HMM with dynamic features on validation set: %.4f" % hmmcl.test(model_pos, model_neg, dynamic_val, labels_val) # dynamic data with RF print "Training RF on the dynamic dataset..." dynamic_as_static_train = dynamic_train.reshape((dynamic_train.shape[0], dynamic_train.shape[1] * dynamic_train.shape[2])) dynamic_as_static_val = dynamic_val.reshape((dynamic_val.shape[0], dynamic_val.shape[1] * dynamic_val.shape[2])) rf = RandomForestClassifier(n_estimators=rf_estimators, n_jobs=-1) rf.fit(dynamic_as_static_train, labels_train) print "RF with dynamic features on validation set: %.4f" % rf.score(dynamic_as_static_val, labels_val) # dynamic data with LSTM lstmcl = LSTMClassifier(2000, 0.5, 'adagrad', lstm_nepochs) model_pos, model_neg = lstmcl.train(dynamic_train, labels_train) print "LSTM with dynamic features on validation set: %.4f" % lstmcl.test(model_pos, model_neg, dynamic_val, labels_val) # joint models print "" print "Splitting data in two halves..." fh_idx = np.random.choice(range(0, dynamic_train.shape[0]), size=np.round(dynamic_train.shape[0] * 0.5, 0), replace=False) sh_idx = list(set(range(0, dynamic_train.shape[0])) - set(fh_idx)) fh_data = dynamic_train[fh_idx, :, :] fh_labels = labels_train[fh_idx] sh_data = dynamic_train[sh_idx, :, :] sh_labels = labels_train[sh_idx] # RF+HMM print "Evaluating RF+HMM model:"
labels_all = np.concatenate((labels_train, labels_val), axis=0)
nsamples = static_all.shape[0]
# prepare where to store the ratios
ratios_all_lstm = np.empty(len(labels_all))
# split indices into folds
enrich_idx_list = np.array_split(range(nsamples), nfolds)
# CV for feature enrichment
for fid, enrich_idx in enumerate(enrich_idx_list):
print "Current fold is %d / %d" % (fid + 1, nfolds)
train_idx = list(set(range(nsamples)) - set(enrich_idx))
# extract ratios using LSTM on dynamic
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs, lstmbatchsize)
model_pos, model_neg = lstmcl.train(dynamic_all[train_idx], labels_all[train_idx])
ratios_all_lstm[enrich_idx] = lstmcl.pos_neg_ratios(model_pos, model_neg, dynamic_all[enrich_idx])
# dataset for hybrid learning
enriched_by_lstm = np.concatenate((static_all, np.matrix(ratios_all_lstm).T), axis=1)
# CV for accuracy estimation
val_idx_list = np.array_split(range(nsamples), nfolds)
scores = []
for fid, val_idx in enumerate(val_idx_list):
print "Current fold is %d / %d" % (fid + 1, nfolds)
train_idx = list(set(range(nsamples)) - set(val_idx))
# Hybrid on features enriched by HMM (3)
rf = RandomForestClassifier(n_estimators=nestimators)
activations_generative = np.empty((len(labels_all), g_lstmsize * 2)) # LSTM size in generative case
activations_discriminative = np.empty((len(labels_all), 100)) # FC layer in disciminative case
predict_idx_list = np.array_split(range(nsamples), nfolds)
for fid, predict_idx in enumerate(predict_idx_list):
print "Current fold is %d" % fid
train_idx = list(set(range(nsamples)) - set(predict_idx))
#
# Generative LSTM
#
print " Extracting ratios and activations from generative LSTM..."
# train the models
lstmcl = LSTMClassifier(g_lstmsize, g_lstmdropout, g_lstmoptim, g_lstmnepochs, g_lstmbatch)
model_pos, model_neg = lstmcl.train(dynamic_all[train_idx], labels_all[train_idx])
# extract ratios
mse_pos, mse_neg = lstmcl.predict_mse(model_pos, model_neg, dynamic_all[predict_idx])
ratios_generative[predict_idx] = lstmcl.pos_neg_ratios(model_pos, model_neg, dynamic_all[predict_idx])
# extract activations
activations_pos = lstmcl.activations(model_pos, dynamic_all[predict_idx])
activations_neg = lstmcl.activations(model_neg, dynamic_all[predict_idx])
activations_generative[predict_idx] = np.concatenate((activations_pos[:, -1, :], activations_neg[:, -1, :]), axis=1)
#
# Discriminative LSTM
#
#
# k-fold CV for performance estimation
#
val_idx_list = np.array_split(range(nsamples), nfolds)
scores = {1: [], 2: [], 3: [], 4: [], 5: [], 6: [], 7: [], 8: [], 9: [], 10: [], 11: []}
for fid, val_idx in enumerate(val_idx_list):
print "Current fold is %d / %d" % (fid + 1, nfolds)
train_idx = list(set(range(nsamples)) - set(val_idx))
# RF on dynamic features (6)
rf = RandomForestClassifier(n_estimators=nestimators)
rf.fit(dynamic_as_static[train_idx], labels_all[train_idx])
scores[6].append(rf.score(dynamic_as_static[val_idx], labels_all[val_idx]))
# HMM on dynamic features (7)
hmmcl = HMMClassifier()
model_pos, model_neg = hmmcl.train(nhmmstates, nhmmiter, hmmcovtype, dynamic_all[train_idx], labels_all[train_idx])
acc, auc = hmmcl.test(model_pos, model_neg, dynamic_all[val_idx], labels_all[val_idx])
scores[7].append(acc)
# LSTM on dynamic features (8)
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs, lstmbatchsize)
model_pos, model_neg = lstmcl.train(dynamic_all[train_idx], labels_all[train_idx])
scores[8].append(lstmcl.test(model_pos, model_neg, dynamic_all[val_idx], labels_all[val_idx]))
print "===> (6) RF on dynamic (spatialized) features: %.4f (+/- %.4f) %s" % (np.mean(scores[6]), np.std(scores[6]), scores[6])
print "===> (7) HMM on dynamic features: %.4f (+/- %.4f) %s" % (np.mean(scores[7]), np.std(scores[7]), scores[7])
print "===> (8) LSTM on dynamic features: %.4f (+/- %.4f) %s" % (np.mean(scores[8]), np.std(scores[8]), scores[8])
rf.fit(trainA_static, trainA_labels) predictions_trainB_rf = rf.predict_log_proba(trainB_static) predictions_trainB_rf[predictions_trainB_rf == -inf] = np.min(predictions_trainB_rf[predictions_trainB_rf != -inf]) predictions_test_rf = rf.predict_log_proba(test_static) predictions_test_rf[predictions_test_rf == -inf] = np.min(predictions_test_rf[predictions_test_rf != -inf]) # extract predictions using HMM on dynamic hmmcl = HMMClassifier() model_pos, model_neg = hmmcl.train(nhmmstates, nhmmiter, hmmcovtype, trainA_dynamic, trainA_labels) predictions_trainB_hmm = hmmcl.predict_log_proba(model_pos, model_neg, trainB_dynamic) ratios_trainB_hmm = hmmcl.pos_neg_ratios(model_pos, model_neg, trainB_dynamic) predictions_test_hmm = hmmcl.predict_log_proba(model_pos, model_neg, test_dynamic) ratios_test_hmm = hmmcl.pos_neg_ratios(model_pos, model_neg, test_dynamic) # extract predictions using LSTM on dynamic lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs, lstmbatchsize) model_pos, model_neg = lstmcl.train(trainA_dynamic, trainA_labels) mse_pos, mse_neg = lstmcl.predict_mse(model_pos, model_neg, trainB_dynamic) predictions_trainB_lstm = np.vstack((mse_pos, mse_neg)).T ratios_trainB_lstm = lstmcl.pos_neg_ratios(model_pos, model_neg, trainB_dynamic) mse_pos, mse_neg = lstmcl.predict_mse(model_pos, model_neg, test_dynamic) predictions_test_lstm = np.vstack((mse_pos, mse_neg)).T ratios_test_lstm = lstmcl.pos_neg_ratios(model_pos, model_neg, test_dynamic) # # Prepare combined datasets for the future experiments # # datasets for ensemble learning trainB_predictions_combined_rf_hmm = np.concatenate((predictions_trainB_rf, ratios_trainB_hmm.reshape((ratios_trainB_hmm.shape[0], 1))), axis=1)
def bpic(lstmsize, lstmdropout, lstmoptim, nestimators):
lstmsize = lstmsize[0] * 10
lstmdropout = lstmdropout[0]
lstmoptim = lstmoptim[0]
lstmnepochs = 50
lstmbatchsize = 256
nestimators = nestimators[0] * 100
nfolds = 5
print lstmsize, lstmdropout, lstmoptim, nestimators
# Load the dataset
static_train = np.load(
'/storage/hpc_anna/GMiC/Data/BPIChallenge/f1/preprocessed/train_static.npy'
)
dynamic_train = np.load(
'/storage/hpc_anna/GMiC/Data/BPIChallenge/f1/preprocessed/train_dynamic.npy'
)
static_val = np.load(
'/storage/hpc_anna/GMiC/Data/BPIChallenge/f1/preprocessed/test_static.npy'
)
dynamic_val = np.load(
'/storage/hpc_anna/GMiC/Data/BPIChallenge/f1/preprocessed/test_dynamic.npy'
)
labels_train = np.load(
'/storage/hpc_anna/GMiC/Data/BPIChallenge/f1/preprocessed/train_labels.npy'
)
labels_val = np.load(
'/storage/hpc_anna/GMiC/Data/BPIChallenge/f1/preprocessed/test_labels.npy'
)
# Merge train and test
static_all = np.concatenate((static_train, static_val), axis=0)
dynamic_all = np.concatenate((dynamic_train, dynamic_val), axis=0)
labels_all = np.concatenate((labels_train, labels_val), axis=0)
nsamples = static_all.shape[0]
# prepare where to store the ratios
ratios_all_lstm = np.empty(len(labels_all))
# split indices into folds
enrich_idx_list = np.array_split(range(nsamples), nfolds)
# run CV
for fid, enrich_idx in enumerate(enrich_idx_list):
train_idx = list(set(range(nsamples)) - set(enrich_idx))
# extract predictions using LSTM on dynamic
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs,
lstmbatchsize)
model_pos, model_neg = lstmcl.train(dynamic_all[train_idx],
labels_all[train_idx])
ratios_all_lstm[enrich_idx] = lstmcl.pos_neg_ratios(
model_pos, model_neg, dynamic_all[enrich_idx])
# dataset for hybrid learning
enriched_by_lstm = np.concatenate(
(static_all, np.matrix(ratios_all_lstm).T), axis=1)
# (2.) k-fold cross validation to obtain accuracy
val_idx_list = np.array_split(range(nsamples), nfolds)
scores = []
for fid, val_idx in enumerate(val_idx_list):
train_idx = list(set(range(nsamples)) - set(val_idx))
# Hybrid on features enriched by HMM (3)
rf = RandomForestClassifier(n_estimators=nestimators)
rf.fit(enriched_by_lstm[train_idx], labels_all[train_idx])
scores.append(rf.score(enriched_by_lstm[val_idx], labels_all[val_idx]))
print 'Result: %.4f' % np.mean(scores)
return -np.mean(scores)
rf = RandomForestClassifier(n_estimators=nestimators)
rf.fit(static_all[train_idx], labels_all[train_idx])
predictions_all_rf[predict_idx] = rf.predict_log_proba(static_all[predict_idx])
predictions_all_rf[predictions_all_rf == -inf] = np.min(predictions_all_rf[predictions_all_rf != -inf])
# extract predictions using HMM on dynamic
print " Extracting predictions on dynamic data with HMM..."
hmmcl = HMMClassifier()
model_pos, model_neg = hmmcl.train(nhmmstates, nhmmiter, hmmcovtype,
dynamic_all[train_idx], labels_all[train_idx])
predictions_all_hmm[predict_idx] = hmmcl.predict_log_proba(model_pos, model_neg, dynamic_all[predict_idx])
ratios_all_hmm[predict_idx] = hmmcl.pos_neg_ratios(model_pos, model_neg, dynamic_all[predict_idx])
# extract predictions using LSTM on dynamic
print " Extracting predictions on dynamic data with LSTM..."
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs, lstmbatchsize)
model_pos, model_neg = lstmcl.train(dynamic_all[train_idx], labels_all[train_idx])
mse_pos, mse_neg = lstmcl.predict_mse(model_pos, model_neg, dynamic_all[predict_idx])
predictions_all_lstm[predict_idx] = np.vstack((mse_pos, mse_neg)).T
ratios_all_lstm[predict_idx] = lstmcl.pos_neg_ratios(model_pos, model_neg, dynamic_all[predict_idx])
#
# Prepare combined datasets for the future experiments
#
# datasets for ensemble learning
predictions_combined_rf_hmm = np.concatenate((predictions_all_rf, predictions_all_hmm), axis=1)
predictions_combined_rf_lstm = np.concatenate((predictions_all_rf, predictions_all_lstm), axis=1)
# datasets for hybrid learning
labels_train = np.load(
'/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/train_labels.npy')
labels_val = np.load(
'/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/test_labels.npy')
nsamples = dynamic_train.shape[0]
# split indices into folds
val_idx_list = np.array_split(range(nsamples), nfolds)
# run CV
scores = []
for fid, val_idx in enumerate(val_idx_list):
print "Current fold is %d/%d" % (fid + 1, nfolds)
train_idx = list(set(range(nsamples)) - set(val_idx))
# train the model and report performance
lstmcl = LSTMClassifier(lstmsize,
lstmdropout,
lstmoptim,
lstmnepochs,
lstmbatch,
validation_split=0.3)
model_pos, model_neg = lstmcl.train(dynamic_train[train_idx],
labels_train[train_idx])
scores.append(
lstmcl.test(model_pos, model_neg, dynamic_train[val_idx],
labels_train[val_idx]))
print 'Generative LSTM classifier on dynamic features: %.4f (+- %.4f) %s' % (
np.mean(scores), np.std(scores), scores)
lstmoptim = 'adadelta'
lstmnepochs = 20
lstmbatch = 64
# load the dataset
print 'Loading the dataset..'
static_train = np.load('/storage/hpc_anna/GMiC/Data/ECoGmixed/fourier/train_data.npy')
dynamic_train = np.load('/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/train_data.npy')
static_val = np.load('/storage/hpc_anna/GMiC/Data/ECoGmixed/fourier/test_data.npy')
dynamic_val = np.load('/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/test_data.npy')
labels_train = np.load('/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/train_labels.npy')
labels_val = np.load('/storage/hpc_anna/GMiC/Data/ECoGmixed/preprocessed/test_labels.npy')
nsamples = dynamic_train.shape[0]
# split indices into folds
val_idx_list = np.array_split(range(nsamples), nfolds)
# run CV
scores = []
for fid, val_idx in enumerate(val_idx_list):
print "Current fold is %d/%d" % (fid + 1, nfolds)
train_idx = list(set(range(nsamples)) - set(val_idx))
# train the model and report performance
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs, lstmbatch, validation_split=0.3)
model_pos, model_neg = lstmcl.train(dynamic_train[train_idx], labels_train[train_idx])
scores.append(lstmcl.test(model_pos, model_neg, dynamic_train[val_idx], labels_train[val_idx]))
print 'Generative LSTM classifier on dynamic features: %.4f (+- %.4f) %s' % (np.mean(scores), np.std(scores), scores)
static_val = np.load(
'/storage/hpc_anna/GMiC/Data/syn_lstm_wins/test_static.npy')
dynamic_val = np.load(
'/storage/hpc_anna/GMiC/Data/syn_lstm_wins/test_dynamic.npy')
labels_train = np.load(
'/storage/hpc_anna/GMiC/Data/syn_lstm_wins/train_labels.npy')
labels_val = np.load(
'/storage/hpc_anna/GMiC/Data/syn_lstm_wins/test_labels.npy')
# transform static features into "fake" sequences
dynamized_static_train = np.zeros(
(static_train.shape[0], static_train.shape[1], dynamic_train.shape[2]))
for i in range(static_train.shape[0]):
dynamized_static_train[i, :, :] = np.tile(static_train[i, :],
(dynamic_train.shape[2], 1)).T
dynamized_static_val = np.zeros(
(static_val.shape[0], static_val.shape[1], dynamic_val.shape[2]))
for i in range(static_val.shape[0]):
dynamized_static_val[i, :, :] = np.tile(static_val[i, :],
(dynamic_val.shape[2], 1)).T
# meld dynamized static and dynamic features together
all_train = np.concatenate((dynamized_static_train, dynamic_train), axis=1)
all_val = np.concatenate((dynamized_static_val, dynamic_val), axis=1)
# dynamic data with LSTM
lstmcl = LSTMClassifier(2000, 0.5, 'adagrad', lstm_nepochs)
model_pos, model_neg = lstmcl.train(all_train, labels_train)
print "LSTM with dynamized static and dynamic features on validation set: %.4f" % lstmcl.test(
model_pos, model_neg, all_val, labels_val)
import numpy as np
from LSTM.lstm_classifier import LSTMClassifier
# parameters
lstmsize = 512
lstmdropout = 0.0
lstmoptim = 'rmsprop'
lstmnepochs = 50
lstmbatch = 32
# load the dataset
print 'Loading the dataset..'
static_train = np.load('/storage/hpc_anna/GMiC/Data/syn_lstm_wins/train_static.npy')
dynamic_train = np.load('/storage/hpc_anna/GMiC/Data/syn_lstm_wins/train_dynamic.npy')
static_val = np.load('/storage/hpc_anna/GMiC/Data/syn_lstm_wins/test_static.npy')
dynamic_val = np.load('/storage/hpc_anna/GMiC/Data/syn_lstm_wins/test_dynamic.npy')
labels_train = np.load('/storage/hpc_anna/GMiC/Data/syn_lstm_wins/train_labels.npy')
labels_val = np.load('/storage/hpc_anna/GMiC/Data/syn_lstm_wins/test_labels.npy')
nsamples = dynamic_train.shape[0]
# split the data into training and test
train_idx = np.random.choice(range(0, nsamples), size=np.round(nsamples * 0.7, 0), replace=False)
test_idx = list(set(range(0, nsamples)) - set(train_idx))
# train the model and report performance
print 'Training the model...'
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs, lstmbatch)
model_pos, model_neg = lstmcl.train(dynamic_train[train_idx], labels_train[train_idx])
print 'Generative LSTM classifier on dynamic features: %.4f' % lstmcl.test(model_pos, model_neg, dynamic_train[test_idx], labels_train[test_idx])
labels_val = np.load(
'/storage/hpc_anna/GMiC/Data/syn_multisame/test_labels.npy')
#
# Sanity Checks
#
print "Expected performance of a lonely model is 0.75, of the joint model 1.0"
# a) static data classification
rf = RandomForestClassifier(n_estimators=100)
rf.fit(static_train, labels_train)
print "Random Forest with static features on validation set: %.4f" % rf.score(
static_val, labels_val)
# b) dynamic data classification
lstmcl = LSTMClassifier(2000, 0.5, 'adagrad', 20)
model_pos, model_neg = lstmcl.train(dynamic_train, labels_train)
print "LSTM with dynamic features on validation set: %.4f" % lstmcl.test(
model_pos, model_neg, dynamic_val, labels_val)
#
# Evaluating Joint Model
#
print ""
print "Evaluating joint model:"
print "Splitting data in two halves..."
fh_idx = np.random.choice(range(0, dynamic_train.shape[0]),
size=np.round(dynamic_train.shape[0] * 0.5, 0),
replace=False)
sh_idx = list(set(range(0, dynamic_train.shape[0])) - set(fh_idx))
fh_data = dynamic_train[fh_idx, :, :]
#
train_half = train_nsamples / 2
trainA_static = train_static[:train_half]
trainB_static = train_static[train_half:]
trainA_dynamic = train_dynamic[:train_half]
trainB_dynamic = train_dynamic[train_half:]
trainA_labels = train_labels[:train_half]
trainB_labels = train_labels[train_half:]
#
# Train enrichment models on trainA
#
print 'Training enrichment models...'
# train LSTM activations extractor
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs,
lstmbatchsize)
model_pos, model_neg = lstmcl.train(trainA_dynamic, trainA_labels)
trainB_activations_pos = lstmcl.activations(model_pos, trainB_dynamic)
trainB_activations_neg = lstmcl.activations(model_neg, trainB_dynamic)
trainB_activations = np.concatenate((trainB_activations_pos[:, seqlen - 1, :],
trainB_activations_neg[:, seqlen - 1, :]),
axis=1)
test_activations_pos = lstmcl.activations(model_pos, test_dynamic)
test_activations_neg = lstmcl.activations(model_neg, test_dynamic)
test_activations = np.concatenate((test_activations_pos[:, seqlen - 1, :],
test_activations_neg[:, seqlen - 1, :]),
axis=1)
#
# run CV
for fid, predict_idx in enumerate(predict_idx_list):
print "Enrichment fold %d / %d" % (fid + 1, nfolds)
train_idx = list(set(range(nsamples)) - set(predict_idx))
# extract predictions using HMM on dynamic
hmmcl = HMMClassifier()
model_pos, model_neg = hmmcl.train(nhmmstates, nhmmiter, hmmcovtype,
dynamic_all[train_idx],
labels_all[train_idx])
ratios_all_hmm[predict_idx] = hmmcl.pos_neg_ratios(
model_pos, model_neg, dynamic_all[predict_idx])
# extract predictions using LSTM on dynamic
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs,
lstmbatchsize)
model_pos, model_neg = lstmcl.train(dynamic_all[train_idx],
labels_all[train_idx])
mse_pos, mse_neg = lstmcl.predict_mse(model_pos, model_neg,
dynamic_all[predict_idx])
ratios_all_lstm[predict_idx] = lstmcl.pos_neg_ratios(
model_pos, model_neg, dynamic_all[predict_idx])
# datasets for hybrid learning
enriched_by_hmm = np.concatenate((static_all, np.matrix(ratios_all_hmm).T),
axis=1)
enriched_by_lstm = np.concatenate(
(static_all, np.matrix(ratios_all_lstm).T), axis=1)
# k-fold CV for performance estimation
val_idx_list = np.array_split(range(nsamples), nfolds)
predictions_all_rf[predictions_all_rf != -inf])
# extract predictions using HMM on dynamic
print " Extracting predictions on dynamic data with HMM..."
hmmcl = HMMClassifier()
model_pos, model_neg = hmmcl.train(nhmmstates, nhmmiter, hmmcovtype,
dynamic_all[train_idx],
labels_all[train_idx])
predictions_all_hmm[predict_idx] = hmmcl.predict_log_proba(
model_pos, model_neg, dynamic_all[predict_idx])
ratios_all_hmm[predict_idx] = hmmcl.pos_neg_ratios(
model_pos, model_neg, dynamic_all[predict_idx])
# extract predictions using LSTM on dynamic
print " Extracting predictions on dynamic data with LSTM..."
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs,
lstmbatchsize)
model_pos, model_neg = lstmcl.train(dynamic_all[train_idx],
labels_all[train_idx])
mse_pos, mse_neg = lstmcl.predict_mse(model_pos, model_neg,
dynamic_all[predict_idx])
predictions_all_lstm[predict_idx] = np.vstack((mse_pos, mse_neg)).T
ratios_all_lstm[predict_idx] = lstmcl.pos_neg_ratios(
model_pos, model_neg, dynamic_all[predict_idx])
#
# Prepare combined datasets for the future experiments
#
# datasets for ensemble learning
predictions_combined_rf_hmm = np.concatenate(
(predictions_all_rf, predictions_all_hmm), axis=1)
val_idx_list = np.array_split(range(nsamples), nfolds)
scores = {1: [], 2: [], 3: [], 4: [], 5: [], 6: [], 7: [], 8: [], 9: [], 10: [], 11: []}
for fid, val_idx in enumerate(val_idx_list):
print "Current fold is %d / %d" % (fid + 1, nfolds)
train_idx = list(set(range(nsamples)) - set(val_idx))
# RF on dynamic features (6)
rf = RandomForestClassifier(n_estimators=nestimators)
rf.fit(dynamic_as_static[train_idx], labels_all[train_idx])
scores[6].append(rf.score(dynamic_as_static[val_idx], labels_all[val_idx]))
# HMM on dynamic features (7)
hmmcl = HMMClassifier()
model_pos, model_neg = hmmcl.train(nhmmstates, nhmmiter, hmmcovtype, dynamic_all[train_idx], labels_all[train_idx])
acc, auc = hmmcl.test(model_pos, model_neg, dynamic_all[val_idx], labels_all[val_idx])
scores[7].append(acc)
# LSTM on dynamic features (8)
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs, lstmbatchsize)
model_pos, model_neg = lstmcl.train(dynamic_all[train_idx], labels_all[train_idx])
scores[8].append(lstmcl.test(model_pos, model_neg, dynamic_all[val_idx], labels_all[val_idx]))
print "===> (6) RF on dynamic (spatialized) features: %.4f (+/- %.4f) %s" % (
np.mean(scores[6]),
np.std(scores[6]),
scores[6],
)
print "===> (7) HMM on dynamic features: %.4f (+/- %.4f) %s" % (np.mean(scores[7]), np.std(scores[7]), scores[7])
print "===> (8) LSTM on dynamic features: %.4f (+/- %.4f) %s" % (np.mean(scores[8]), np.std(scores[8]), scores[8])
labels_all = np.concatenate((labels_train, labels_val), axis=0)
nsamples = static_all.shape[0]
# prepare where to store the ratios
activations_all = np.empty((len(labels_all), lstmsize * 2))
# split indices into folds
enrich_idx_list = np.array_split(range(nsamples), nfolds)
# CV for feature enrichment
for fid, enrich_idx in enumerate(enrich_idx_list):
print "Enrichment fold %d / %d" % (fid + 1, nfolds)
train_idx = list(set(range(nsamples)) - set(enrich_idx))
# train models for enrichment
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs, lstmbatchsize)
model_pos, model_neg = lstmcl.train(dynamic_all[train_idx], labels_all[train_idx])
# extract activations
activations_pos = lstmcl.activations(model_pos, dynamic_all[enrich_idx])
activations_neg = lstmcl.activations(model_neg, dynamic_all[enrich_idx])
activations_all[enrich_idx] = np.concatenate((activations_pos[:, -1, :], activations_neg[:, -1, :]), axis=1)
# dataset for hybrid learning
enriched_by_lstm = np.concatenate((static_all, activations_all), axis=1)
print static_all.shape
print enriched_by_lstm.shape
# CV for accuracy estimation
val_idx_list = np.array_split(range(nsamples), nfolds)
scores = []
# run CV
for fid, predict_idx in enumerate(predict_idx_list):
print "Current fold is %d" % fid
train_idx = list(set(range(nsamples)) - set(predict_idx))
# extract predictions using RF on static
print " Extracting predictions on static data with RF..."
rf = RandomForestClassifier(n_estimators=nestimators)
rf.fit(static_all[train_idx], labels_all[train_idx])
predictions_all_rf[predict_idx] = rf.predict_log_proba(static_all[predict_idx])
predictions_all_rf[predictions_all_rf == -inf] = np.min(predictions_all_rf[predictions_all_rf != -inf])
# extract predictions using LSTM on dynamic
print " Extracting predictions on dynamic data with LSTM..."
lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs, lstmbatchsize)
model_pos, model_neg = lstmcl.train(dynamic_all[train_idx], labels_all[train_idx])
mse_pos, mse_neg = lstmcl.predict_mse(model_pos, model_neg, dynamic_all[predict_idx])
predictions_all_lstm[predict_idx] = np.vstack((mse_pos, mse_neg)).T
ratios_all_lstm[predict_idx] = lstmcl.pos_neg_ratios(model_pos, model_neg, dynamic_all[predict_idx])
#
# Prepare combined datasets for the future experiments
#
# datasets for ensemble learning
predictions_combined_rf_lstm = np.concatenate((predictions_all_rf, predictions_all_lstm), axis=1)
# datasets for hybrid learning
enriched_by_lstm = np.concatenate((static_all, np.matrix(ratios_all_lstm).T), axis=1)
rf.fit(trainA_static, trainA_labels) predictions_trainB_rf = rf.predict_log_proba(trainB_static) predictions_trainB_rf[predictions_trainB_rf == -inf] = np.min(predictions_trainB_rf[predictions_trainB_rf != -inf]) predictions_test_rf = rf.predict_log_proba(test_static) predictions_test_rf[predictions_test_rf == -inf] = np.min(predictions_test_rf[predictions_test_rf != -inf]) # extract predictions using HMM on dynamic hmmcl = HMMClassifier() model_pos, model_neg = hmmcl.train(nhmmstates, nhmmiter, hmmcovtype, trainA_dynamic, trainA_labels) predictions_trainB_hmm = hmmcl.predict_log_proba(model_pos, model_neg, trainB_dynamic) ratios_trainB_hmm = hmmcl.pos_neg_ratios(model_pos, model_neg, trainB_dynamic) predictions_test_hmm = hmmcl.predict_log_proba(model_pos, model_neg, test_dynamic) ratios_test_hmm = hmmcl.pos_neg_ratios(model_pos, model_neg, test_dynamic) # extract predictions using LSTM on dynamic lstmcl = LSTMClassifier(lstmsize, lstmdropout, lstmoptim, lstmnepochs, lstmbatchsize) model_pos, model_neg = lstmcl.train(trainA_dynamic, trainA_labels) mse_pos, mse_neg = lstmcl.predict_mse(model_pos, model_neg, trainB_dynamic) predictions_trainB_lstm = np.vstack((mse_pos, mse_neg)).T ratios_trainB_lstm = lstmcl.pos_neg_ratios(model_pos, model_neg, trainB_dynamic) trainB_activations_pos = lstmcl.activations(model_pos, trainB_dynamic) trainB_activations_neg = lstmcl.activations(model_neg, trainB_dynamic) trainB_activations = np.concatenate((trainB_activations_pos[:, -1, :], trainB_activations_neg[:, -1, :]), axis=1) mse_pos, mse_neg = lstmcl.predict_mse(model_pos, model_neg, test_dynamic) predictions_test_lstm = np.vstack((mse_pos, mse_neg)).T ratios_test_lstm = lstmcl.pos_neg_ratios(model_pos, model_neg, test_dynamic) test_activations_pos = lstmcl.activations(model_pos, test_dynamic) test_activations_neg = lstmcl.activations(model_neg, test_dynamic) test_activations = np.concatenate((test_activations_pos[:, -1, :], test_activations_neg[:, -1, :]), axis=1)
