def train():
from judgments import judgments_from_file, judgments_by_qid
es = elastic_connection(timeout=1000)
# Load features into Elasticsearch
init_default_store()
load_features(FEATURE_SET_NAME)
# Parse a judgments
movieJudgments = judgments_by_qid(
judgments_from_file(filename=JUDGMENTS_FILE))
# Use proposed Elasticsearch queries (1.json.jinja ... N.json.jinja) to generate a training set
# output as "sample_judgments_wfeatures.txt"
log_features(es, judgments_dict=movieJudgments, search_index=INDEX_NAME)
build_features_judgments_file(movieJudgments,
filename=JUDGMENTS_FILE_FEATURES)
# Train each ranklib model type
for modelType in [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]:
# 0, MART
# 1, RankNet
# 2, RankBoost
# 3, AdaRank
# 4, coord Ascent
# 6, LambdaMART
# 7, ListNET
# 8, Random Forests
# 9, Linear Regression
Logger.logger.info("*** Training %s " % modelType)
train_model(judgments_with_features_file=JUDGMENTS_FILE_FEATURES,
model_output='model.txt',
which_model=modelType)
save_model(script_name="gsearch_model_%s" % modelType,
feature_set=FEATURE_SET_NAME,
model_fname='model.txt')
def gogo_kpca(fxpath, mpath):
kpca_params = {
'n_components': 256,
'kernel': 'rbf',
'gamma': None,
'degree': 3,
'coef0': 1,
'kernel_params': None,
'alpha': 1.0,
'fit_inverse_transform': False,
'eigen_solver': 'auto',
'tol': 0,
'max_iter': None,
'remove_zero_eig': True
}
kpca_fname = '%s/kpca_rbf_{0}_{1}.pkl' % mpath
for i in range(7):
if i < 5:
nbreed = 1
sbreed = 'dog'
nsubject = i + 1
else:
nbreed = 2
sbreed = 'human'
nsubject = 1 + abs(5 - i)
print('breed%d.subject%d..' % (nbreed, nsubject))
X_ictal = load_features(fxpath, nbreed, nsubject, 1)
X_inter = load_features(fxpath, nbreed, nsubject, 2)
X = vstack((X_inter, X_ictal))
del X_inter, X_ictal
gc.collect()
X_test = load_features(fxpath, nbreed, nsubject, 3)
X = vstack((X, X_test))
del X_test
gc.collect()
kpca = KernelPCA(**kpca_params)
skip_interval = get_skip_interval(X)
X = kpca_preprocess_features(X)
kpca.fit(X[::skip_interval])
with open(kpca_fname.format(sbreed, nsubject), 'wb') as f:
pickle.dump(kpca, f)
del X, kpca
gc.collect()
def gogo_kpca( fxpath, mpath ):
kpca_params = {'n_components':256,
'kernel':'rbf',
'gamma':None,
'degree':3,
'coef0':1,
'kernel_params':None,
'alpha':1.0,
'fit_inverse_transform':False,
'eigen_solver':'auto',
'tol':0,
'max_iter':None,
'remove_zero_eig':True}
kpca_fname = '%s/kpca_rbf_{0}_{1}.pkl' % mpath
for i in range(7):
if i < 5:
nbreed = 1
sbreed = 'dog'
nsubject = i+1
else:
nbreed = 2
sbreed = 'human'
nsubject = 1 + abs(5-i)
print 'breed%d.subject%d..' % ( nbreed, nsubject )
X_ictal = load_features( fxpath, nbreed, nsubject, 1 )
X_inter = load_features( fxpath, nbreed, nsubject, 2 )
X = vstack((X_inter, X_ictal))
del X_inter, X_ictal; gc.collect()
X_test = load_features( fxpath, nbreed, nsubject, 3 )
X = vstack((X, X_test))
del X_test; gc.collect()
kpca = KernelPCA(**kpca_params)
skip_interval = get_skip_interval(X)
X = kpca_preprocess_features(X)
kpca.fit(X[::skip_interval])
with open(kpca_fname.format(sbreed,nsubject),'wb') as f:
cPickle.dump(kpca,f)
del X, kpca; gc.collect()
def main():
data = load_data()
features_cnt,label_features = load_features.load_features(data["inputFile"])
#for label in label_features:
# label_features[label] = random.shuffle(label_features[label])
training_data = {}
for label in label_features:
training_data.update({label:label_features[label][:data["traininSamplesCnt"]]})
testing_data = {}
for label in label_features:
testing_data.update({label:label_features[label][data["traininSamplesCnt"]:]})
network = NeuralNetwork(features_cnt,
data["hiddenLayers"]["cnt"],
data["hiddenLayers"]["layersNeuronsCnt"],
data["hiddenLayers"]["layersActivFns"],
data["outputLayer"]["neorunsCnt"],
data["outputLayer"]["activFn"],
data["withBias"])
model,loss_curve = network.train(training_data,
data["eta"],
data["epochsNo"],
data["stopMSE"],
data["MSE"])
draw_result.draw_training(loss_curve)
x = network.test(testing_data,data["outputLayer"]["neorunsCnt"])*100
print "Accuracy: %f%%"%x
def training_pipeline():
from utils import elastic_connection
es = elastic_connection()
file_judgments = parse_data_and_get_judgement()
print(file_judgments)
init_default_store()
load_features(FEATURE_SET_NAME)
log_features(es, judgments_dict=file_judgments, search_index=INDEX_NAME)
build_features_judgments_file(file_judgments,
filename=JUDGMENTS_FILE_FEATURES)
for modelType in [6, 7, 9]:
# for modelType in [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]:
Logger.logger.info("*** Training %s " % modelType)
train_model(judgments_with_features_file=JUDGMENTS_FILE_FEATURES,
model_output='model.txt',
which_model=modelType)
save_model(script_name="test_%s" % modelType,
feature_set=FEATURE_SET_NAME,
model_fname='model.txt')
def upload_model():
init_default_store()
load_features(FEATURE_SET_NAME)
save_model(script_name="test_9",
feature_set=FEATURE_SET_NAME,
model_fname='model.txt')
data=json.dumps(model_payload),
headers=head,
auth=ES_AUTH,
verify=False)
Logger.logger.info(resp.status_code)
if resp.status_code >= 300:
Logger.logger.error(resp.text)
if __name__ == "__main__":
from judgments import judgments_from_file, judgments_by_qid
es = elastic_connection(timeout=1000)
# Load features into Elasticsearch
init_default_store()
load_features(FEATURE_SET_NAME)
# Parse a judgments
movieJudgments = judgments_by_qid(
judgments_from_file(filename=JUDGMENTS_FILE))
# Use proposed Elasticsearch queries (1.json.jinja ... N.json.jinja) to generate a training set
# output as "sample_judgments_wfeatures.txt"
log_features(es, judgments_dict=movieJudgments, search_index=INDEX_NAME)
build_features_judgments_file(movieJudgments,
filename=JUDGMENTS_FILE_FEATURES)
# Train each ranklib model type
#for modelType in [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]:
for modelType in [6]:
# 0, MART
# 1, RankNet
# 2, RankBoost
# 3, AdaRank
# to change when small data: n_train and n_test in utils.py, n_components in fisher_feature_extractor.py
folder_name = 'data/'
# folder_name = 'data_small/'
nclasses = 10
classifier = 'svm_ovo'
do_validation = True
validation = 0.2
do_prediction = False
svm_kernel = LinearKernel()
#svm_kernel = LaplacianRBFKernel(1.6)
C = 1
Xtrain, Ytrain, Xtest = load_features(feature_extractor, overwrite_features,
overwrite_kpca, kernel_pca,
kernel_pca_kernel, cut_percentage,
folder_name)
#Xtrain = numpy.reshape(Xtrain, (Xtrain.shape[0], -1))
#Xtest = numpy.reshape(Xtest, (Xtest.shape[0], -1))
print(Xtrain.shape)
print(Xtest.shape)
assert Xtrain.ndim == 2 and Xtrain.shape[1] == Xtest.shape[1]
print("Fitting on training data")
if classifier == 'cross_entropy':
Xtrain = concat_bias(Xtrain)
Xtest = concat_bias(Xtest)
model = CrossEntropyClassifier(nclasses)
iterations = 500
lr = 0.01
def gogo_bagged_svm( fxpath, mpath, spath ):
transform = True
svc_params = {'penalty':'l2',
'loss':'l2',
'dual':False,
'C':33.0,
'intercept_scaling':1e4,
'class_weight':'auto',
'random_state':42}
bc_params = {'base_estimator':LinearSVC(**svc_params),
'n_estimators':96,
'max_samples':0.1,
'max_features':0.8,
'oob_score':False,
# if you have tons of memory (i.e. 32gb ram + 32gb swap)
# incresaing this parameter may help performance. else,
# increasing it may cause "out of memory" errors.
'n_jobs':1,
#'n_jobs':8,
'verbose':1,
'random_state':42}
'''
lr_params = {'C':1e6,#tr(-3,3,7),
'penalty':'l2',
'class_weight':'auto',
'intercept_scaling':1e6}#tr(-1,6,7)}
'''
preds = []
kpca_fname = '%s/kpca_rbf_{0}_{1}.pkl' % mpath
s_fname = '%s/kpca_linear_svm{0}_{1}_preds.csv' % spath
for i in range(7):
if i < 5:
nbreed = 1
sbreed = 'dog'
nsubject = i+1
else:
nbreed = 2
sbreed = 'human'
nsubject = 1 + abs(5-i)
print('breed%d.subject%d..' % ( nbreed, nsubject ))
X_ictal = load_features( fxpath, nbreed, nsubject, 1 )
X_inter = load_features( fxpath, nbreed, nsubject, 2 )
X_train = vstack((X_inter, X_ictal))
Y = [0 for x in X_inter] + [1 for x in X_ictal]
wi = 1.0/len(X_inter) * 1000
wp = 1.0/len(X_ictal) * 1000
W = array([wp if y else wi for y in Y])
del X_inter, X_ictal; gc.collect()
with open(kpca_fname.format(sbreed,nsubject),'rb') as f:
kpca = pickle.load(f)
if transform:
X_train = kpca_preprocess_features(X_train)
X_train = kpca_incremental_transform(kpca,X_train)
gc.collect()
X_test = load_features( fxpath, nbreed, nsubject, 3 )
if transform:
X_test = kpca_preprocess_features(X_test)
X_test = kpca_incremental_transform(kpca,X_test)
gc.collect()
bc = BC(**bc_params)
bc.fit(X_train,Y)
#print 'oob_score: ', bc.oob_score_
subject_preds = bc.predict_proba(X_test)[:,1]
preds.append(subject_preds)
subject_preds = pd.DataFrame(subject_preds)
subject_preds.to_csv(s_fname.format(sbreed,nsubject),index=False,header=None)
del X_train, X_test; gc.collect()
sys.stdout.flush()
def gogo_bagged_svm(fxpath, mpath, spath):
transform = True
svc_params = {
"penalty": "l2",
"loss": "l2",
"dual": False,
"C": 33.0,
"intercept_scaling": 1e4,
"class_weight": "auto",
"random_state": 42,
}
bc_params = {
"base_estimator": LinearSVC(**svc_params),
"n_estimators": 96,
"max_samples": 0.1,
"max_features": 0.8,
"oob_score": False,
# if you have tons of memory (i.e. 32gb ram + 32gb swap)
# incresaing this parameter may help performance. else,
# increasing it may cause "out of memory" errors.
"n_jobs": 1,
#'n_jobs':8,
"verbose": 1,
"random_state": 42,
}
"""
lr_params = {'C':1e6,#tr(-3,3,7),
'penalty':'l2',
'class_weight':'auto',
'intercept_scaling':1e6}#tr(-1,6,7)}
"""
preds = []
kpca_fname = "%s/kpca_rbf_{0}_{1}.pkl" % mpath
s_fname = "%s/kpca_linear_svm{0}_{1}_preds.csv" % spath
for i in range(7):
if i < 5:
nbreed = 1
sbreed = "dog"
nsubject = i + 1
else:
nbreed = 2
sbreed = "human"
nsubject = 1 + abs(5 - i)
print "breed%d.subject%d.." % (nbreed, nsubject)
X_ictal = load_features(fxpath, nbreed, nsubject, 1)
X_inter = load_features(fxpath, nbreed, nsubject, 2)
X_train = vstack((X_inter, X_ictal))
Y = [0 for x in X_inter] + [1 for x in X_ictal]
wi = 1.0 / len(X_inter) * 1000
wp = 1.0 / len(X_ictal) * 1000
W = array([wp if y else wi for y in Y])
del X_inter, X_ictal
gc.collect()
with open(kpca_fname.format(sbreed, nsubject), "rb") as f:
kpca = cPickle.load(f)
if transform:
X_train = kpca_preprocess_features(X_train)
X_train = kpca_incremental_transform(kpca, X_train)
gc.collect()
X_test = load_features(fxpath, nbreed, nsubject, 3)
if transform:
X_test = kpca_preprocess_features(X_test)
X_test = kpca_incremental_transform(kpca, X_test)
gc.collect()
bc = BC(**bc_params)
bc.fit(X_train, Y)
# print 'oob_score: ', bc.oob_score_
subject_preds = bc.predict_proba(X_test)[:, 1]
preds.append(subject_preds)
subject_preds = pd.DataFrame(subject_preds)
subject_preds.to_csv(s_fname.format(sbreed, nsubject), index=False, header=None)
del X_train, X_test
gc.collect()
sys.stdout.flush()
