def __init__(self,
init_dat,
train_dat,
multi_t=False,
random=True,
do_active=False,
rand_task=False):
self.trained_x = init_dat['feature']
self.trained_y = init_dat['label']
self.train_pool = gen_land_pool(train_dat, multi_t)
self.pool_size = util.dat_size(train_dat)
self.total_pool = self.pool_size
self.init_size = util.dat_size(init_dat)
self.multi_task = multi_t
self.rand_task = rand_task
# Keep track of tasks that have training instances
# In matlab format where all tasks start with one
self.task_online = [t for t in range(0, T)]
if random:
if multi_t:
for task_pool in self.train_pool:
np.random.shuffle(task_pool)
else:
np.random.shuffle(self.train_pool)
self.do_active = do_active
def run_stl_landm(pool_dat, init_dat, test_dat, do_active):
###### Train Initial Model ######
init_size = dat_size(init_dat)
learner = MLLearner(LogisticRegression(), init_dat)
print "Start training..."
print "pool", len(train_pool)
total_pool_size = len(train_pool)
test_acc = []
learned_size = []
prof = Professor(init_dat, pool_dat, random=True, do_active=do_active)
total_pool_size = prof.get_pool_size()
init_acc = util.model_roc_score(learner, test_dat)
test_acc = [init_acc]
learned_size = [init_size]
### Training Until No more data available OR Reach the set N_ITER ###
count = N_ITER
while prof.has_next():
# print "pool", len(train_pool)
selected_x, selected_y, tasks = prof.next_train_set(INS_SIZE,
learner=learner)
# Group selected training set by tasks
next_train_x = [[] for i in range(0, T)]
next_train_y = [[] for i in range(0, T)]
for i in range(0, selected_x.shape[0]):
t = int(tasks[i])
next_train_x[t].append(selected_x[i, :])
next_train_y[t].append(selected_y[i])
for t in range(0, T):
if next_train_x[t]:
learner.refine_model(np.array(next_train_x[t]),
np.array(next_train_y[t]), t)
# acc = util.model_roc_score(learner, test_dat)
acc = util.model_score(learner, test_dat)
test_acc.append(acc)
learned_size.append(total_pool_size - prof.get_pool_size() + init_size)
if ITER_ENABLE:
if count < 0: break
count -= 1
return test_acc, learned_size
from data_process import gen_init
import numpy as np
import pickle as pk
from config import T
from util import dat_size
import util
source = 'fera_forella'
## Load From Source ##
train_dat, test_dat = load_dat(source)
print "init dat"
init_dat, pool_dat = gen_init(train_dat, 5)
print "train size", dat_size(train_dat)
print "test size", dat_size(test_dat)
print "pool size", dat_size(pool_dat)
print "init size", dat_size(init_dat)
print "----> Writing to files..."
# Save to files
pool_f = open("data/pool", "wb")
pk.dump(pool_dat, pool_f)
pool_f.close()
test_f = open("data/test", "wb")
pk.dump(test_dat, test_f)
test_f.close()
from util import load_test from util import load_init from util import load_pool ######## Panda ###### # Comparing Multiple Active Learner vs ELLA + ATS vs ELLA + ATS + AL vs # ELLA + AL # This file runs Pure Multiple Active Learner ##################### ###### Load Data ###### pool_dat = load_pool() init_dat = load_init() test_dat = load_test() init_size = dat_size(init_dat) train_pool = np.array(gen_land_pool(pool_dat)) shuffle(train_pool) ###### Train Initial Model ###### learner = MLLearner(LogisticRegression(), init_dat) print "Start training..." print "pool", len(train_pool) total_pool_size = len(train_pool) test_acc = [] learned_size = []
eng.addpath(eng.genpath(ELLA_DIR)) # res = eng.runExperimentActiveTask() # print res ######## Panda ###### # Comparing Multiple Active Learner vs ELLA + ATS vs ELLA + ATS + AL vs # ELLA + AL # This file runs ELLA + Active Task Selection ##################### ## Load all files test_dat = util.add_bias(load_test()) pool_dat = util.add_bias(load_pool()) init_dat = util.add_bias(load_init()) init_size = util.dat_size(init_dat) ## Init ELLA Model with init set ## ella_model = ELLA(eng, init_dat) init_acc = ella_score(ella_model, test_dat) test_acc = [init_acc] learned_size = [init_size] prof = Professor(init_dat, pool_dat, multi_t=True, random=True) total_pool_size = prof.get_pool_size() print "train pool size", total_pool_size # ### Training Until No more data available OR Reach the set N_ITER ### count = N_ITER while prof.has_next():
landmine = loadmat('data/LandmineData.mat')
#### Data Summary ####
# counts = np.zeros(2)
# for t in range(0, T):
# counts += np.bincount(landmine['label'][:, t][0].reshape(landmine['label'][:, t][0].shape[0]))
# print counts
# exit()
print "----> Loading & Splitting data..."
land_train, land_test = load_landset()
print "init dat"
init_dat, pool_dat = gen_init(land_train, 5)
print "train size", dat_size(land_train)
print "test size", dat_size(land_test)
print "pool size", dat_size(pool_dat)
print "init size", dat_size(init_dat)
print "----> Writing to files..."
# Save to files
pool_f = open("data/pool", "wb")
pk.dump(pool_dat, pool_f)
pool_f.close()
test_f = open("data/test", "wb")
pk.dump(land_test, test_f)
test_f.close()
def run_ml_landmine(pool_dat,
init_dat,
test_dat,
learner_class,
engine=None,
do_active=False,
task_ctrl=TASK_NONE,
evaluation=config.EVAL_ACC):
if task_ctrl == TASK_NONE:
active_task = False
rand_task = False
elif task_ctrl == TASK_RAND:
active_task = False
rand_task = True
elif task_ctrl == TASK_ACTIVE:
active_task = True
rand_task = False
# Copy all data
test_dat_cp = copy.deepcopy(test_dat)
init_dat_cp = copy.deepcopy(init_dat)
pool_dat_cp = copy.deepcopy(pool_dat)
###### Train Initial Model ######
init_size = dat_size(init_dat_cp)
if learner_class is ELLA:
# Add bias term
test_dat_cp = util.add_bias(test_dat_cp)
pool_dat_cp = util.add_bias(pool_dat_cp)
init_dat_cp = util.add_bias(init_dat_cp)
learner = learner_class(engine, init_dat_cp)
else:
learner = learner_class(LogisticRegression(), init_dat_cp)
print "Start training..."
test_acc = []
learned_size = []
prof = Professor(init_dat_cp,
pool_dat_cp,
random=True,
do_active=do_active,
multi_t=active_task or rand_task,
rand_task=rand_task)
total_pool_size = prof.get_pool_size()
init_acc = model_eval(learner, test_dat_cp, evaluation)
test_acc = [init_acc]
learned_size = [init_size]
task_rec = []
## Training Until No more data available OR Reach the set N_ITER ###
count = N_ITER
while prof.has_next():
# print "pool", prof.get_pool_size()
selected_x, selected_y, tasks = prof.next_train_set(INS_SIZE,
learner=learner)
task_rec.append(tasks[0])
# Group selected training set by tasks
next_train_x = [[] for i in range(0, T)]
next_train_y = [[] for i in range(0, T)]
for i in range(0, selected_x.shape[0]):
t = int(tasks[i])
next_train_x[t].append(selected_x[i, :])
next_train_y[t].append(selected_y[i])
for t in range(0, T):
if next_train_x[t]:
learner.refine_model(np.array(next_train_x[t]),
np.array(next_train_y[t]), t)
acc = model_eval(learner, test_dat_cp, evaluation)
# print acc
test_acc.append(acc)
# print learner.get_trained_size()
learned_size.append(total_pool_size - prof.get_pool_size() + init_size)
if ITER_ENABLE:
if count < 0: break
count -= 1
if count % 20 == 0:
if engine is not None:
engine.clean_mem()
util.curve_to_csv("res/task_save" + str(do_active) + " .csv", task_rec,
[0])
return test_acc, learned_size
