TEST_FILE = MID_FILE + '.test'
try:
# shutil.rmtree(SWEEP_DIR)
os.mkdir(SWEEP_DIR)
except:
pass
# build the feature vector ONCE if necessary
os.system('./%s %s' % (BUILD_PROG, TRAIN_FILE))
CSteps = [0.25, 0.5, 1, 2, 4]
GSteps = [0.25, 0.5, 1, 2, 4, 8, 16, 32, 64, 128]
# CSteps = [0.25, 0.5, 1, 2, 4]
# GSteps = [0.25, 0.5, 1, 2, 4]
for C in CSteps:
for G in GSteps:
print '\n\n** tuning paramers {C=%d, gamma=%d} **\n\n' % (C, G)
os.system('./%s %s %s' % (TRAIN_PROG, str(C), str(G)))
os.system('./%s %s' % (TEST_PROG, TEST_FILE))
# copy training & testing results
# pdb.set_trace()
copy_anything(RESULTS_DIR, os.path.join(SWEEP_DIR, '%s.C%s.G%s' % (RESULTS_DIR, str(C), str(G))))
# cleanup
os.system('./clean_test_svm.sh')
movie_id = f_mlist.readline().strip()
idx = 1
while (movie_id):
if (idx in test_idx):
f_test.write('%s\n' % movie_id)
else:
f_train.write('%s\n' % movie_id)
movie_id = f_mlist.readline().strip()
idx += 1
f_train.close()
f_test.close()
# now train/test on this partition
if model_type == 'svm':
BUILD_PROG = 'build_svm_imdb.py'
os.system('./%s %s' % (BUILD_PROG, train_file))
os.system('./%s 2 2' % TRAIN_PROG)
else:
os.system('./%s %s' % (TRAIN_PROG, train_file))
os.system('./%s %s' % (TEST_PROG, test_file))
# copy training & testing results
copy_anything(MODEL_DIR, os.path.join(XVAL_DIR, '%s.K%s' % (MODEL_DIR, k)))
copy_anything(RESULTS_DIR, os.path.join(XVAL_DIR, '%s.K%s' % (RESULTS_DIR, k)))
# clean training & testing results
call(['./clean_train_%s.sh' % model_type])
call(['./clean_test_%s.sh' % model_type])
