logger.info(f'Splitting data set:')
train_frac, val_frac, test_frac = args.train_val_test_frac
X_train, X_val_test, y_train, y_val_test = model_selection.train_test_split(
samples,
encoded_labels,
test_size=val_frac + test_frac,
random_state=RANDOM_SEED,
shuffle=True)
val_split = int(len(X_val_test) * val_frac / (val_frac + test_frac))
X_val, y_val = X_val_test[:val_split], y_val_test[:val_split]
X_test, y_test = X_val_test[val_split:], y_val_test[val_split:]
logger.info(f'...training samples: {len(X_train)}')
logger.info(f'...validation samples: {len(X_val)}')
logger.info(f'...test samples: {len(X_test)}')
proj_mask = common.ProjMask(*args.proj_mask)
logger.info(f'Projection mask: {proj_mask}')
logger.info(f'Augment epochs: {args.epochs}')
logger.info(f'Online learning: {args.online_learn}')
if not args.use_svc:
logger.info('Using SVM algo: SGDClassifier.')
clf = sgd_fit(train=(X_train, y_train),
test=(X_test, y_test),
proj_mask=args.proj_mask,
online_learn=args.online_learn,
svm_model=args.svm_model,
epochs=args.epochs)
else:
logger.info('Using SVM algo: SVC.')
clf = svc_fit(train=(X_train, y_train),
from sys import maxsize as SYS_MAXSIZE
#np.set_printoptions(threshold=SYS_MAXSIZE)
# Output SVM confusion matrix name.
SVM_CM = 'train-results/svm_cm.png'
# Output XGBoost confusion matrix name.
XGB_CM = 'train-results/xgb_cm.png'
# Define a seed so random operations are the same from run to run.
RANDOM_SEED = 1234
# Define number of folds for the Stratified K-Folds cross-validator.
FOLDS = 5
# Number of parameters to combine for xgb random search.
PARA_COMB = 20
# Radar 2-D projections to use for predictions.
PROJ_MASK = common.ProjMask(xy=True, xz=True, yz=True)
def evaluate_model(model, X_test, y_test, target_names, cm_name):
""" Generate model confusion matrix and classification report. """
print('\n Evaluating model.')
y_pred = model.predict(X_test)
cm = confusion_matrix(y_test, y_pred)
print(f'\n Confusion matrix:\n{cm}')
cm_figure = plot_confusion_matrix(cm, class_names=target_names)
cm_figure.savefig(path.join(common.PRJ_DIR, cm_name))
cm_figure.clf()
print('\n Classification matrix:')
print(classification_report(y_test, y_pred, target_names=target_names))
return
from sklearn import svm, linear_model
logger = logging.getLogger(__name__)
# Radar detection threshold.
RADAR_THRESHOLD = 5
# Set to True if using Moving Target Identification (MTI) filter.
MTI = True
# Radar scan arena in polar coords.
# This can be different than arena used for training.
R_MIN, R_MAX, R_RES = 10, 360, 2
THETA_MIN, THETA_MAX, THETA_RES = -42, 42, 4
PHI_MIN, PHI_MAX, PHI_RES = -30, 30, 2
# Radar 2-D projections to use for predictions.
PROJ_MASK = common.ProjMask(xy=True, xz=True, yz=True)
LOG_FILE = 'predict.log'
def calc_proj_zoom(train_size_x, train_size_y, train_size_z, size_x, size_y,
size_z):
""" Calculate projection zoom factors for prediction radar arena.
Args:
train_size_{x,y,z} (int): Size of image array used for training.
size_{x,y,z} (int): Size of sample image array.
Returns:
ProjZoom (tuple of list of floats): Zoom factors per projection.
"""