def compute_n_crystal_found(filename, test_range=range(0, 101, 10)):
X, y = read_data(filename)
scores = []
for t in test_range:
ind = N_INIT_POINTS - 1 + t
scores.append(np.sum(y[N_INIT_POINTS:ind] == CRYSTAL_CLASS))
return test_range, scores
def compute_volume_convex_hull(filename, test_range=range(0, 101, 10)):
X, y = read_data(filename)
scores = []
for t in test_range:
ind = N_INIT_POINTS - 1 + t
X_train = X[0:ind]
y_train = y[0:ind]
points = X_train[y_train == CRYSTAL_CLASS, :]
hull = ConvexHull(points)
scores.append(hull.volume)
return test_range, scores
def compute_volume_neighbors(filename, radius, test_range=range(0, 101, 10)):
X, y = read_data(filename)
scores = []
for t in test_range:
ind = N_INIT_POINTS - 1 + t
X_train = X[0:ind]
y_train = y[0:ind]
points = X_train[y_train == CRYSTAL_CLASS, :]
tree = BallTree(points)
score, _ = average_n_neighbors(tree, points, radius)
scores.append(score)
return test_range, scores
if len(sys.argv) != 2:
print 'Please specify a root folder as argument'
# seed
seed = int(sys.argv[1])
random.seed(seed)
np.random.seed(seed)
#
root_folder = os.path.join(HERE_PATH, sys.argv[1])
filetools.ensure_dir(root_folder)
# load data
current_datafile = os.path.join(HERE_PATH, 'init_data.csv')
X, y = read_data(current_datafile)
# check everything is fine
np.testing.assert_array_almost_equal(np.sum(X, axis=1),
TOTAL_VOLUME_IN_ML,
decimal=N_DECIMAL_EQUAL)
#
X_selected = np.random.rand(N_GENERATED, X.shape[1])
X_selected = proba_normalize_row(X_selected)
X_selected = TOTAL_VOLUME_IN_ML * X_selected
# save new csv
# xout
X_out = np.vstack((X, X_selected))
def get_data_xp(xp_key):
return read_data(FILENAMES[xp_key])
def get_new_data(filename):
X, y = read_data(filename)
return X[N_INIT_POINTS:, :], y[N_INIT_POINTS:]
def get_init_data(filename):
X, y = read_data(filename)
return X[:N_INIT_POINTS, :], y[:N_INIT_POINTS]
def get_min_sample_per_class(y):
min_sample_per_class = np.inf
for class_number in np.unique(y):
n_sample = np.sum(y == class_number)
if n_sample < min_sample_per_class:
min_sample_per_class = n_sample
return min_sample_per_class
def compute_learning_curve(filename, (X_test, y_test),
blank_clf,
test_range=range(0, 101, 10)):
X, y = read_data(filename)
scores = []
confusions = []
for t in test_range:
ind = N_INIT_POINTS - 1 + t
X_train = X[0:ind]
y_train = y[0:ind]
clf = clone(blank_clf)
clf.fit(X_train, y_train)
prediction_accuracy = clf.score(X_test, y_test)
scores.append(prediction_accuracy)
