def data_splitting():
split_type = "general" #"all" or "general"
test_percent = 0.2
type_labeling = "per_group" #"per_group" or "per_spp"
raw_data_file = os.path.join(results_dir, "counts_per_animal.pkl")
counts_data = rws.loadData(raw_data_file)
splitted = splitting.split_data(counts_data, split_type, test_percent,
type_labeling)
splitted_data_file = os.path.join(
partitions_dir, "split_{}_{}_{}.pkl".format(split_type,
str(test_percent),
type_labeling))
rws.saveData(splitted, splitted_data_file)
def train_model(type_crossval="k-fold", type_classif="qda"):
""" Function to train a classifier depending on:
type_crossval: Type of cross validation iterator
- "k-fold": make just homogeneus k-fold
- "shuffleSplit": Add a shuffle representation of k-fold
- "groups_k-fold": Generates cross-validation depending on some groups
- "L_P_Groups_out": Leave P groups Out
tyoe_classif: Type of classifier use to train the model
- "qda"
- "rf
- "svc" """
split_type="general" #"all" or "general"
test_percent=0.2
type_labeling="per_group" #"per_group" or "per_spp"
splitted_data_file = os.path.join(partitions_dir,
"split_{}_{}_{}.pkl".format(split_type,str(test_percent),type_labeling))
train_dict, test_dict = rws.loadData(splitted_data_file)
train_data = train_dict['data']
train_labels = train_dict['labels']
cross_val_test_size=0.2
if type_classif == 'qda':
kwargs = {}
elif type_classif == 'rf':
kwargs = {"n_trees" : 100, "boots" : True}
elif type_classif == 'svc':
kwargs = {"c" : 1.0, "kernel_type" : 'linear', "gamma_value" : 'scale'}
else:
raise ValueError("Select an aproppriate type of classifier")
if split_type == 'all':
num_splits = 5
if type_crossval == 'k-fold':
train_results = cv.k_fold_iter(train_data, train_labels, num_ss=num_splits,
clf=type_classif, **kwargs)
elif type_crossval == 'shuffleSplit':
test_size = cross_val_test_size
train_results = cv.shuffle_split_iter(train_data, train_labels, num_ss=5,
t_size=test_size, clf=type_classif, **kwargs)
else:
raise ValueError("If split_type: 'all'," +
" you have to choose type_crossval: k-fold or shuffleSplit")
if split_type == "general":
s_labels = train_dict['s_labs']
train_grp_animals = train_dict['train_grps']
n_groups = round(((1-test_percent) * 10)/2)
if type_crossval == 'groups_k-fold':
train_results = cv.groups_k_fold_iter(train_data, train_labels, s_labels,
train_grp_animals, num_groups=n_groups, clf=type_classif, **kwargs)
elif type_crossval == 'L_P_Groups_out':
n_groups = 2
train_results = leave_P_out_iter(train_data, train_labels, s_labels, train_grp_animals,
num_groups=n_groups, clf=type_classif, **kwargs)
else:
raise ValueError("If split_type: 'general'," +
" you have to choose type_crossval: groups_k-fold or L_P_Groups_out")
print(len(train_results))
print(train_results[0][1].shape)
print(train_results[0][2].shape)
print(train_results[0][2])
if len(train_results[0]) == 5:
print(train_results[0][3].shape)
print(train_results[0][3])
print(train_results[0][4])
else:
print(train_results[0][3])
acc_values = np.zeros(len(train_results))
for i in range(len(train_results)):
if len(train_results[0]) == 5:
acc_values[i] = train_results[i][4]
else:
acc_values[i] = train_results[i][3]
print(np.mean(acc_values))
get_metrics(train_results, type_crossval, type_classif, **kwargs)