"loss": {
"name": loss,
"alpha": alpha
},
"max_depth": max_depth,
"n_estimators": n_estimators
}
}
# Part 1 - Data Preprocessing
# Importing the dataset
dataset = pd.read_csv('train.csv')
# feature selection
if feature_selection == "infogain":
categorical_features = get_cached_features(parameters["feature_selection"])
continuous_values = []
categorical_features_count = len(categorical_features)
selected_features = categorical_features + continuous_values
X = dataset.iloc[:, selected_features].values
y = dataset.iloc[:, 1].values
column_ranges = []
print("replacing missing values and encode categorical features")
t0 = time.time()
print("number of examples: " + str(len(X[:, 0])))
for i in range(len(X[0, :])):
if i <= categorical_features_count:
def preproc(dataset,
mode,
oneHot,
scale=False,
scaler=None,
feature_selection=False):
# categorical and binary features
categorical_features = [
3, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 17, 18, 19, 23, 24, 25, 26, 27,
28, 29, 30, 31, 32, 33, 53, 54, 55, 56, 57, 58
]
# continuous values
continuous_values = [
2, 4, 15, 16, 20, 21, 22, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44,
45, 46, 47, 48, 49, 50, 51, 52
]
if feature_selection:
feat = "infogain"
else:
feat = "none"
number_of_features = 10
alpha = 32
max_depth = 4
n_estimators = 100
loss = "rank:pairwise"
parameters = {
"feature_selection": {
"name": feat,
"number_of_features": number_of_features
},
"classifier": {
"name": "xgboost",
"loss": {
"name": loss,
"alpha": alpha
},
"max_depth": max_depth,
"n_estimators": n_estimators
}
}
if feat == "infogain":
categorical_features = get_cached_features(
parameters["feature_selection"])
continuous_values = []
categorical_features_count = len(categorical_features)
if not feature_selection:
if mode == 'train':
selected_features = categorical_features + continuous_values
elif mode == 'test':
selected_features = categorical_features + continuous_values
selected_features = np.array(selected_features) - 1
else:
Warning(
"Mode must be train or set, otherwise it will lead to local"
" a variable referenced before assignment error")
else:
if mode == 'train':
selected_features = categorical_features
elif mode == 'test':
selected_features = categorical_features
selected_features = np.array(selected_features) - 1
else:
Warning(
"Mode must be train or set, otherwise it will lead to local"
" a variable referenced before assignment error")
X = dataset.iloc[:, selected_features].values
y = dataset.iloc[:, 1].values
column_ranges = []
print("replacing missing values")
t0 = time.time()
print("number of examples: " + str(len(X[:, 0])))
for i in range(len(X[0, :])):
if i <= categorical_features_count:
# si c'est une variable de catégories, on prend comme stratégie de remplacer par la
# valeur la plus fréquente
(values, counts) = np.unique(X[:, i], return_counts=True)
counts = [
counts[i] if values[i] >= 0 else 0 for i in range(len(values))
]
ind = np.argmax(counts)
column_ranges.append(max(values))
replacement_value = values[ind]
else:
# sinon on prend simplement la moyenne
replacement_value = np.mean(X[:, i])
for j in range(len(X[:, i])):
if X[j, i] < -0.5:
X[j, i] = replacement_value
if oneHot:
print("One hot encoding")
# NB: les nouvelles colonnes sont placées juste devant l'autre colonne
onehotencoder = OneHotEncoder(
categorical_features=range(categorical_features_count))
X = onehotencoder.fit_transform(X).toarray()
# pour éviter le piège de la "dummy variable" on va retirer une colonne pour chaque ajout de colonnes
# typiquement pour les trucs binaires on aurait pas du rajouter de colonnes (on verra ca plus tard)
# donc la c'est un peu technique, il faudra relire ca en détail
to_delete = []
t = 0
for i in range(categorical_features_count):
to_delete.append(t)
t += column_ranges[i]
mask = []
for s in range(len(X[0, :])):
if s not in to_delete:
mask.append(s)
X = X[:, mask]
if scale:
if scaler == None:
sc = StandardScaler()
X = sc.fit_transform(X)
return X, y, sc
else:
X = scaler.transform(X)
return X, y