def return_prediction(data, clf_filename='clf_rf'):
"""
Returns the predicted category of an object by using a trained network.
Parameters:
data (list): List of the 4 parameters (float or integer) of the object. These parameters
are the height, the width, the depth and the weight of the concerned object.
Raise a Value_Error exception if the 4 parameters describe abnormal object dimensions.
clf_filename (str): The file name of a trained classifier without its extension.
Default extension is a joblib file. (Default : 'clf_rf')
Returns:
flatten_list (str): Predicted category of the object. Can be either 'mlp', 'deco' or 'meuble'.
"""
# We put the data into a nested list
inputs = [data]
# We scale the data by retrieving the scaler computed on the training dataset
print("directory", os.getcwd())
try:
test_scaled = scale_test(inputs, filename='src/scaler_mdm.joblib')
except FileNotFoundError:
test_scaled = scale_test(inputs, filename='../src/scaler_mdm.joblib')
# We load the trained classifier and make the prediction on the data
try:
clf = load('src/' + clf_filename + '.joblib')
except FileNotFoundError:
clf = load('../src/' + clf_filename + '.joblib')
print(test_scaled)
prediction = clf.predict(test_scaled)
# We load the encoder to decode the prediction into a string
load_onehot = OneHotEncoder()
load_onehot.drop_idx_ = None
try:
load_onehot.categories_ = np.load('src/classes_onehot.npy',
allow_pickle=True)
except FileNotFoundError:
load_onehot.categories_ = np.load('../src/classes_onehot.npy',
allow_pickle=True)
# Raises ValueError Exception if the values are outliers
try:
str_pred = load_onehot.inverse_transform(prediction)
except ValueError:
return "Values are too different from training dataset"
# Return the category predicted
flatten_list = list(chain.from_iterable(str_pred))
print("Catégorie prédite : ", flatten_list[0])
return flatten_list[0]
def encoder_for(field):
if field.get('optype') != 'categorical':
return 'passthrough'
encoder = OneHotEncoder()
encoder.categories_ = np.array([self.field_mapping[field.get('name')][1].categories])
encoder._legacy_mode = False
return encoder
def complete_y_tranformation(self, Y):
# Y to matrix
y_encoder = None
Y = Y.astype(np.float32)
if len(Y.shape) == 1:
Y = Y.reshape(-1, 1)
# encode Y
if self.encode_Y:
y_encoder = OneHotEncoder(sparse=False,
categories="auto",
handle_unknown='ignore')
y_encoder.categories_ = np.array([])
Y = y_encoder.fit_transform(Y)
return Y, y_encoder
def fit(self, pipeline_config, X_train, X_valid, Y_train, Y_valid,
categorical_features):
ohe = OneHotEncoder(categories="auto",
sparse=False,
handle_unknown="ignore")
encoder = ColumnTransformer(transformers=[
("ohe", ohe, [i for i, f in enumerate(categorical_features) if f])
],
remainder="passthrough")
encoder.categories_ = np.array([])
encoder.categorical_features = categorical_features
if any(categorical_features) and not scipy.sparse.issparse(X_train):
# encode X
X_train = encoder.fit_transform(X_train)
if (X_valid is not None):
X_valid = encoder.transform(X_valid)
encoder.categories_ = encoder.transformers_[0][1].categories_
# Y to matrix
y_encoder = None
Y_train = Y_train.astype(np.float32)
if len(Y_train.shape) == 1:
Y_train = Y_train.reshape(-1, 1)
if Y_valid is not None and len(Y_valid.shape) == 1:
Y_valid = Y_valid.reshape(-1, 1)
# encode Y
if self.encode_Y and not scipy.sparse.issparse(Y_train):
y_encoder = OneHotEncoder(sparse=False,
categories="auto",
handle_unknown='ignore')
y_encoder.categories_ = np.array([])
Y_train = y_encoder.fit_transform(Y_train)
if Y_valid is not None:
Y_valid = y_encoder.transform(Y_valid)
return {
'X_train': X_train,
'X_valid': X_valid,
'one_hot_encoder': encoder,
'Y_train': Y_train,
'Y_valid': Y_valid,
'y_one_hot_encoder': y_encoder,
'categorical_features': None
}
