def transform(self, data):
"""Return imputed data by trained TGAIN model.
Args:
- data: incomplete 3d numpy array
Returns:
- data_imputed: complete 3d numpy array after imputation
"""
# Only after fitting
assert self.gain_model is not None
# Data preprocessing
no, time, dim = data.shape
data_concat = np.reshape(data, [no * time, dim])
# Only with non-padded data
idx = np.where(data_concat[:, 1] != -1)[0]
# Apply TGAIN
data_imputed = data_concat.copy()
data_imputed[idx, :] = self.gain_model.transform(data_concat[idx, :])
# Rounding
data_imputed = rounding(data_concat, data_imputed)
# Return to 3d array
data_imputed = np.reshape(data_imputed, [no, time, dim])
return data_imputed
def transform(self, data):
"""Return imputed data by trained GAIN model.
Args:
- data: 2d numpy array with missing data
Returns:
- imputed data: 2d numpy array without missing data
"""
# Only after fitting
assert self.norm_parameters is not None
assert os.path.exists(self.save_file_directory) is True
min_val = self.norm_parameters["min"]
max_val = self.norm_parameters["max"]
no, dim = data.shape
# Set missing
m = 1 - (1 * (np.isnan(data)))
x = np.nan_to_num(data)
## Imputed data
z = self.sample_Z(no, dim)
x = m * x + (1 - m) * z
# Restore the trained GAIN model
graph = tf.Graph()
with graph.as_default():
with tf.Session() as sess:
tf.saved_model.loader.load(sess, [tag_constants.SERVING],
self.save_file_directory)
X = graph.get_tensor_by_name("Placeholder_2:0")
M = graph.get_tensor_by_name("Placeholder:0")
G_sample = graph.get_tensor_by_name("Sigmoid:0")
imputed_data = sess.run(G_sample, feed_dict={X: x, M: m})
# Renormalize
for i in range(dim):
imputed_data[:, i] = imputed_data[:, i] * (max_val[i] + 1e-8)
imputed_data[:, i] = imputed_data[:, i] + min_val[i]
# Rounding
imputed_data = rounding(x, imputed_data)
return imputed_data
def transform(self, dataset):
"""Return imputed dataset by standard imputation.
Args:
- dataset: incomplete dataset
Returns:
- dataset: imputed dataset by standard imputation.
"""
assert self.imputation_model is not None
if dataset.static_feature is not None:
# Standard imputation
data_imputed = self.imputation_model.transform(
dataset.static_feature)
# Rounding
dataset.static_feature = rounding(dataset.static_feature,
data_imputed)
return dataset
def transform (self, data):
"""Return imputed data by trained MRNN model.
Args:
- data: 3d numpy array with missing data
Returns:
- data_imputed: 3d numpy array with imputed data
"""
# Only after fit
assert self.median_vals is not None
assert self.mrnn_model is not None
# Data preprocess for mrnn
x, m, t, data_new, _ = self.mrnn_data_preprocess(data)
# Impute missing data
imputed_x = self.mrnn_model.transform(x, m, t, self.median_vals)
data_imputed = (1-m) * imputed_x + m * data_new
# Rounding
data_imputed = rounding(data_new, data_imputed)
return data_imputed