Newer
Older
DONUT
=====
Donut is an anomaly detection algorithm for periodic KPIs.
Installation
------------
Checkout this repository and execute:
.. code-block:: bash
pip install git+https://github.com/thu-ml/zhusuan.git
pip install git+https://github.com/korepwx/tfsnippet.git
pip install .
This will first install `ZhuSuan <https://github.com/thu-ml/zhusuan>`_ and
`TFSnippet <https://github.com/korepwx/tfsnippet>`_, the two major dependencies
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
of Donut, then install the Donut package itself.
API Usage
---------
To prepare the data:
.. code-block:: python
import numpy as np
from donut import complete_timestamp, standardize_kpi
# Read the raw data.
timestamp, values, labels = ...
# If there is no label, simply use all zeros.
labels = np.zeros_like(values, dtype=np.int32)
# Complete the timestamp, and obtain the missing point indicators.
timestamp, missing, (values, labels) = \
complete_timestamp(timestamp, (values, labels))
# Split the training and testing data.
test_portion = 0.3
test_n = int(len(values) * test_portion)
train_values, test_values = values[:-test_n], values[-test_n:]
train_labels, test_labels = labels[:-test_n], labels[-test_n:]
train_missing, test_missing = missing[:-test_n], missing[-test_n:]
# Standardize the training and testing data.
train_values, mean, std = standardize_kpi(
train_values, excludes=np.logical_or(train_labels, train_missing))
test_values, _, _ = standardize_kpi(test_values, mean=mean, std=std)
To construct a Donut model:
.. code-block:: python
import tensorflow as tf
from donut import Donut
from tensorflow import keras as K
from tfsnippet.modules import Sequential
# We build the entire model within the scope of `model_vs`,
# it should hold exactly all the variables of `model`, including
# the variables created by Keras layers.
with tf.variable_scope('model') as model_vs:
model = Donut(
h_for_p_x=Sequential([
K.layers.Dense(100, W_regularizer=K.regularizers.l2(0.001),
activation=tf.nn.relu),
K.layers.Dense(100, W_regularizer=K.regularizers.l2(0.001),
activation=tf.nn.relu),
]),
h_for_q_z=Sequential([
K.layers.Dense(100, W_regularizer=K.regularizers.l2(0.001),
activation=tf.nn.relu),
K.layers.Dense(100, W_regularizer=K.regularizers.l2(0.001),
activation=tf.nn.relu),
]),
x_dims=120,
z_dims=5,
)
To train the Donut model:
.. code-block:: python
from donut import DonutTrainer
trainer = DonutTrainer(model=model, model_vs=model_vs)
with tf.Session().as_default():
trainer.fit(train_values, train_labels, train_missing, mean, std)
To use a trained Donut model for prediction:
.. code-block:: python
from donut import DonutPredictor
predictor = DonutPredictor(model)
with tf.Session().as_default():
# Remember to train the model before using the predictor,
# or to restore the saved model.
...
# Now we can use the predictor.
test_score = predictor.get_score(test_values, test_missing)
To save and restore a trained model:
.. code-block:: python
from tfsnippet.utils import get_variables_as_dict, VariableSaver
with tf.Session().as_default():
# Train the model.
...
# Remember to get the model variables after the birth of a
# `predictor` or a `trainer`. The :class:`Donut` instances
# does not build the graph until :meth:`Donut.get_score` or
# :meth:`Donut.get_training_objective` is called, which is
# done in the `predictor` or the `trainer`.
var_dict = get_variables_as_dict(model_vs)
# save variables to `save_dir`
saver = VariableSaver(var_dict, save_dir)
saver.save()
with tf.Session().as_default():
# Restore variables from `save_dir`.
saver = VariableSaver(get_variables_as_dict(model_vs), save_dir)
saver.restore()
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
If you need more advanced outputs from the model, you may derive the outputs
by using `model.vae` directly, for example:
.. code-block:: python
from donut import iterative_masked_reconstruct
# Obtain the reconstructed `x`, with MCMC missing data imputation.
# See also:
# :meth:`donut.Donut.get_score`
# :func:`donut.iterative_masked_reconstruct`
# :meth:`tfsnippet.modules.VAE.reconstruct`
input_x = ... # 2-D `float32` :class:`tf.Tensor`, input `x` windows
input_y = ... # 2-D `int32` :class:`tf.Tensor`, missing point indicators
# for the `x` windows
x_r = model.vae.reconstruct(
iterative_masked_reconstruct(
reconstruct=model.vae.reconstruct,
x=input_x,
mask=input_y,
iter_count=mcmc_iteration,
back_prop=False
)
)
# `x_r` is a :class:`tfsnippet.stochastic.StochasticTensor`, from which
# you may derive many useful outputs, for example:
x_r.tensor # the `x` samples
x_r.log_prob(group_ndims=0) # element-wise log p(x|z)
x_r.distribution.mean, x_r.distribution.std # mean and std of p(x|z)