def test_train(self):
global i
def side_effect(*args, **kwargs):
global i
if kwargs == {}:
effect_collector.append('init')
else:
i += 1
effect_collector.append(effects[i])
test_cases = [
(
# RNN Model to be used
RNNModel(2, 3),
# X train to be used
np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]]),
# y train to be used
np.array([0, 1, 1]),
# sequence lengths
np.array([5, 7, 2])),
(
# RNN Model to be used
RNNModel(2, 3, num_epochs=10),
# X train to be used
np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]]),
# y train to be used
np.array([0, 1, 1]),
# sequence lengths
np.array([5, 7, 2])),
(
# RNN Model to be used
RNNModel(3, 3, num_epochs=10),
# X train to be used
np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]]),
# y train to be used
np.array([0, 1, 2]),
# sequence lengths
np.array([5, 7, 2]))
]
for rnn_model, X_train, y_train, sequence_lengths in test_cases:
i = -1
effect_collector = []
effects = ['train' + str(i) for i in range(rnn_model.num_epochs)]
self.assertEqual(len(effects), rnn_model.num_epochs)
preprocess.upsample = MagicMock()
rnn_model.spawn_clf = MagicMock()
tf.global_variables_initializer = MagicMock()
rnn_model.sess = MagicMock()
rnn_model.sess.run.side_effect = side_effect
rnn_model.opt = MagicMock()
rnn_model.x = MagicMock()
rnn_model.sequence_lengths = MagicMock()
rnn_model.y = MagicMock()
rnn_model.train(X_train, y_train, sequence_lengths)
self.assertListEqual(effect_collector, ["init"] + effects)
def test_predict(self):
test_cases = [
(
# Feature Matrix X
np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]]),
# the predictions to check for
np.array([0, 1, 1]),
# the RNN Model to use
RNNModel(2, 3)),
(
# Feature Matrix X
np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]]),
# the predictions to check for
np.array([0, 1, 2]),
# the RNN Model to use
RNNModel(3, 3))
]
for X, true_predictions, rnn_model in test_cases:
# mock rnn attributes that would be set by spawn_clf
rnn_model.sess = MagicMock()
rnn_model.prediction = MagicMock()
rnn_model.x = MagicMock()
rnn_model.sequence_lengths = MagicMock()
rnn_model.sess.run.return_value = true_predictions
self.assertTrue(
np.array_equal(rnn_model.predict(X, []), true_predictions))
def test_get_incremental_score(self):
test_cases = [
# scores to be returned in order by mock of model.score
np.array(
[0, 0, 0.5, 0.3, 0.8, 0.6, 0.9, 0.7, 1, 0.8, 1, 0.9, 1, 1]),
np.array([0, 144, 222, -0.29, 1, 1])
]
for scores in test_cases:
model = RNNModel(3, 8, num_epochs=int(len(scores) / 2))
model.score = MagicMock()
model.score.side_effect = scores
model.spawn_clf = MagicMock()
# mock tensorflow network attributes
model.sess = MagicMock()
model.sess.run = MagicMock()
model.opt = MagicMock()
model.x = MagicMock()
model.y = MagicMock()
model.sequence_lengths = MagicMock()
preprocess.up_sample = MagicMock()
# none of the inputs are actually used due to mocking so just give dummy inputs
train_scores, test_scores = model.incremental_score(
[], [], [], [],
train_sequence_lengths=[],
test_sequence_lengths=[])
self.assertListEqual(
train_scores,
list(scores[[i for i in range(0, len(scores), 2)]]))
self.assertListEqual(
test_scores,
list(scores[[i for i in range(1, len(scores), 2)]]))
def test_get_confusion_matrix(self):
test_cases = [(np.array([0, 0, 0, 1, 1, 1, 2, 2,
2]), np.array([0, 1, 2, 0, 1, 2, 0, 1,
2]), np.array([[1] * 3] * 3)),
(np.array([0, 0, 0, 1, 1, 1, 2, 2,
2]), np.array([0, 0, 0, 1, 1, 1, 2, 2, 2]),
np.array([[3, 0, 0], [0, 3, 0], [0, 0, 3]])),
(np.array([0, 0, 0, 1, 1, 1, 2, 2,
2]), np.array([1, 1, 1, 2, 2, 2, 0, 0, 0]),
np.array([[0, 3, 0], [0, 0, 3], [3, 0, 0]]))]
for y_true, y_pred, conf_mat in test_cases:
model = RNNModel(3, 8)
self.assertTrue(
np.array_equal(
model.get_confusion_matrix(y_true, y_pred=y_pred),
conf_mat))
def check_no_sequence_lengths_all_methods(self):
rnn_model = RNNModel(2, 3)
X_train = [[1, 2, 3], [4, 5, 6]]
y_train = [0, 1]
self.assertRaises(ValueError, rnn_model.train, X_train, y_train)
self.assertRaises(ValueError, rnn_model.predict, X_train)
self.assertRaises(ValueError, rnn_model.score, X_train, y_train)
self.assertRaises(ValueError, rnn_model.incremental_score, X_train,
y_train, X_train, y_train)
self.assertRaises(ValueError,
rnn_model.incremental_score,
X_train,
y_train,
X_train,
y_train,
train_sequence_lengths=[])
self.assertRaises(ValueError,
rnn_model.incremental_score,
X_train,
y_train,
X_train,
y_train,
test_sequence_lengths=[])
self.assertRaises(ValueError, rnn_model.get_confusion_matrix, X_train,
y_train)
def test_fill_default_params(self):
#check filling params with all defaults
correct_default_params = {
'rnn_cell_type': tf.contrib.rnn.GRUCell,
'learning_rate': 0.001,
'fully_connected_cells': 8,
'num_stacks': 2,
'num_rnn_units': 128
}
model = RNNModel(3, 8)
self.assertDictEqual(correct_default_params, model.clf_params)
# check some params filled in
clf_params = {
'num_stacks': 8,
'rnn_cell_type': tf.contrib.rnn.LSTMCell
}
model = RNNModel(3, 8, clf_params=clf_params)
correct_params = {
'rnn_cell_type': tf.contrib.rnn.LSTMCell,
'learning_rate': 0.001,
'fully_connected_cells': 8,
'num_stacks': 8,
'num_rnn_units': 128
}
self.assertDictEqual(correct_params, model.clf_params)
# check second level dict param filling
clf_params = {'num_stacks': 8, "random_param": 789}
model = RNNModel(3, 8, clf_params=clf_params)
correct_params = {
'rnn_cell_type': tf.contrib.rnn.GRUCell,
'learning_rate': 0.001,
'random_param': 789,
'fully_connected_cells': 8,
'num_stacks': 8,
'num_rnn_units': 128
}
self.assertDictEqual(correct_params, model.clf_params)
def test_score(self):
test_cases = [
(
# True values for y
np.array([1, 0, 0, 0, 1]),
# Predicted values for y (used as return value for mocked prediction method - tested separately)
np.array([0, 1, 0, 0, 1]),
# score
0.6),
(np.array([0, 0, 0, 1, 1, 1, 2, 2,
2]), np.array([0, 1, 2, 0, 1, 2, 0, 1, 2]), 3 / 9),
(np.array([0, 0, 0, 1, 1, 1, 2, 2,
2]), np.array([1, 1, 1, 2, 2, 2, 0, 0, 0]), 0),
(np.array([0, 0, 0, 1, 1, 1, 2, 2,
2]), np.array([0, 0, 0, 1, 1, 1, 2, 2, 2]), 1),
]
for y_true, y_pred, score in test_cases:
model = RNNModel(3, 8)
model.predict = MagicMock()
model.predict.return_value = y_pred
# no params except y_true are used so just input dummy params
self.assertAlmostEqual(
model.score([], y_true, sequence_lengths=[]), score)
"""
This program creates the log files necessary for TensorBoard to visualise the architecture. It provides code for
changing the parameters of the model, the number of windows, number of features and number of classes. Different
versions were experimented with to ensure the model worked properly
In order to display the visualisation follow the steps below. The logfiles are stored in a folder called tmp in the
directory containing this python file.
1 - Go to the terminal
2 - Use the tensorboard command to display the visualisation on a local server. If the path to the tmp directory created
is path/to/tmp/ then run the following command -
tensorboard --logdir=path/to/tmp/ --port 6006
3 - Open the following link in your browser - http://localhost:6006/
"""
import sys, os
sys.path.append(os.path.join(sys.path[0], "../.."))
from analysis.rnn_model import RNNModel
import settings
settings.num_windows = settings.max_reading_length
num_classes = 3
num_features = len(settings.used_sensors) * len(
settings.used_sensor_attributes) * settings.num_windows
clf_params = None
model = RNNModel(num_classes, num_features, clf_params=clf_params)
model.spawn_clf()
model.write_graph_for_visualisation()