def make_prediction(webcam, params, timestamp):
"""
Make a prediction using NearestNeighbors algorithm.
Parameters
----------
webcam : Webcam instance
The source of pictures
params : PredictionParams instance
The parameters to use to compute the prediction
timestamp : int
The Unix-Epoch-timestamp (UTC) of when the prediction is made
Return
------
Prediction instance
"""
cam_id = webcam.webcam_id
result = None
with webcam_fs.get_dataset(cam_id) as dataset:
ex_set = dataset.get_set(params.name)
new_input = dataset.make_input(ex_set, timestamp)
if new_input is not None and len(ex_set.input) > 0:
neighbors = NearestNeighbors()
neighbors.fit(ex_set.input)
output_ref = neighbors.predict(new_input)
output = dataset.output_img(ex_set, output_ref)
# FIXME reference existing image (data and file)
imgpath = webcam_fs.prediction_path(cam_id, params.name, timestamp)
result = Prediction(params=params, path=imgpath)
result.comp_date = datetime.fromtimestamp(timestamp, utc)
result.sci_bytes = output
result.create()
return result
def test_predict_wminkowski(fixed_array):
"Weighing the metric should be taken into account"
nn = NearestNeighbors(metric='wminkowski', w=np.array([10, 1, 1]))
nn.fit(*fixed_array)
y4 = nn.predict(np.array([4.4, 4.4, 4.6]))
y5 = nn.predict(np.array([4.6, 4.4, 4.4]))
assert(y4 == 4)
assert(y5 == 5)
def test_predict_extendedarray(variable_array, fixed_array):
"Extending the array after fitting it should not be a problem"
X, y = variable_array
nn = NearestNeighbors()
nn.fit(X, y)
newX, newy = fixed_array
X.append(newX)
y.append(newy)
y = nn.predict(np.array([4.2, 4.2, 4.2]))
assert(y == 4)
def test_predict_notexisting(fixed_array):
"Find the nearest neighbor of an unexisting row"
nn = NearestNeighbors()
nn.fit(*fixed_array)
y = nn.predict(np.array([4.6, 4.4, 4.4]))
assert(y == 4)
def test_predict_multibatch(small_batch, fixed_array):
"An existing row should be found, only one batch"
nn = NearestNeighbors()
nn.fit(*fixed_array)
y = nn.predict(np.array([17, 17, 17]))
assert(y == 17)