"TRIP_ID", "START_POINT", "GRID_POLYLINE",
"TRUNC_GRID_POLYLINE"
],
converters={
"START_POINT": lambda x: eval(x),
"GRID_POLYLINE": lambda x: eval(x),
"TRUNC_GRID_POLYLINE": lambda x: eval(x)
})
# Loop through every trip to make a prediction
for nr, partial_trip in validation.iterrows():
# Select the last segment of the partial trip
end = partial_trip.TRUNC_GRID_POLYLINE[-k:]
# Create a DestinationGrid object
posterior = DestinationGrid(N, M)
# Match this segment with trips in the training set
for i, chunk in enumerate(train):
# Only look at trips that start from the same position
trips = chunk[chunk.START_CELL == partial_trip.GRID_POLYLINE[0]]
for idx, row in trips.iterrows():
if match(end, row.GRID_POLYLINE, k):
destination = row.GRID_POLYLINE[-1]
posterior._table[destination] += 1
posterior.normalizeProbs()
# Export the probabilities
dests = [id_to_nr(dest) for dest in posterior]
for simulation in xrange(S):
# Initialize the walker
wlk = Walker(lattice = ltc, start = start_simulation, dest_threshold = 1)
# Simulate a random walk and record the destination
destinations.append(wlk.simulateWalker())
# Use pandas facilities
table = pd.DataFrame({"DEST": pd.Series(destinations)})
table["COUNT"] = 1
prob_table = table.groupby(["DEST"], as_index = False).aggregate({"COUNT": "sum"})
prob_table.COUNT = prob_table.COUNT / np.sum(prob_table.COUNT)
# Store the probabilities in a DestinationGrid object
posterior = DestinationGrid(N, M)
for idx, row in prob_table.iterrows():
posterior.setProb(row.DEST, row.COUNT)
# Plot the new distribution
plt.subplot(1,2,1)
plt.imshow(np.log(posterior.as_array() + trip_to_array(partial_trip.GRID_POLYLINE, N, M)), interpolation = "nearest")
plt.title("Complete trip superimposed")
plt.subplot(1,2,2)
plt.imshow(np.log(posterior.as_array() + trip_to_array(partial_trip.TRUNC_GRID_POLYLINE, N, M)), interpolation = "nearest")
plt.title("Partial trip superimposed")
# Initialize the walker
wlk = Walker(lattice=ltc, start=start_simulation, dest_threshold=1)
# Simulate a random walk and record the destination
destinations.append(wlk.simulateWalker())
# Use pandas facilities
table = pd.DataFrame({"DEST": pd.Series(destinations)})
table["COUNT"] = 1
prob_table = table.groupby(["DEST"],
as_index=False).aggregate({"COUNT": "sum"})
prob_table.COUNT = prob_table.COUNT / np.sum(prob_table.COUNT)
# Store the probabilities in a DestinationGrid object
posterior = DestinationGrid(N, M)
for idx, row in prob_table.iterrows():
posterior.setProb(row.DEST, row.COUNT)
# Plot the new distribution
plt.subplot(1, 2, 1)
plt.imshow(np.log(posterior.as_array() +
trip_to_array(partial_trip.GRID_POLYLINE, N, M)),
interpolation="nearest")
plt.title("Complete trip superimposed")
plt.subplot(1, 2, 2)
plt.imshow(np.log(posterior.as_array() +
trip_to_array(partial_trip.TRUNC_GRID_POLYLINE, N, M)),
interpolation="nearest")
# Import the validation data
validation = pd.read_csv(filepath_or_buffer = filepath_val,
sep = ",",
usecols = ["TRIP_ID", "START_POINT", "GRID_POLYLINE", "TRUNC_GRID_POLYLINE"],
converters = {"START_POINT": lambda x: eval(x),
"GRID_POLYLINE": lambda x: eval(x),
"TRUNC_GRID_POLYLINE": lambda x: eval(x)})
# Loop through every trip to make a prediction
for nr, partial_trip in validation.iterrows():
# Select the last segment of the partial trip
end = partial_trip.TRUNC_GRID_POLYLINE[-k:]
# Create a DestinationGrid object
posterior = DestinationGrid(N, M)
# Match this segment with trips in the training set
for i, chunk in enumerate(train):
# Only look at trips that start from the same position
trips = chunk[chunk.START_CELL == partial_trip.GRID_POLYLINE[0]]
for idx, row in trips.iterrows():
if match(end, row.GRID_POLYLINE, k):
destination = row.GRID_POLYLINE[-1]
posterior._table[destination] += 1
posterior.normalizeProbs()
# Export the probabilities
dests = [id_to_nr(dest) for dest in posterior]
{ #"TIMESTAMP" : lambda x: datetime.datetime.fromtimestamp(x),
"POLYLINE": lambda x: json.loads(x),
"START_POINT": lambda x: eval(x),
"GRID_POLYLINE": lambda x: eval(x),
"TRUNC_POLYLINE": lambda x: eval(x),
"TRUNC_GRID_POLYLINE": lambda x: eval(x)
})
# Select a partial trip
partial_trip = test.loc[0]
# Select the last segment of the partial trip
end = partial_trip.TRUNC_GRID_POLYLINE[-k:]
# Create a DestinationGrid object
grid = DestinationGrid(N, M)
# Match this segment with trips in the training set
for i, chunk in enumerate(trips):
for idx, row in chunk.iterrows():
if match(end, row.GRID_POLYLINE, k):
dest = row.GRID_POLYLINE[-1]
grid._table[dest] += 1
print "Processed chunk %d" % i
grid.normalizeProbs()
# Plot the new distribution
plt.subplot(1, 2, 1)
plt.imshow(np.log(grid.as_array() +
trips["MATCH"] = False
# Match this segment with trips in the training set
for idx, row in trips.iterrows():
if match(end, row.GRID_POLYLINE, k):
trips.set_value(idx, 'MATCH', True)
# We are only really interested in the matched trips now
trips = trips[trips.MATCH]
# Compute the destination distribution
trips_agg = trips.groupby(["DEST_CELL"],
as_index=False).aggregate({"MATCH": "sum"})
trips_agg.MATCH = trips_agg.MATCH / np.sum(trips_agg.MATCH)
# Store it in a DestinationGrid object
grid = DestinationGrid(N, M)
grid.setProbs(trips_agg.DEST_CELL.values, trips_agg.MATCH.values)
# Plot the new distribution
plt.subplot(1, 2, 1)
plt.imshow(np.log(grid.as_array() +
trip_to_array(partial_trip.GRID_POLYLINE, N, M)),
interpolation="nearest")
plt.title("Complete trip superimposed")
plt.subplot(1, 2, 2)
plt.imshow(np.log(grid.as_array() +
trip_to_array(partial_trip.TRUNC_GRID_POLYLINE, N, M)),
interpolation="nearest")
plt.title("Partial trip superimposed")
nrows = 10,
converters = {#"TIMESTAMP" : lambda x: datetime.datetime.fromtimestamp(x),
"POLYLINE": lambda x: json.loads(x),
"START_POINT": lambda x: eval(x),
"GRID_POLYLINE": lambda x: eval(x),
"TRUNC_POLYLINE": lambda x: eval(x),
"TRUNC_GRID_POLYLINE": lambda x: eval(x)})
# Select a partial trip
partial_trip = test.loc[0]
# Select the last segment of the partial trip
end = partial_trip.TRUNC_GRID_POLYLINE[-k:]
# Create a DestinationGrid object
grid = DestinationGrid(N, M)
# Match this segment with trips in the training set
for i, chunk in enumerate(trips):
for idx, row in chunk.iterrows():
if match(end, row.GRID_POLYLINE, k):
dest = row.GRID_POLYLINE[-1]
grid._table[dest] += 1
print "Processed chunk %d" % i
grid.normalizeProbs()
# Plot the new distribution
plt.subplot(1,2,1)
plt.imshow(np.log(grid.as_array() + trip_to_array(partial_trip.GRID_POLYLINE, N, M)), interpolation = "nearest")
# Select the last segment of the partial trip
end = partial_trip.TRUNC_GRID_POLYLINE[-k:]
# Create a dummy flag to indicate if a trip matches or not
trips["MATCH"] = False
# Match this segment with trips in the training set
for idx, row in trips.iterrows():
if match(end, row.GRID_POLYLINE, k):
trips.set_value(idx, 'MATCH', True)
# We are only really interested in the matched trips now
trips = trips[trips.MATCH]
# Compute the destination distribution
trips_agg = trips.groupby(["DEST_CELL"], as_index = False).aggregate({"MATCH": "sum"})
trips_agg.MATCH = trips_agg.MATCH / np.sum(trips_agg.MATCH)
# Store it in a DestinationGrid object
grid = DestinationGrid(N, M)
grid.setProbs(trips_agg.DEST_CELL.values, trips_agg.MATCH.values)
# Plot the new distribution
plt.subplot(1,2,1)
plt.imshow(np.log(grid.as_array() + trip_to_array(partial_trip.GRID_POLYLINE, N, M)), interpolation = "nearest")
plt.title("Complete trip superimposed")
plt.subplot(1,2,2)
plt.imshow(np.log(grid.as_array() + trip_to_array(partial_trip.TRUNC_GRID_POLYLINE, N, M)), interpolation = "nearest")
plt.title("Partial trip superimposed")
