taxi

commit 222971123f2741ee3689092fb4396dac83a13338
parent f6d2c6fc47f93b158b70b5c0c9a45324041ca4d5
Author: Alex Auvolat <alex.auvolat@ens.fr>
Date:   Thu, 21 May 2015 17:05:07 -0400

Implement cut-based iteration scheme (SLOW!!!)

Diffstat:
A
data/cut.py
 | 
31
+++++++++++++++++++++++++++++++
A
data/make_time_index.py
 | 
50
++++++++++++++++++++++++++++++++++++++++++++++++++
M
model/mlp.py
 | 
5
+++--

3 files changed, 84 insertions(+), 2 deletions(-)
diff --git a/data/cut.py b/data/cut.py
@@ -0,0 +1,31 @@
+from fuel.schemes import IterationScheme
+import sqlite3
+import random
+import os
+from picklable_itertools import iter_
+
+import data
+
+first_time = 1372636853
+last_time = 1404172787
+
+
+class TaxiTimeCutScheme(IterationScheme):
+    def __init__(self, dbfile=None, use_cuts=None):
+        self.dbfile = os.path.join(data.path, 'time_index.db') if dbfile == None else dbfile
+        self.use_cuts = use_cuts
+
+    def get_request_iterator(self):
+        cuts = self.use_cuts
+        if cuts == None:
+            cuts = [random.randrange(first_time, last_time) for _ in range(100)]
+
+        l = []
+        with sqlite3.connect(self.dbfile) as db:
+            c = db.cursor()
+            for cut in cuts:
+                l = l + [i for (i,) in
+                    c.execute('SELECT trip FROM trip_times WHERE begin <= ? AND end >= ?', (cut, cut))]
+                
+        return iter_(l)
+
diff --git a/data/make_time_index.py b/data/make_time_index.py
@@ -0,0 +1,50 @@
+#!/usr/bin/env python
+# Make a valid dataset by cutting the training set at specified timestamps
+
+import os
+import sys
+import importlib
+
+import h5py
+import numpy
+
+import data
+from data.hdf5 import taxi_it
+
+import sqlite3
+
+def make_valid(outpath):
+    times = []
+    for i, line in enumerate(taxi_it('train')):
+        time = line['timestamp']
+        latitude = line['latitude']
+
+        if len(latitude) == 0:
+            continue
+
+        duration = 15 * (len(latitude) - 1)
+
+        times.append((i, int(time), int(time + duration)))
+        if i % 1000 == 0:
+            print times[-1]
+
+
+    with sqlite3.connect(outpath) as timedb:
+        c = timedb.cursor()
+        c.execute('''
+                CREATE TABLE trip_times
+                    (trip INTEGER, begin INTEGER, end INTEGER)
+        ''')
+        print "Adding data..."
+        c.executemany('INSERT INTO trip_times(trip, begin, end) VALUES(?, ?, ?)', times)
+        timedb.commit()
+        print "Creating index..."
+        c.execute('''CREATE INDEX trip_time_index ON trip_times (begin, end)''')
+
+
+if __name__ == '__main__':
+    if len(sys.argv) < 1 or len(sys.argv) > 2:
+        print >> sys.stderr, 'Usage: %s [outfile]' % sys.argv[0]
+        sys.exit(1)
+    outpath = os.path.join(data.path, 'time_index.db') if len(sys.argv) < 2 else sys.argv[1]
+    make_valid(outpath)
diff --git a/model/mlp.py b/model/mlp.py
@@ -8,6 +8,7 @@ from blocks.bricks import application, MLP, Rectifier, Initializable
 import data
 from data import transformers
 from data.hdf5 import TaxiDataset, TaxiStream
+from data.cut import TaxiTimeCutScheme
 from model import ContextEmbedder
 
 
@@ -51,13 +52,13 @@ class Stream(object):
 
     def train(self, req_vars):
         stream = TaxiDataset('train')
-        stream = DataStream(stream, iteration_scheme=ShuffledExampleScheme(stream.num_examples))
+        stream = DataStream(stream, iteration_scheme=TaxiTimeCutScheme())
 
         valid = TaxiDataset(self.config.valid_set, 'valid.hdf5', sources=('trip_id',))
         valid_trips_ids = valid.get_data(None, slice(0, valid.num_examples))[0]
 
         stream = transformers.TaxiExcludeTrips(valid_trips_ids, stream)
-        stream = transformers.TaxiGenerateSplits(stream, max_splits=100)
+        stream = transformers.TaxiGenerateSplits(stream, max_splits=1)
 
         stream = transformers.TaxiAddDateTime(stream)
         stream = transformers.TaxiAddFirstLastLen(self.config.n_begin_end_pts, stream)