Merge branch 'master' of github.com:adbrebs/taxi

4 files changed, 33 insertions, 15 deletions

diff --git a/README.md b/README.md
new file mode 100644
index 0000000..00eb60b
--- /dev/null
+++ b/README.md
@@ -0,0 +1,3 @@
+Winning entry to the Kaggle ECML/PKDD destination competition.
+
+https://www.kaggle.com/c/pkdd-15-predict-taxi-service-trajectory-i
diff --git a/config/dest_mlp_tgtcls_1_cswdtx_batchshuffle.py b/config/dest_mlp_tgtcls_1_cswdtx_batchshuffle.py
index a4db33c..b816930 100644
--- a/config/dest_mlp_tgtcls_1_cswdtx_batchshuffle.py
+++ b/config/dest_mlp_tgtcls_1_cswdtx_batchshuffle.py
@@ -23,14 +23,14 @@ dim_embeddings = [
 ]
 
 dim_input = n_begin_end_pts * 2 * 2 + sum(x for (_, _, x) in dim_embeddings)
-dim_hidden = [1000]
+dim_hidden = [500]
 dim_output = tgtcls.shape[0]
 
 embed_weights_init = IsotropicGaussian(0.01)
 mlp_weights_init = IsotropicGaussian(0.1) 
 mlp_biases_init = Constant(0.01)
 
-step_rule = Momentum(learning_rate=0.01, momentum=0.9)
+step_rule = Momentum(learning_rate=0.001, momentum=0.99)
 
 batch_size = 200
 
diff --git a/data_analysis/cluster_arrival.py b/data_analysis/cluster_arrival.py
index fd4ea04..5e990cd 100644..100755
--- a/data_analysis/cluster_arrival.py
+++ b/data_analysis/cluster_arrival.py
@@ -1,20 +1,31 @@
-import matplotlib.pyplot as plt
+#!/usr/bin/env python
 import numpy
 import cPickle
 import scipy.misc
+import os
 
 from sklearn.cluster import MeanShift, estimate_bandwidth
 from sklearn.datasets.samples_generator import make_blobs
 from itertools import cycle
 
-print "Reading arrival point list"
-with open("arrivals.pkl") as f:
-    pts = cPickle.load(f)
+import data
+from data.hdf5 import taxi_it
+from data.transformers import add_destination
+
+print "Generating arrival point list"
+dests = []
+for v in taxi_it("train"):
+    if len(v['latitude']) == 0: continue
+    dests.append([v['latitude'][-1], v['longitude'][-1]])
+pts = numpy.array(dests)
+
+with open(os.path.join(data.path, "arrivals.pkl"), "w") as f:
+    cPickle.dump(pts, f, protocol=cPickle.HIGHEST_PROTOCOL)
 
 print "Doing clustering"
 bw = estimate_bandwidth(pts, quantile=.1, n_samples=1000)
 print bw
-bw = 0.001
+bw = 0.001 # (
 
 ms = MeanShift(bandwidth=bw, bin_seeding=True, min_bin_freq=5)
 ms.fit(pts)
@@ -22,6 +33,6 @@ cluster_centers = ms.cluster_centers_
 
 print "Clusters shape: ", cluster_centers.shape
 
-with open("arrival-cluters.pkl", "w") as f:
+with open(os.path.join(data.path, "arrival-clusters.pkl"), "w") as f:
     cPickle.dump(cluster_centers, f, protocol=cPickle.HIGHEST_PROTOCOL)
 
diff --git a/model/mlp.py b/model/mlp.py
index 1f53e8c..7d04c82 100644
--- a/model/mlp.py
+++ b/model/mlp.py
@@ -52,6 +52,12 @@ class FFMLP(Initializable):
     def predict_inputs(self):
         return self.inputs
 
+class UniformGenerator(object):
+    def __init__(self):
+        self.rng = numpy.random.RandomState(123)
+    def __call__(self, *args):
+        return float(self.rng.uniform())
+
 class Stream(object):
     def __init__(self, config):
         self.config = config
@@ -69,17 +75,15 @@ class Stream(object):
 
         stream = transformers.TaxiExcludeTrips(stream, valid_trips_ids)
         stream = transformers.TaxiGenerateSplits(stream, max_splits=self.config.max_splits)
-        stream = transformers.add_destination(stream)
-
-        stream = transformers.taxi_add_datetime(stream)
-        stream = transformers.taxi_add_first_last_len(stream, self.config.n_begin_end_pts)
-        stream = transformers.Select(stream, tuple(req_vars))
 
         if hasattr(self.config, 'shuffle_batch_size'):
             stream = transformers.Batch(stream, iteration_scheme=ConstantScheme(self.config.shuffle_batch_size))
-            rng = numpy.random.RandomState(123)
-            stream = Mapping(stream, SortMapping(lambda x: float(rng.uniform())))
+            stream = Mapping(stream, SortMapping(key=UniformGenerator()))
             stream = Unpack(stream)
+
+        stream = transformers.taxi_add_datetime(stream)
+        stream = transformers.taxi_add_first_last_len(stream, self.config.n_begin_end_pts)
+        stream = transformers.Select(stream, tuple(req_vars))
         
         stream = Batch(stream, iteration_scheme=ConstantScheme(self.config.batch_size))