Add target class based model for time prediction (seems to work)

3 files changed, 111 insertions, 3 deletions

diff --git a/config/time_simple_mlp_tgtcls_2_cswdtx.py b/config/time_simple_mlp_tgtcls_2_cswdtx.py
new file mode 100644
index 0000000..4579df3
--- /dev/null
+++ b/config/time_simple_mlp_tgtcls_2_cswdtx.py
@@ -0,0 +1,41 @@
+import model.time_simple_mlp_tgtcls as model
+
+from blocks.initialization import IsotropicGaussian, Constant
+
+import data
+
+n_begin_end_pts = 5     # how many points we consider at the beginning and end of the known trajectory
+n_end_pts = 5
+
+n_valid = 1000
+
+# generate target classes as a Fibonacci sequence
+tgtcls = [1, 2]
+for i in range(22):
+    tgtcls.append(tgtcls[-1] + tgtcls[-2])
+
+dim_embeddings = [
+    ('origin_call', data.origin_call_size+1, 10),
+    ('origin_stand', data.stands_size+1, 10),
+    ('week_of_year', 52, 10),
+    ('day_of_week', 7, 10),
+    ('qhour_of_day', 24 * 4, 10),
+    ('day_type', 3, 10),
+    ('taxi_id', 448, 10),
+]
+
+dim_input = n_begin_end_pts * 2 * 2 + sum(x for (_, _, x) in dim_embeddings)
+dim_hidden = [500, 100]
+dim_output = len(tgtcls)
+
+embed_weights_init = IsotropicGaussian(0.001)
+mlp_weights_init = IsotropicGaussian(0.01)
+mlp_biases_init = Constant(0.001)
+
+exp_base = 1.5
+
+learning_rate = 0.0001
+momentum = 0.99
+batch_size = 32
+
+valid_set = 'cuts/test_times_0'
diff --git a/model/time_simple_mlp_tgtcls.py b/model/time_simple_mlp_tgtcls.py
new file mode 100644
index 0000000..1f1eab7
--- /dev/null
+++ b/model/time_simple_mlp_tgtcls.py
@@ -0,0 +1,67 @@
+from blocks.bricks import MLP, Rectifier, Linear, Sigmoid, Identity, Softmax
+from blocks.bricks.lookup import LookupTable
+
+import numpy
+import theano
+from theano import tensor
+
+import data
+import error
+
+class Model(object):
+    def __init__(self, config):
+        # The input and the targets
+        x_firstk_latitude = (tensor.matrix('first_k_latitude') - data.train_gps_mean[0]) / data.train_gps_std[0]
+        x_firstk_longitude = (tensor.matrix('first_k_longitude') - data.train_gps_mean[1]) / data.train_gps_std[1]
+
+        x_lastk_latitude = (tensor.matrix('last_k_latitude') - data.train_gps_mean[0]) / data.train_gps_std[0]
+        x_lastk_longitude = (tensor.matrix('last_k_longitude') - data.train_gps_mean[1]) / data.train_gps_std[1]
+
+        input_list = [x_firstk_latitude, x_firstk_longitude, x_lastk_latitude, x_lastk_longitude]
+        embed_tables = []
+
+        self.require_inputs = ['first_k_latitude', 'first_k_longitude', 'last_k_latitude', 'last_k_longitude']
+
+        for (varname, num, dim) in config.dim_embeddings:
+            self.require_inputs.append(varname)
+            vardata = tensor.lvector(varname)
+            tbl = LookupTable(length=num, dim=dim, name='%s_lookup'%varname)
+            embed_tables.append(tbl)
+            input_list.append(tbl.apply(vardata))
+
+        y = tensor.lvector('travel_time')
+
+        # Define the model
+        mlp = MLP(activations=[Rectifier() for _ in config.dim_hidden] + [Softmax()],
+                           dims=[config.dim_input] + config.dim_hidden + [config.dim_output])
+        classes = theano.shared(numpy.array(config.tgtcls, dtype=theano.config.floatX), name='classes')
+
+        # Create the Theano variables
+        inputs = tensor.concatenate(input_list, axis=1)
+        # inputs = theano.printing.Print("inputs")(inputs)
+        cls_probas = mlp.apply(inputs)
+        outputs = tensor.dot(cls_probas, classes)
+
+        # outputs = theano.printing.Print("outputs")(outputs)
+        # y = theano.printing.Print("y")(y)
+
+        outputs.name = 'outputs'
+
+        # Calculate the cost
+        cost = error.rmsle(outputs.flatten(), y.flatten())
+        cost.name = 'cost'
+
+        # Initialization
+        for tbl in embed_tables:
+            tbl.weights_init = config.embed_weights_init
+        mlp.weights_init = config.mlp_weights_init
+        mlp.biases_init = config.mlp_biases_init
+
+        for tbl in embed_tables:
+            tbl.initialize()
+        mlp.initialize()
+
+        self.cost = cost
+        self.monitor = [cost]
+        self.outputs = outputs
+        self.pred_vars = ['travel_time']
diff --git a/train.py b/train.py
index 40449f0..4e2b983 100755
--- a/train.py
+++ b/train.py
@@ -108,7 +108,7 @@ def main():
                 # Checkpoint('model.pkl', every_n_batches=100),
                 Dump('model_data/' + model_name, every_n_batches=1000),
                 LoadFromDump('model_data/' + model_name),
-                FinishAfter(after_epoch=42),
+                # FinishAfter(after_epoch=42),
                 ]
 
     main_loop = MainLoop(
@@ -124,12 +124,12 @@ def main():
 
     outfile = open("output/test-output-%s.csv" % model_name, "w")
     outcsv = csv.writer(outfile)
-    if model.pred_vars == ['time']:
+    if model.pred_vars == ['travel_time']:
         outcsv.writerow(["TRIP_ID", "TRAVEL_TIME"])
         for out in apply_model.Apply(outputs=outputs, stream=test_stream, return_vars=['trip_id', 'outputs']):
             time = out['outputs']
             for i, trip in enumerate(out['trip_id']):
-                outcsv.writerow([trip, int(time[i, 0])])
+                outcsv.writerow([trip, int(time[i])])
     else:
         outcsv.writerow(["TRIP_ID", "LATITUDE", "LONGITUDE"])
         for out in apply_model.Apply(outputs=outputs, stream=test_stream, return_vars=['trip_id', 'outputs']):