Merge pull request #62 from maciejkula/goodbooks

Add goodbooks dataset

docs/changelog.rst

@@ -2,6 +2,14 @@
 Changelog
 =========
 
+unreleased (unreleased)
+-----------------------
+
+Added
+~~~~~
+
+* Goodbooks dataset.
+
 v0.1.2 (2017-09-10)
 -------------------
 

docs/datasets/datasets.rst

@@ -10,3 +10,4 @@ Datasets
 
    Synthetic <synthetic>
    Movielens <movielens>
+   Goodbooks <goodbooks>

docs/datasets/goodbooks.rst

@@ -0,0 +1,7 @@
+Goodbooks
+=========
+
+
+.. automodule:: spotlight.datasets.goodbooks
+   :members:
+   :undoc-members:

spotlight/datasets/goodbooks.py

@@ -0,0 +1,48 @@
+"""
+Utilities for fetching the Goodbooks-10K dataset [1]_.
+
+References
+----------
+
+.. [1] https://github.com/zygmuntz/goodbooks-10k
+"""
+
+import h5py
+
+import numpy as np
+
+from spotlight.datasets import _transport
+from spotlight.interactions import Interactions
+
+
+def _get_dataset():
+
+    path = _transport.get_data('https://github.com/zygmuntz/goodbooks-10k/'
+                               'releases/download/v1.0/goodbooks-10k.hdf5',
+                               'goodbooks',
+                               'goodbooks.hdf5')
+
+    with h5py.File(path, 'r') as data:
+        return (data['ratings'][:, 0],
+                data['ratings'][:, 1],
+                data['ratings'][:, 2],
+                np.arange(len(data['ratings']), dtype=np.int32))
+
+
+def get_goodbooks_dataset():
+    """
+    Download and return the goodbooks-10K dataset [2]_.
+
+    Returns
+    -------
+
+    Interactions: :class:`spotlight.interactions.Interactions`
+        instance of the interactions class
+
+    References
+    ----------
+
+    .. [2] https://github.com/zygmuntz/goodbooks-10k
+    """
+
+    return Interactions(*_get_dataset())

spotlight/interactions.py

@@ -299,3 +299,14 @@ def __init__(self,
             self.num_items = sequences.max() + 1
         else:
             self.num_items = num_items
+
+    def __repr__(self):
+
+        num_sequences, sequence_length = self.sequences.shape
+
+        return ('<Sequence interactions dataset ({num_sequences} '
+                'sequences x {sequence_length} sequence length)>'
+                .format(
+                    num_sequences=num_sequences,
+                    sequence_length=sequence_length,
+                ))

spotlight/sequence/implicit.py

@@ -269,9 +269,10 @@ def _get_negative_prediction(self, shape, user_representation):
     def _get_multiple_negative_predictions(self, shape, user_representation,
                                            n=5):
         batch_size, sliding_window = shape
+        size = (n,) + (1,) * (user_representation.dim() - 1)
         negative_prediction = self._get_negative_prediction(
             (n * batch_size, sliding_window),
-            user_representation.repeat(n, 1, 1))
+            user_representation.repeat(*size))
 
         return negative_prediction.view(n, batch_size, sliding_window)
 
@@ -314,7 +315,8 @@ def predict(self, sequences, item_ids=None):
         item_var = Variable(gpu(item_ids, self._use_cuda))
 
         _, sequence_representations = self._net.user_representation(sequence_var)
-        out = self._net(sequence_representations.repeat(len(item_var), 1),
+        size = (len(item_var),) + sequence_representations.size()[1:]
+        out = self._net(sequence_representations.expand(*size),
                         item_var)
 
         return cpu(out.data).numpy().flatten()