Feature/fullbatch linkgenerator (stellargraph#543)

* Added GCN link model and added new full-batch link generator * Fixed bug in full-batch generators. Updated unit tests. Updated full-batch generators to return adjacency matrix in same node order as the features in the StellarGraph object. This fixes a bug that was introduced by using the 'get_index_for_node` method. * Added LinkEmbedding layer. * Updated all full-batch classes to use `build` All full-batch classes (GCN, GAT, APPNP, PPNP) now support creating a link model * Updated demos to use `build` rather than `node_model` or `link_model` * Formatted GCN link prediction notebook. * Removed erroneous format_notebook script * Fixed bug in stellargraph using deprecated `.node` method * Updated docstrings and CHANGELOG.md * Bugfix for issue introduced in previous commit * Removed Dense layer from LinkEmbedding layer and moved to associated function. Cleaned up link tests. Misc spelling errors. * Addressed review comments: updated notebooks to match new full-batch node order, raise error if FullBatchGenerator is instanciated. * Added docstring describing axis to LinkEmbedding and removed argument from link_inference. * Spelling errors fixed. * Fixed bug in stellargraph to_adjacency_matrix when nodes are specified * Fixed bug in unit test. Removed erroneous spaces. Updated copyright header. * More spaces removed * Updated documentation Co-authored-by: Yuriy Tyshetskiy <youph@users.noreply.github.com>
tatras-ocelot · Jan 23, 2020 · bcf6d0a · bcf6d0a
1 parent 10e6200
commit bcf6d0a
Show file tree

Hide file tree

Showing 41 changed files with 7,113 additions and 1,714 deletions.
diff --git a/demos/ensembles/ensemble-node-classification-example.ipynb b/demos/ensembles/ensemble-node-classification-example.ipynb
diff --git a/demos/interpretability/gat/node-link-importance-demo-gat.ipynb b/demos/interpretability/gat/node-link-importance-demo-gat.ipynb
@@ -353,8 +353,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# Expose the input and output tensors of the GAT model for node prediction, via GAT.node_model() method:\n",
-    "x_inp, predictions = gat.node_model()"
+    "# Expose the input and output tensors of the GAT model for node prediction, via GAT.build() method:\n",
+    "x_inp, predictions = gat.build()"
    ]
   },
   {

diff --git a/demos/interpretability/gcn/hateful-twitters-interpretability.ipynb b/demos/interpretability/gcn/hateful-twitters-interpretability.ipynb
@@ -679,7 +679,7 @@
     "    dropout=0.5,\n",
     "    activations=[\"elu\", \"elu\"]\n",
     ")\n",
-    "x_inp, x_out = base_model.node_model()\n",
+    "x_inp, x_out = base_model.build()\n",
     "prediction = layers.Dense(units=1, activation=\"sigmoid\")(x_out)"
    ]
   },

diff --git a/demos/interpretability/gcn/node-link-importance-demo-gcn-sparse.ipynb b/demos/interpretability/gcn/node-link-importance-demo-gcn-sparse.ipynb
@@ -266,8 +266,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# Expose the input and output tensors of the GCN model for node prediction, via GCN.node_model() method:\n",
-    "x_inp, x_out = gcn.node_model()\n",
+    "# Expose the input and output tensors of the GCN model for node prediction, via GCN.build() method:\n",
+    "x_inp, x_out = gcn.build()\n",
     "# Snap the final estimator layer to x_out\n",
     "x_out = layers.Dense(units=train_targets.shape[1], activation=\"softmax\")(x_out)"
    ]

diff --git a/demos/interpretability/gcn/node-link-importance-demo-gcn.ipynb b/demos/interpretability/gcn/node-link-importance-demo-gcn.ipynb
@@ -268,8 +268,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# Expose the input and output tensors of the GCN model for node prediction, via GCN.node_model() method:\n",
-    "x_inp, x_out = gcn.node_model()\n",
+    "# Expose the input and output tensors of the GCN model for node prediction, via GCN.build() method:\n",
+    "x_inp, x_out = gcn.build()\n",
     "# Snap the final estimator layer to x_out\n",
     "x_out = layers.Dense(units=train_targets.shape[1], activation=\"softmax\")(x_out)"
    ]