Merge pull request #111 from jrzaurin/fix_additive_attention

Fix additive attention (#110)

Author: jrzaurin

VERSION

@@ -1 +1 @@
-1.2.0
+1.2.1

examples/notebooks/10_3rd_party_integration-RayTune_WnB.ipynb

@@ -143,6 +143,7 @@
     "        quantity.\n",
     "\n",
     "    \"\"\"\n",
+    "\n",
     "    def __init__(\n",
     "        self,\n",
     "        wb: object,\n",
@@ -1061,7 +1062,7 @@
     "    \"wandb\": {\n",
     "        \"project\": \"test\",\n",
     "        # \"api_key_file\": os.getcwd() + \"/wandb_api.key\",\n",
-    "        \"api_key\": \"WNB_API_KEY\", \n",
+    "        \"api_key\": \"WNB_API_KEY\",\n",
     "    },\n",
     "}\n",
     "\n",
@@ -1080,7 +1081,12 @@
     "    trainer = Trainer(\n",
     "        model,\n",
     "        objective=\"binary_focal_loss\",\n",
-    "        callbacks=[RayTuneReporter, WnBReportBest(wb=wandb), early_stopping, model_checkpoint],\n",
+    "        callbacks=[\n",
+    "            RayTuneReporter,\n",
+    "            WnBReportBest(wb=wandb),\n",
+    "            early_stopping,\n",
+    "            model_checkpoint,\n",
+    "        ],\n",
     "        lr_schedulers={\"deeptabular\": deep_sch},\n",
     "        initializers={\"deeptabular\": XavierNormal},\n",
     "        optimizers={\"deeptabular\": deep_opt},\n",

mkdocs/sources/examples/10_3rd_party_integration-RayTune_WnB.ipynb

@@ -143,6 +143,7 @@
     "        quantity.\n",
     "\n",
     "    \"\"\"\n",
+    "\n",
     "    def __init__(\n",
     "        self,\n",
     "        wb: object,\n",
@@ -1061,7 +1062,7 @@
     "    \"wandb\": {\n",
     "        \"project\": \"test\",\n",
     "        # \"api_key_file\": os.getcwd() + \"/wandb_api.key\",\n",
-    "        \"api_key\": \"WNB_API_KEY\", \n",
+    "        \"api_key\": \"WNB_API_KEY\",\n",
     "    },\n",
     "}\n",
     "\n",
@@ -1080,7 +1081,12 @@
     "    trainer = Trainer(\n",
     "        model,\n",
     "        objective=\"binary_focal_loss\",\n",
-    "        callbacks=[RayTuneReporter, WnBReportBest(wb=wandb), early_stopping, model_checkpoint],\n",
+    "        callbacks=[\n",
+    "            RayTuneReporter,\n",
+    "            WnBReportBest(wb=wandb),\n",
+    "            early_stopping,\n",
+    "            model_checkpoint,\n",
+    "        ],\n",
     "        lr_schedulers={\"deeptabular\": deep_sch},\n",
     "        initializers={\"deeptabular\": XavierNormal},\n",
     "        optimizers={\"deeptabular\": deep_opt},\n",

pytorch_widedeep/models/tabular/transformers/_attention_layers.py

@@ -217,14 +217,14 @@ def forward(self, X: Tensor) -> Tensor:
         v = self.qv_proj(X) if self.share_qv_weights else self.v_proj(X)
         k = self.k_proj(X)
 
-        alphas = (self.W_q(q) / math.sqrt(self.head_dim)).softmax(dim=-1)
+        alphas = (self.W_q(q) / math.sqrt(self.head_dim)).softmax(dim=1)
         q_r = einops.rearrange(q, "b s (h d) -> b s h d", h=self.n_heads)
         global_query = einsum(" b s h, b s h d -> b h d", alphas, q_r)
         global_query = einops.rearrange(global_query, "b h d -> b () (h d)")
 
         p = k * global_query
 
-        betas = (self.W_k(p) / math.sqrt(self.head_dim)).softmax(dim=-1)
+        betas = (self.W_k(p) / math.sqrt(self.head_dim)).softmax(dim=1)
         p_r = einops.rearrange(p, "b s (h d) -> b s h d", h=self.n_heads)
         global_key = einsum(" b s h, b s h d -> b h d", betas, p_r)
         global_key = einops.rearrange(global_key, "b h d -> b () (h d)")

pytorch_widedeep/version.py

@@ -1 +1 @@
-__version__ = "1.2.0"
+__version__ = "1.2.1"