.

harvardnlp · Oct 28, 2019 · 31cfe95 · 31cfe95
1 parent 50c9284
commit 31cfe95
Show file tree

Hide file tree

Showing 4 changed files with 60 additions and 16 deletions.
diff --git a/docs/source/model.ipynb b/docs/source/model.ipynb
@@ -147,13 +147,6 @@
     ".. autoclass:: torch_struct.HMM"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": []
-  },
   {
    "cell_type": "code",
    "execution_count": 6,
@@ -651,7 +644,10 @@
     "raw_mimetype": "text/restructuredtext"
    },
    "source": [
-    "    .. autoclass:: torch_struct.Autoregressive"
+    "    .. autoclass:: torch_struct.Autoregressive\n",
+    "         :members: sample,sample_without_replacement,greedy_max,beam_topk\n",
+    "    .. autoclass:: torch_struct.AutoregressiveModel\n",
+    "    "
    ]
   },
   {

diff --git a/docs/source/refs.bib b/docs/source/refs.bib
@@ -92,7 +92,7 @@ @incollection{eisner2000bilexical
   publisher={Springer}
 }
 
-    @inproceedings{eisner2002parameter,
+@inproceedings{eisner2002parameter,
   title={Parameter estimation for probabilistic finite-state transducers},
   author={Eisner, Jason},
   booktitle={Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics},

diff --git a/docs/source/semiring.ipynb b/docs/source/semiring.ipynb
@@ -97,6 +97,13 @@
     "    show(m.sum(-1).sum(0).sum(0).transpose(0,1), \"Alignment example\")"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Log"
+   ]
+  },
   {
    "cell_type": "raw",
    "metadata": {
@@ -164,6 +171,13 @@
     "run(torch_struct.LogSemiring)"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Max"
+   ]
+  },
   {
    "cell_type": "raw",
    "metadata": {
@@ -231,6 +245,13 @@
     "run(torch_struct.MaxSemiring)"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## K-Max"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": 6,
@@ -289,6 +310,13 @@
     "run(torch_struct.KMaxSemiring(3))"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Counting"
+   ]
+  },
   {
    "cell_type": "raw",
    "metadata": {
@@ -356,6 +384,13 @@
     "run(torch_struct.StdSemiring)"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Sampled"
+   ]
+  },
   {
    "cell_type": "raw",
    "metadata": {
@@ -490,6 +525,13 @@
     "run(torch_struct.MultiSampledSemiring)"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Entropy"
+   ]
+  },
   {
    "cell_type": "raw",
    "metadata": {
@@ -557,6 +599,13 @@
     "run(torch_struct.EntropySemiring)"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Sparsemax"
+   ]
+  },
   {
    "cell_type": "raw",
    "metadata": {
@@ -643,7 +692,6 @@
     }
    ],
    "source": [
-    "\n",
     "all_sr = [(\"marginals\", torch_struct.LogSemiring), \n",
     "          (\"argmax\", torch_struct.MaxSemiring),\n",
     "          (\"sample\", torch_struct.SampledSemiring),\n",

diff --git a/torch_struct/distributions.py b/torch_struct/distributions.py
@@ -194,7 +194,7 @@ class LinearChainCRF(StructDistribution):
 
     Implementation uses linear-scan, forward-pass only.
 
-    * Parallel Time: `O(\log(N))` parallel merges.
+    * Parallel Time: :math:`O(\log(N))` parallel merges.
     * Forward Memory: :math:`O(N \log(N) C^2)`
 
     """
@@ -217,7 +217,7 @@ class AlignmentCRF(StructDistribution):
 
     Implementation uses linear-scan.
 
-    * Parallel Time: `O(\log (M +N))` parallel merges.
+    * Parallel Time: :math:`O(\log (M +N))` parallel merges.
     * Forward Memory: :math:`O((M+N)^3)`
 
     """
@@ -263,7 +263,7 @@ class SemiMarkovCRF(StructDistribution):
 
     Implementation uses linear-scan, forward-pass only.
 
-    * Parallel Time: `O(\log(N))` parallel merges.
+    * Parallel Time: :math:`O(\log(N))` parallel merges.
     * Forward Memory: :math:`O(N \log(N) C^2 K^2)`
 
     """
@@ -292,7 +292,7 @@ class DependencyCRF(StructDistribution):
 
     Implementation uses linear-scan, forward-pass only.
 
-    * Parallel Time: `O(N)` parallel merges.
+    * Parallel Time: :math:`O(N)` parallel merges.
     * Forward Memory: :math:`O(N \log(N) C^2 K^2)`
 
     """
@@ -318,7 +318,7 @@ class TreeCRF(StructDistribution):
 
     Implementation uses width-batched, forward-pass only
 
-    * Parallel Time: `O(N)` parallel merges.
+    * Parallel Time: :math:`O(N)` parallel merges.
     * Forward Memory: :math:`O(N^2)`
 
     Compact representation:  *N x N x NT* long tensor (Same)
@@ -342,7 +342,7 @@ class SentCFG(StructDistribution):
 
     Implementation uses width-batched, forward-pass only
 
-    * Parallel Time: `O(N)` parallel merges.
+    * Parallel Time: :math:`O(N)` parallel merges.
     * Forward Memory: :math:`O(N^2 (NT+T))`
 
     Compact representation:  *N x N x NT* long tensor