add space and model hub for molecule generation with VAE example (ker…

…as-team#974)
satpalsr · Jul 7, 2022 · a01a0e3 · a01a0e3
1 parent 1382cca
commit a01a0e3
Show file tree

Hide file tree

Showing 3 changed files with 28 additions and 15 deletions.
diff --git a/examples/generative/ipynb/molecule_generation.ipynb b/examples/generative/ipynb/molecule_generation.ipynb
@@ -127,10 +127,10 @@
    "source": [
     "## Dataset\n",
     "\n",
-    "We use the [**ZINC \u2013 A Free Database of Commercially Available Compounds for\n",
+    "We use the [**ZINC – A Free Database of Commercially Available Compounds for\n",
     "Virtual Screening**](https://bit.ly/3IVBI4x) dataset. The dataset comes with molecule\n",
     "formula in SMILE representation along with their respective molecular properties such as\n",
-    "**logP** (water\u2013octanal partition coefficient), **SAS** (synthetic\n",
+    "**logP** (water–octanal partition coefficient), **SAS** (synthetic\n",
     "accessibility score) and **QED** (Qualitative Estimate of Drug-likeness)."
    ]
   },
@@ -260,8 +260,7 @@
     "    if flag != Chem.SanitizeFlags.SANITIZE_NONE:\n",
     "        return None\n",
     "\n",
-    "    return molecule\n",
-    ""
+    "    return molecule\n"
    ]
   },
   {
@@ -355,8 +354,7 @@
     "        # Reduce bond types dim\n",
     "        x_reduced = tf.reduce_sum(x, axis=1)\n",
     "        # Apply non-linear transformation\n",
-    "        return self.activation(x_reduced)\n",
-    ""
+    "        return self.activation(x_reduced)\n"
    ]
   },
   {
@@ -376,9 +374,9 @@
     "non-linearly transformed neighbourhood aggregations. We can define these layers as\n",
     "follows:\n",
     "\n",
-    "`H_hat**(l+1) = \u03c3(D_hat**(-1) * A_hat * H_hat**(l+1) * W**(l))`\n",
+    "`H_hat**(l+1) = σ(D_hat**(-1) * A_hat * H_hat**(l+1) * W**(l))`\n",
     "\n",
-    "Where `\u03c3` denotes the non-linear transformation (commonly a ReLU activation), `A` the\n",
+    "Where `σ` denotes the non-linear transformation (commonly a ReLU activation), `A` the\n",
     "adjacency tensor, `H_hat**(l)` the feature tensor at the `l-th` layer, `D_hat**(-1)` the\n",
     "inverse diagonal degree tensor of `A_hat`, and `W_hat**(l)` the trainable weight tensor\n",
     "at the `l-th` layer. Specifically, for each bond type (relation), the degree tensor\n",
@@ -452,8 +450,7 @@
     "        latent_inputs, outputs=[x_adjacency, x_features], name=\"decoder\"\n",
     "    )\n",
     "\n",
-    "    return decoder\n",
-    ""
+    "    return decoder\n"
    ]
   },
   {
@@ -480,8 +477,7 @@
     "        batch = tf.shape(z_log_var)[0]\n",
     "        dim = tf.shape(z_log_var)[1]\n",
     "        epsilon = tf.keras.backend.random_normal(shape=(batch, dim))\n",
-    "        return z_mean + tf.exp(0.5 * z_log_var) * epsilon\n",
-    ""
+    "        return z_mean + tf.exp(0.5 * z_log_var) * epsilon\n"
    ]
   },
   {
@@ -639,8 +635,7 @@
     "\n",
     "        property_pred = self.property_prediction_layer(z_mean)\n",
     "\n",
-    "        return z_mean, log_var, property_pred, gen_adjacency, gen_features\n",
-    ""
+    "        return z_mean, log_var, property_pred, gen_adjacency, gen_features\n"
    ]
   },
   {
@@ -766,7 +761,13 @@
     "while the later paper considers SMILES inputs as graphs (a combination of adjacency\n",
     "matrices and feature matrices) and seeks to generate molecules as graphs.\n",
     "\n",
-    "This hybrid approach enables a new type of directed gradient-based search through chemical space."
+    "This hybrid approach enables a new type of directed gradient-based search through chemical space.\n",
+    "\n",
+    "Example available on HuggingFace\n",
+    "\n",
+    "| Trained Model | Demo |\n",
+    "| :--: | :--: |\n",
+    "| [![Generic badge](https://img.shields.io/badge/%F0%9F%A4%97%20Model-molecule%20generation%20with%20VAE-black.svg)](https://huggingface.co/keras-io/drug-molecule-generation-with-VAE) | [![Generic badge](https://img.shields.io/badge/%F0%9F%A4%97%20Spaces-molecule%20generation%20with%20VAE-black.svg)](https://huggingface.co/spaces/keras-io/generating-drug-molecule-with-VAE) |"
    ]
   }
  ],

diff --git a/examples/generative/md/molecule_generation.md b/examples/generative/md/molecule_generation.md
@@ -737,3 +737,9 @@ while the later paper considers SMILES inputs as graphs (a combination of adjace
 matrices and feature matrices) and seeks to generate molecules as graphs.
 
 This hybrid approach enables a new type of directed gradient-based search through chemical space.
+
+Example available on HuggingFace
+
+| Trained Model | Demo |
+| :--: | :--: |
+| [![Generic badge](https://img.shields.io/badge/%F0%9F%A4%97%20Model-molecule%20generation%20with%20VAE-black.svg)](https://huggingface.co/keras-io/drug-molecule-generation-with-VAE) | [![Generic badge](https://img.shields.io/badge/%F0%9F%A4%97%20Spaces-molecule%20generation%20with%20VAE-black.svg)](https://huggingface.co/spaces/keras-io/generating-drug-molecule-with-VAE) |
diff --git a/examples/generative/molecule_generation.py b/examples/generative/molecule_generation.py
@@ -597,4 +597,10 @@ def plot_latent(vae, data, labels):
 matrices and feature matrices) and seeks to generate molecules as graphs.
 
 This hybrid approach enables a new type of directed gradient-based search through chemical space.
+
+Example available on HuggingFace
+
+| Trained Model | Demo |
+| :--: | :--: |
+| [![Generic badge](https://img.shields.io/badge/%F0%9F%A4%97%20Model-molecule%20generation%20with%20VAE-black.svg)](https://huggingface.co/keras-io/drug-molecule-generation-with-VAE) | [![Generic badge](https://img.shields.io/badge/%F0%9F%A4%97%20Spaces-molecule%20generation%20with%20VAE-black.svg)](https://huggingface.co/spaces/keras-io/generating-drug-molecule-with-VAE) |
 """