Add the ability to generate TF-format guides (keras-team#87)

* Add the ability to generate TF-format guides

* Add link conversion to tf notebooks

* Use fully-qualified links

guides/ipynb/customizing_what_happens_in_fit.ipynb

@@ -47,7 +47,7 @@
     "API. You can do this whether you're building `Sequential` models, Functional API\n",
     "models, or subclassed models.\n",
     "\n",
-    "Let's see how that works.\n"
+    "Let's see how that works."
    ]
   },
   {
@@ -56,7 +56,8 @@
     "colab_type": "text"
    },
    "source": [
-    "## Setup\n"
+    "## Setup\n",
+    "Requires TensorFlow 2.2 or later."
    ]
   },
   {
@@ -68,7 +69,7 @@
    "outputs": [],
    "source": [
     "import tensorflow as tf\n",
-    "from tensorflow import keras\n"
+    "from tensorflow import keras"
    ]
   },
   {
@@ -100,7 +101,7 @@
     "\n",
     "Similarly, we call `self.compiled_metrics.update_state(y, y_pred)` to update the state\n",
     "of the metrics that were passed in `compile()`, and we query results from\n",
-    "`self.metrics` at the end to retrieve their current value.\n"
+    "`self.metrics` at the end to retrieve their current value."
    ]
   },
   {
@@ -133,7 +134,7 @@
     "        self.compiled_metrics.update_state(y, y_pred)\n",
     "        # Return a dict mapping metric names to current value\n",
     "        return {m.name: m.result() for m in self.metrics}\n",
-    "\n"
+    ""
    ]
   },
   {
@@ -142,7 +143,7 @@
     "colab_type": "text"
    },
    "source": [
-    "Let's try this out:\n"
+    "Let's try this out:"
    ]
   },
   {
@@ -164,7 +165,7 @@
     "# Just use `fit` as usual\n",
     "x = np.random.random((1000, 32))\n",
     "y = np.random.random((1000, 1))\n",
-    "model.fit(x, y, epochs=3)\n"
+    "model.fit(x, y, epochs=3)"
    ]
   },
   {
@@ -177,7 +178,7 @@
     "\n",
     "Naturally, you could just skip passing a loss function in `compile()`, and instead do\n",
     "everything *manually* in `train_step`. Likewise for metrics. Here's a lower-level\n",
-    "example, that only uses `compile()` to configure the optimizer:\n"
+    "example, that only uses `compile()` to configure the optimizer:"
    ]
   },
   {
@@ -225,7 +226,7 @@
     "# Just use `fit` as usual -- you can use callbacks, etc.\n",
     "x = np.random.random((1000, 32))\n",
     "y = np.random.random((1000, 1))\n",
-    "model.fit(x, y, epochs=3)\n"
+    "model.fit(x, y, epochs=3)"
    ]
   },
   {
@@ -243,7 +244,7 @@
     "- Unpack `sample_weight` from the `data` argument\n",
     "- Pass it to `compiled_loss` & `compiled_metrics` (of course, you could also just apply\n",
     "it manually if you don't rely on `compile()` for losses & metrics)\n",
-    "- That's it. That's the list.\n"
+    "- That's it. That's the list."
    ]
   },
   {
@@ -301,7 +302,7 @@
     "x = np.random.random((1000, 32))\n",
     "y = np.random.random((1000, 1))\n",
     "sw = np.random.random((1000, 1))\n",
-    "model.fit(x, y, sample_weight=sw, epochs=3)\n"
+    "model.fit(x, y, sample_weight=sw, epochs=3)"
    ]
   },
   {
@@ -313,7 +314,7 @@
     "## Providing your own evaluation step\n",
     "\n",
     "What if you want to do the same for calls to `model.evaluate()`? Then you would\n",
-    "override `test_step` in exactly the same way. Here's what it looks like:\n"
+    "override `test_step` in exactly the same way. Here's what it looks like:"
    ]
   },
   {
@@ -349,7 +350,7 @@
     "# Evaluate with our custom test_step\n",
     "x = np.random.random((1000, 32))\n",
     "y = np.random.random((1000, 1))\n",
-    "model.evaluate(x, y)\n"
+    "model.evaluate(x, y)"
    ]
   },
   {
@@ -369,7 +370,7 @@
     "\"real\").\n",
     "- One optimizer for each.\n",
     "- A loss function to train the discriminator.\n",
-    "\n"
+    ""
    ]
   },
   {
@@ -412,7 +413,7 @@
     "        layers.Conv2D(1, (7, 7), padding=\"same\", activation=\"sigmoid\"),\n",
     "    ],\n",
     "    name=\"generator\",\n",
-    ")\n"
+    ")"
    ]
   },
   {
@@ -422,7 +423,7 @@
    },
    "source": [
     "Here's a feature-complete GAN class, overriding `compile()` to use its own signature,\n",
-    "and implementing the entire GAN algorithm in 17 lines in `train_step`:\n"
+    "and implementing the entire GAN algorithm in 17 lines in `train_step`:"
    ]
   },
   {
@@ -490,7 +491,7 @@
     "        grads = tape.gradient(g_loss, self.generator.trainable_weights)\n",
     "        self.g_optimizer.apply_gradients(zip(grads, self.generator.trainable_weights))\n",
     "        return {\"d_loss\": d_loss, \"g_loss\": g_loss}\n",
-    "\n"
+    ""
    ]
   },
   {
@@ -499,7 +500,7 @@
     "colab_type": "text"
    },
    "source": [
-    "Let's test-drive it:\n"
+    "Let's test-drive it:"
    ]
   },
   {
@@ -528,7 +529,7 @@
     "\n",
     "# To limit execution time, we only train on 100 batches. You can train on\n",
     "# the entire dataset. You will need about 20 epochs to get nice results.\n",
-    "gan.fit(dataset.take(100), epochs=1)\n"
+    "gan.fit(dataset.take(100), epochs=1)"
    ]
   },
   {
@@ -537,7 +538,7 @@
     "colab_type": "text"
    },
    "source": [
-    "The idea behind deep learning are simple, so why should their implementation be painful?\n"
+    "The idea behind deep learning are simple, so why should their implementation be painful?"
    ]
   }
  ],