Add a few images as it is done in other notebooks

Signed-off-by: Guillaume Tauzin <guillaumetauzin.ut@gmail.com>

examples/MNIST_classification.ipynb

@@ -31,19 +31,16 @@
    },
    "outputs": [],
    "source": [
-    "from gtda.images import Binarizer, Inverter, ImageToPointCloud, HeightFiltration, DilationFiltration, RadialFiltration, ErosionFiltration, SignedDistanceFiltration\n",
+    "from gtda.images import Binarizer, Inverter, HeightFiltration, DilationFiltration, RadialFiltration, ErosionFiltration, SignedDistanceFiltration\n",
     "from gtda.homology import CubicalPersistence\n",
     "from gtda.diagrams import ForgetDimension, Amplitude, Scaler, PersistenceEntropy, BettiCurve, PersistenceLandscape, HeatKernel\n",
     "from gtda.plotting import plot_heatmap, plot_betti_curves, plot_diagram\n",
     "from sklearn.pipeline import Pipeline, make_pipeline, FeatureUnion, make_union\n",
     "from sklearn.ensemble import RandomForestClassifier\n",
     "from sklearn.model_selection import GridSearchCV\n",
     "from sklearn.metrics import classification_report\n",
-    "import numpy as np\n",
-    "\n",
-    "# This would be quite nice but is available with sklearn >= 0.23\n",
-    "# from sklearn import set_config\n",
-    "#set_config(display='diagram') "
+    "from sklearn import set_config\n",
+    "set_config(display='diagram') "
    ]
   },
   {
@@ -152,7 +149,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "inverter = Inverter(n_jobs=4)\n",
+    "inverter = Inverter(n_jobs=-1)\n",
     "X_train_inverted = inverter.fit_transform(X_train_binarized)"
    ]
   },
@@ -322,7 +319,11 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "## Building a pipeline to extract features"
+    "## Building a pipeline to extract features\n",
+    "\n",
+    "We can summarize all the previous step into a single giotto Pipeline pictured below:\n",
+    "\n",
+    "![example_pipeline_images](images/example_pipeline_images.png \"TDA pipeline for MNIST images.\")"
    ]
   },
   {
@@ -333,8 +334,9 @@
    "source": [
     "steps = [\n",
     "    ('binarizer', Binarizer(threshold=0.4)),\n",
-    "    ('filtration', SignedDistanceFiltration(n_iterations=28), n_jobs=-1),\n",
+    "    ('filtration', RadialFiltration(center=np.array([20, 6])), n_jobs=-1),\n",
     "    ('diagram', CubicalPersistence(n_jobs=-1)),\n",
+    "    ('rescaling'), Scaler(),\n",
     "    ('amplitude', Amplitude(metric='wasserstein', metric_params={'p': 2}, n_jobs=-1))\n",
     "]\n",
     "\n",
@@ -400,7 +402,12 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "We can go full-scale and extract a large number of features. Careful, some of them will be highly correlated, so it would be good to use a feature selection algorithm to reduce their number before passing them to a classifier.\n",
+    "We can go full-scale and extract a large number of features by creating a full-blown pipeline such as the one below.\n",
+    "\n",
+    "![diagram_pipeline_images](images/diagram_pipeline_images.png \"Full-blown TDA pipeline for MNIST images.\")\n",
+    "\n",
+    "\n",
+    "Careful, some of them will be highly correlated, so it would be good to use a feature selection algorithm to reduce their number before passing them to a classifier.\n",
     "\n",
     "Note that if you have selected a large number of training/testing samples, this part will take a lot of resources to run. Consider using a cluster."
    ]