dynamic_pca_mwe.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "334c32bd",
   "metadata": {},
   "source": [
    "# Minimal working example of a dynamic PCA Layer for TensorFlow"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "217b081a",
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "class PCALayer(tf.keras.layers.Layer):\n",
    "    n_components : int = None   # No dimension reduction with None\n",
    "    \n",
    "    def __init__(self, inverse_layer : bool, **options):\n",
    "        self.inverse_layer = inverse_layer\n",
    "        self._fitted = False\n",
    "        options['dynamic']   = True  # Is dynamic\n",
    "        options['trainable'] = False # but not trainable\n",
    "        super().__init__(**options)  # further Layer init\n",
    "    \n",
    "    def fit(self, XTrain : tf.Tensor) -> None:\n",
    "        \"\"\"\n",
    "        Calculates the SVD, afterwards the PCA can be applied\n",
    "        \"\"\"\n",
    "        self.mean = tf.reduce_mean(XTrain, axis=0)\n",
    "        XTrain = tf.subtract(XTrain, self.mean)     # Centering of data\n",
    "        # Sidenote, tf, numpy and scipys svds can be different\n",
    "        S, U, V = tf.linalg.svd(XTrain, full_matrices=False)\n",
    "        V = tf.transpose(V)\n",
    "        # Store singular values and transformation matrix V\n",
    "        self.S, self.V = S, V  # TODO: Should be done as tf weights\n",
    "        self._fitted = True       \n",
    "    \n",
    "    def transform(self, X : tf.Tensor) -> tf.Tensor:\n",
    "        \"\"\"\n",
    "        Applies the PCA transformation with choosen dimension\n",
    "        which is set by the attribute n_components.\n",
    "        Output shape: (X.shape[0], self.n_components)\n",
    "        \"\"\"\n",
    "        X = tf.subtract(X, self.mean) # Center data\n",
    "        # Transform with submatrix of V to choosen dimension\n",
    "        Xk = tf.matmul(X, tf.transpose(self.V[:self.n_components, :]))\n",
    "        return Xk\n",
    "        \n",
    "    def inverse_transform(self, Xk : tf.Tensor) -> tf.Tensor:\n",
    "        \"\"\"\n",
    "        Reverse the PCA transformation from Xk with dimension k\n",
    "        to original dimension n.\n",
    "        Output shape: (Xk.shape[0], self.S.shape[0])\n",
    "        \"\"\"\n",
    "        X = tf.matmul(Xk, self.V[:Xk.shape[-1], :]) # Always returns to full dimension of V\n",
    "        X = X + self.mean                           # Reverse the centering\n",
    "        return X\n",
    "        \n",
    "    def call(self, input : tf.Tensor) -> tf.Tensor:\n",
    "        \"\"\"\n",
    "        This function is called when data passes the layer\n",
    "        \"\"\"\n",
    "        if not self._fitted :   # No action if not fitted\n",
    "            return input\n",
    "        if self.inverse_layer:  # Reverse transformation\n",
    "            return self.inverse_transform(input)\n",
    "        return self.transform(input) # reduce dimension with PCA\n",
    "    \n",
    "    def compute_output_shape(self, input_shape : tf.Tensor) -> tf.Tensor:\n",
    "        \"\"\"\n",
    "        This method is not needed if TensorFlow runs eagerly\n",
    "        \"\"\"\n",
    "        if not self._fitted:\n",
    "            return input_shape\n",
    "        if not self.inverse_layer:\n",
    "            return tf.TensorShape([input_shape[0], self.n_components or input_shape[1:]])\n",
    "        return tf.TensorShape([None, self.S.shape[0]])\n",
    "\n",
    "        "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cc6d5a6e",
   "metadata": {},
   "source": [
    "# Possible usage\n",
    "\n",
    "## Autoencoder setup and first training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "8c260ffc",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/2\n",
      "469/469 [==============================] - 41s 86ms/step - loss: 0.1382 - val_loss: 0.0843\n",
      "Epoch 2/2\n",
      "469/469 [==============================] - 40s 85ms/step - loss: 0.0786 - val_loss: 0.0752\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.keras.models import Sequential\n",
    "from tensorflow.keras.layers import Dense, Flatten, Reshape, Conv2D, BatchNormalization, UpSampling2D, Input\n",
    "\n",
    "# Load and scale data\n",
    "(XTrain, YTrain), (XTest, YTest) = tf.keras.datasets.mnist.load_data()\n",
    "input_shape=XTrain.shape[1:] # 28x28 here\n",
    "XTrain = XTrain.astype('float32') / 255.\n",
    "XTest = XTest.astype('float32') / 255.\n",
    "\n",
    "# Define PCA Layers\n",
    "pca_transform_layer = PCALayer(inverse_layer=False)\n",
    "pca_inverse_layer = PCALayer(inverse_layer=True)\n",
    "\n",
    "# Define encoder and decoder\n",
    "encoder = Sequential(layers=[\n",
    "    Reshape((-1, 28,28,1)),\n",
    "    BatchNormalization(),\n",
    "    Conv2D(128, kernel_size=(3,3), strides=(2,2), activation=\"relu\"),\n",
    "    Conv2D(64, kernel_size=(3,3), strides=(2,2), activation=\"relu\"),\n",
    "    Flatten(),\n",
    "    Dense(256, activation=\"relu\"),\n",
    "    Dense(128),\n",
    "    BatchNormalization(),\n",
    "    pca_transform_layer\n",
    "    ])\n",
    "\n",
    "decoder = Sequential(layers=[\n",
    "    pca_inverse_layer,\n",
    "    Dense(7**3, activation=\"relu\"),\n",
    "    Reshape((7, 7, 7)),\n",
    "    BatchNormalization(),\n",
    "    Conv2D(64, kernel_size=(3,3), strides=(1,1), activation=\"relu\", padding='same'),\n",
    "    UpSampling2D(),\n",
    "    Conv2D(128, kernel_size=(3,3), strides=(1,1), activation=\"relu\", padding='same'),\n",
    "    UpSampling2D(),\n",
    "    BatchNormalization(),\n",
    "    Conv2D(128, kernel_size=(3,3), strides=(1,1), activation=\"relu\", padding='same'),\n",
    "    Conv2D(1, kernel_size=(3,3), strides=(1,1), activation=\"sigmoid\", padding='same'),\n",
    "    Reshape((28, 28))\n",
    "    ])\n",
    "\n",
    "# Define and train autoencoder\n",
    "model = Sequential(layers=[encoder, decoder])\n",
    "model.compile(optimizer='adam', loss='binary_crossentropy')\n",
    "\n",
    "# %%% Training\n",
    "model.fit(XTrain, XTrain, batch_size=128, epochs=2,\n",
    "         validation_data=(XTest, XTest)) \n",
    "\n",
    "# Encode Training data for PCA\n",
    "encoded = encoder.predict(XTrain)    # <--- Note the memory usage"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "patent-correlation",
   "metadata": {},
   "source": [
    "## Output sample"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "thrown-civilian",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "plt.title(\"Original\")\n",
    "plt.imshow(XTest[:1][0], cmap='gray')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "wireless-basics",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.title(\"Result 1st Training - Latent dim 128\")\n",
    "plt.imshow(model.predict(XTest[:1])[0], cmap='gray')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "editorial-satin",
   "metadata": {},
   "source": [
    "## Apply PCA and define dynamic latent dimension"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "fcac5049",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Fit PCA\n",
    "# Note rerunning this cell yields not the same result\n",
    "# as the pca_layer is part of the encoder\n",
    "encoded = encoder.predict(XTrain)  # <--- Encode the data, Note the memory usage\n",
    "pca_transform_layer.fit(encoded)\n",
    "pca_inverse_layer.fit(encoded)    # <--- Here you see the problem of this mwe. pca needs to be done twice\n",
    "del encoded\n",
    "pca_transform_layer.n_components = None # <-- Full latent shape\n",
    "\n",
    "plt.title(\"Result with PCA - Latent dim 128\")\n",
    "plt.imshow(model.predict(XTest[:1])[0], cmap='gray')\n",
    "plt.show()\n",
    "\n",
    "pca_transform_layer.n_components = 32 # <-- Define a custom latent layer size\n",
    "\n",
    "plt.title(\"Result with PCA - Latent dim 32\")\n",
    "plt.imshow(model.predict(XTest[:1])[0], cmap='gray')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "framed-person",
   "metadata": {},
   "source": [
    "## Possible retraining to improve results at a lower dimension"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "occupational-saudi",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/2\n",
      "469/469 [==============================] - 40s 86ms/step - loss: 0.0802 - val_loss: 0.0808\n",
      "Epoch 2/2\n",
      "469/469 [==============================] - 42s 89ms/step - loss: 0.0775 - val_loss: 0.0760\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Possible Retraining\n",
    "model.fit(XTrain, XTrain, batch_size=128, epochs=2, \n",
    "          validation_data=(XTest, XTest)) # <--- Adjust with appropriate data\n",
    "\n",
    "plt.title(\"Result Retraining - Latent dim 32\")\n",
    "plt.imshow(model.predict(XTest[:1])[0], cmap='gray')\n",
    "plt.show()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.12"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}