Merge pull request #531 from neurodata/shakespeare

ENH & DOC update algorithm descriptions

docs/experiments/fte_bte_aircraft_bird.ipynb

@@ -2,7 +2,6 @@
  "cells": [
   {
    "cell_type": "markdown",
-   "id": "three-dakota",
    "metadata": {},
    "source": [
     "# FTE/BTE Experiment for Aircraft & Birdsnap\n",
@@ -24,7 +23,6 @@
   {
    "cell_type": "code",
    "execution_count": 1,
-   "id": "funny-harvest",
    "metadata": {},
    "outputs": [],
    "source": [
@@ -34,7 +32,6 @@
   },
   {
    "cell_type": "markdown",
-   "id": "tropical-approval",
    "metadata": {},
    "source": [
     "### Load tasks\n",
@@ -51,7 +48,6 @@
   {
    "cell_type": "code",
    "execution_count": 2,
-   "id": "engaged-darwin",
    "metadata": {},
    "outputs": [],
    "source": [
@@ -73,7 +69,6 @@
   {
    "cell_type": "code",
    "execution_count": 3,
-   "id": "bizarre-defendant",
    "metadata": {},
    "outputs": [],
    "source": [
@@ -86,7 +81,6 @@
   },
   {
    "cell_type": "markdown",
-   "id": "anticipated-diving",
    "metadata": {},
    "source": [
     "### Sample images\n",
@@ -97,7 +91,6 @@
   {
    "cell_type": "code",
    "execution_count": 4,
-   "id": "south-government",
    "metadata": {},
    "outputs": [
     {
@@ -119,29 +112,27 @@
   },
   {
    "cell_type": "markdown",
-   "id": "dynamic-terrain",
    "metadata": {},
    "source": [
-    "### Run progressive learning\n",
+    "### Run synergistic learning\n",
     "\n",
-    "Here we provide two options of implementations of progressive learning: \n",
+    "Here we provide two options of implementations of synergistical learning: \n",
     "\n",
-    "- omnidirectional forest (Odif), which uses uncertainty forests  as the base representer\n",
-    "- omnidirectional networks (Odin), which uses a deep network as the base representer.\n",
+    "- synergistic forest (SynF), which uses uncertainty forests  as the base representer\n",
+    "- synergistic network (SynN), which uses a deep network as the base representer.\n",
     "\n",
-    "Use `odif` for omnidirectional forest and `odin` for omnidirectional networks."
+    "Use `synf` for synergistic forest and `synn` for synergistic network."
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 5,
-   "id": "angry-margin",
    "metadata": {},
    "outputs": [],
    "source": [
     "from functions.fte_bte_aircraft_bird_functions import single_experiment\n",
     "\n",
-    "model = \"odif\"  # Choose 'odif' or 'odin'\n",
+    "model = \"synf\"  # Choose 'synf' or 'synn'\n",
     "ntrees = 10  # Number of trees\n",
     "num_repetition = 30\n",
     "\n",
@@ -157,7 +148,6 @@
   },
   {
    "cell_type": "markdown",
-   "id": "fabulous-sunset",
    "metadata": {},
    "source": [
     "### Calculate and plot transfer efficiency"
@@ -166,7 +156,6 @@
   {
    "cell_type": "code",
    "execution_count": 6,
-   "id": "adopted-cuisine",
    "metadata": {},
    "outputs": [],
    "source": [
@@ -178,7 +167,6 @@
   {
    "cell_type": "code",
    "execution_count": 7,
-   "id": "casual-probe",
    "metadata": {},
    "outputs": [
     {
@@ -215,7 +203,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.8.8"
+   "version": "3.8.5"
   }
  },
  "nbformat": 4,

docs/experiments/fte_bte_dtd.ipynb

@@ -8,7 +8,7 @@
     "\n",
     "The progressive learning package utilizes representation ensembling algorithms to sequentially learn a representation for each task and ensemble both old and new representations for all future decisions. \n",
     "\n",
-    "Here, a representation ensembling algorithm based on decision forests (Odif) and an algorithm based on neural networks (Odin) demonstrate forward and backward knowledge transfer of tasks on the Describable Textures Dataset (DTD). The original dataset can be found at https://www.robots.ox.ac.uk/~vgg/data/dtd/.\n",
+    "Here, a representation ensembling algorithm based on decision forests (SynF) and an algorithm based on neural networks (SynN) demonstrate forward and backward knowledge transfer of tasks on the Describable Textures Dataset (DTD). The original dataset can be found at https://www.robots.ox.ac.uk/~vgg/data/dtd/.\n",
     "\n",
     "### Import necessary packages and modules"
    ]
@@ -167,8 +167,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# Choose algorithm (odif or odin)\n",
-    "model = \"odin\""
+    "# Choose algorithm (synf or synn)\n",
+    "model = \"synn\""
    ]
   },
   {
@@ -205,9 +205,9 @@
     "    for z in range(which_task):\n",
     "        for y in range(1):\n",
     "            for x in range(4):\n",
-    "                if model == \"odin\":\n",
+    "                if model == \"synn\":\n",
     "                    acc_x.append(acc[x][y][\"task_accuracy\"][z])\n",
-    "                elif model == \"odif\":\n",
+    "                elif model == \"synf\":\n",
     "                    acc_x.append(acc[0][x][y][\"task_accuracy\"][z])\n",
     "            acc_y.append(np.mean(acc_x))\n",
     "            acc_x = []\n",
@@ -224,9 +224,9 @@
     "    for z in range((which_task - 1), 10):\n",
     "        for y in range(1):\n",
     "            for x in range(4):\n",
-    "                if model == \"odin\":\n",
+    "                if model == \"synn\":\n",
     "                    acc_x.append(acc[x][y][\"task_accuracy\"][z])\n",
-    "                elif model == \"odif\":\n",
+    "                elif model == \"synf\":\n",
     "                    acc_x.append(acc[0][x][y][\"task_accuracy\"][z])\n",
     "            acc_y.append(np.mean(acc_x))\n",
     "            acc_x = []\n",
@@ -267,7 +267,7 @@
    "metadata": {},
    "source": [
     "### Plotting FTE, BTE, TE, and Accuracy\n",
-    "Run cell to generate a figure containing 4 plots of the forward transfer efficiency, backward transfer efficiency, transfer efficiency, and accuracy of the Odif/Odin algorithms. "
+    "Run cell to generate a figure containing 4 plots of the forward transfer efficiency, backward transfer efficiency, transfer efficiency, and accuracy of the SynF/SynN algorithms. "
    ]
   },
   {
@@ -305,9 +305,9 @@
     "n_tasks = 10\n",
     "# clr = [\"#e41a1c\", \"#a65628\", \"#377eb8\", \"#4daf4a\", \"#984ea3\", \"#ff7f00\", \"#CCCC00\"]\n",
     "# c = sns.color_palette(clr, n_colors=len(clr))\n",
-    "if model == \"odin\":\n",
+    "if model == \"synn\":\n",
     "    c = \"blue\"\n",
-    "elif model == \"odif\":\n",
+    "elif model == \"synf\":\n",
     "    c = \"red\"\n",
     "\n",
     "fontsize = 28\n",
@@ -420,7 +420,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.8.8"
+   "version": "3.8.5"
   }
  },
  "nbformat": 4,

docs/experiments/fte_bte_flowers.ipynb

@@ -8,7 +8,7 @@
     "\n",
     "The progressive learning package utilizes representation ensembling algorithms to sequentially learn a representation for each task and ensemble both old and new representations for all future decisions. \n",
     "\n",
-    "Here, a representation ensembling algorithm based on decision forests (Odif) and an algorithm based on neural networks (Odin) demonstrate forward and backward knowledge transfer of tasks on the Flowers dataset. The original dataset can be found at https://www.robots.ox.ac.uk/~vgg/data/flowers/102/index.html.\n",
+    "Here, a representation ensembling algorithm based on decision forests (SynF) and an algorithm based on neural networks (SynN) demonstrate forward and backward knowledge transfer of tasks on the Flowers dataset. The original dataset can be found at https://www.robots.ox.ac.uk/~vgg/data/flowers/102/index.html.\n",
     "\n",
     "### Import necessary packages and modules"
    ]
@@ -156,8 +156,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# Choose algorithm (odif or odin)\n",
-    "model = \"odin\""
+    "# Choose algorithm (synf or synn)\n",
+    "model = \"synn\""
    ]
   },
   {
@@ -251,7 +251,7 @@
    "metadata": {},
    "source": [
     "### Plotting FTE, BTE, TE, and Accuracy\n",
-    "Run cell to generate a figure containing 4 plots of the forward transfer efficiency, backward transfer efficiency, transfer efficiency, and accuracy of the Odif/Odin algorithms."
+    "Run cell to generate a figure containing 4 plots of the forward transfer efficiency, backward transfer efficiency, transfer efficiency, and accuracy of the SynF/SynN algorithms."
    ]
   },
   {
@@ -289,9 +289,9 @@
     "n_tasks = 10\n",
     "# clr = [\"#e41a1c\", \"#a65628\", \"#377eb8\", \"#4daf4a\", \"#984ea3\", \"#ff7f00\", \"#CCCC00\"]\n",
     "# c = sns.color_palette(clr, n_colors=len(clr))\n",
-    "if model == \"odin\":\n",
+    "if model == \"synn\":\n",
     "    c = \"blue\"\n",
-    "elif model == \"odif\":\n",
+    "elif model == \"synf\":\n",
     "    c = \"red\"\n",
     "\n",
     "fontsize = 28\n",
@@ -404,7 +404,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.8.8"
+   "version": "3.8.5"
   }
  },
  "nbformat": 4,

docs/experiments/fte_bte_food101.ipynb

@@ -8,7 +8,7 @@
     "\n",
     "The progressive learning package utilizes representation ensembling algorithms to sequentially learn a representation for each task and ensemble both old and new representations for all future decisions. \n",
     "\n",
-    "Here, a representation ensembling algorithm based on decision forests (Lifelong Forest) demonstrates forward and backward knowledge transfer of tasks on the food-101 dataset. The original dataset can be found at https://data.vision.ee.ethz.ch/cvl/datasets_extra/food-101/static/bossard_eccv14_food-101.pdf and https://www.tensorflow.org/datasets/catalog/food101.\n",
+    "Here, a representation ensembling algorithm based on decision forests (SynF) demonstrates forward and backward knowledge transfer of tasks on the food-101 dataset. The original dataset can be found at https://data.vision.ee.ethz.ch/cvl/datasets_extra/food-101/static/bossard_eccv14_food-101.pdf and https://www.tensorflow.org/datasets/catalog/food101.\n",
     "\n",
     "### Import necessary packages and modules"
    ]
@@ -256,7 +256,7 @@
    "metadata": {},
    "source": [
     "### Plotting FTE, BTE, TE, and Accuracy\n",
-    "Run cell to generate a figure containing 4 plots of the forward transfer efficiency, backward transfer efficiency, transfer efficiency, and accuracy of the Lifelong Classification Forest algorithm. "
+    "Run cell to generate a figure containing 4 plots of the forward transfer efficiency, backward transfer efficiency, transfer efficiency, and accuracy of the SynF algorithm. "
    ]
   },
   {
@@ -358,7 +358,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.8.8"
+   "version": "3.8.5"
   }
  },
  "nbformat": 4,

docs/experiments/functions/fte_bte_aircraft_bird_functions.py

@@ -15,7 +15,7 @@
 )
 from proglearn.voters import TreeClassificationVoter, KNNClassificationVoter
 from sklearn.model_selection import train_test_split
-from tensorflow.keras.backend import clear_session  # To avoid OOM error when using odin
+from tensorflow.keras.backend import clear_session  # To avoid OOM error when using synn
 
 
 def load_tasks(
@@ -107,13 +107,13 @@ def show_image(train_x_task):
 
 
 def single_experiment(
-    train_x_task, test_x_task, train_y_task, test_y_task, ntrees=10, model="odif"
+    train_x_task, test_x_task, train_y_task, test_y_task, ntrees=10, model="synf"
 ):
     num_tasks = 10
     num_points_per_task = 1800
     accuracies = np.zeros(65, dtype=float)
 
-    if model == "odin":
+    if model == "synn":
 
         clear_session()  # clear GPU memory before each run, to avoid OOM error
 
@@ -194,7 +194,7 @@ def single_experiment(
         default_voter_kwargs = {"k": int(np.log2(num_points_per_task))}
         default_decider_class = SimpleArgmaxAverage
 
-    elif model == "odif":
+    elif model == "synf":
         for i in range(num_tasks):
             train_x_task[i] = train_x_task[i].reshape(1080, -1)
             test_x_task[i] = test_x_task[i].reshape(720, -1)
@@ -218,7 +218,7 @@ def single_experiment(
             X=train_x_task[i],
             y=train_y_task[i],
             task_id=i,
-            num_transformers=1 if model == "odin" else ntrees,
+            num_transformers=1 if model == "synn" else ntrees,
             transformer_voter_decider_split=[0.67, 0.33, 0],
             decider_kwargs={"classes": np.unique(train_y_task[i])},
         )
@@ -231,12 +231,12 @@ def single_experiment(
             if j > i:
                 pass  # this is not wrong but misleading, should be continue
             else:
-                odif_predictions = progressive_learner.predict(
+                synf_predictions = progressive_learner.predict(
                     test_x_task[j], task_id=j
                 )
 
             accuracies[10 + j + (i * (i + 1)) // 2] = np.mean(
-                odif_predictions == test_y_task[j]
+                synf_predictions == test_y_task[j]
             )
     # print('single experiment done!')