fix adv debiasing

notebooks/Script Tester.ipynb

@@ -2,27 +2,70 @@
  "cells": [
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 1,
    "metadata": {},
    "outputs": [],
    "source": [
     "from importlib import reload\n",
     "import sys\n",
+    "\n",
     "src_path =   '../src' # change as needed\n",
-    "sys.path.insert(0,src_path)"
+    "sys.path.insert(0,src_path)\n",
+    "\n",
+    "max_length = 128"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 2,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "INFO:absl:Using /tmp/tfhub_modules to cache modules.\n",
+      "100%|██████████| 397080/397080 [02:03<00:00, 3212.44it/s]\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "        tag  cat  occurences\n",
+      "0    B-MISC    0         800\n",
+      "1     I-LOC    1      328160\n",
+      "2    I-MISC    2      139600\n",
+      "3     I-ORG    3      237760\n",
+      "4     I-PER    4      977800\n",
+      "5         O    5     6661880\n",
+      "6  [nerCLS]    6      397080\n",
+      "7  [nerPAD]    7    39623065\n",
+      "8  [nerSEP]    8      397080\n",
+      "9    [nerX]    9     2063015\n",
+      "\n",
+      "                tag  cat  occurences\n",
+      "0  AFRICAN-AMERICAN    0      436788\n",
+      "1          EUROPEAN    1      208467\n",
+      "2         [raceCLS]    2      397080\n",
+      "3         [racePAD]    3    39623065\n",
+      "4         [raceSEP]    4      397080\n",
+      "5           [raceX]    5     9763760\n",
+      "\n",
+      "           tag  cat  occurences\n",
+      "0       FEMALE    0      367299\n",
+      "1         MALE    1      277956\n",
+      "2  [genderCLS]    2      397080\n",
+      "3  [genderPAD]    3    39623065\n",
+      "4  [genderSEP]    4      397080\n",
+      "5    [genderX]    5     9763760\n",
+      "\n"
+     ]
+    }
+   ],
    "source": [
     "import data_generator;reload(data_generator)\n",
     "\n",
-    "#Start session\n",
-    "max_length = 128\n",
-    "\n",
     "train_data, val_data, test_data = data_generator.GetData(max_length)"
    ]
   },
@@ -37,25 +80,31 @@
     "import tensorflow as tf\n",
     "tf.logging.set_verbosity(tf.logging.ERROR)\n",
     "import model_utils; reload(model_utils)\n",
-    "\n",
-    "adam_customized = tf.keras.optimizers.Adam(lr=0.001, beta_1=0.91, beta_2=0.999, epsilon=None, decay=0.1, amsgrad=False)\n",
     "    \n",
     "config = tf.ConfigProto()\n",
     "config.gpu_options.allow_growth = True\n",
     "sess = tf.Session(config=config)\n",
     "\n",
-    "model = model_utils.NER()\n",
+    "model = model_utils.NER(max_length)\n",
     "    \n",
-    "model.generate(max_length, train_layers=4, optimizer = adam_customized, debias=False, debiasWeight=0.95)\n",
-    "\n",
-    "# Instantiate variables\n",
-    "model_utils.initialize_vars(sess)\n",
-    "\n",
+    "model.generate(bert_train_layers=4)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "scrolled": true
+   },
+   "outputs": [],
+   "source": [
     "model.fit(\n",
-    "    train_data, \n",
+    "    sess,\n",
+    "    train_data,\n",
     "    val_data,\n",
     "    epochs=8,\n",
-    "    batch_size=32\n",
+    "    batch_size=32,\n",
+    "    debias=True\n",
     ")"
    ]
   },
@@ -65,7 +114,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "cm = model.score(test_data)"
+    "bias = model.getBiasedPValues(test_data, num_iterations=1000)"
    ]
   },
   {
@@ -74,7 +123,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "bias = model.getBiasedPValues(test_data, num_iterations=10000)"
+    "bias"
    ]
   },
   {
@@ -83,9 +132,16 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "bias"
+    "sess.close()"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
   {
    "cell_type": "code",
    "execution_count": null,