ML solution - WIP [DO NOT MERGE YET]

notebooks/kaggle-satellite-challenge-model.ipynb

@@ -0,0 +1,141 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Model\n",
+    "* load ternaus-net model (or my model for bone seg)\n",
+    "* set frozen pretrained VGG weights\n",
+    "* keras to estimator\n",
+    "* T2T keras model - TODO\n",
+    "* train the rest\n",
+    "* fine tune the whole network - TODO\n",
+    "\n",
+    "References:\n",
+    "* [VGG16 TernausNet](https://github.com/ternaus/TernausNet/blob/master/unet_models.py) \n",
+    "* [Keras Transfer Learning I](https://www.learnopencv.com/keras-tutorial-fine-tuning-using-pre-trained-models)\n",
+    "* [Keras Transfer Learning II](https://flyyufelix.github.io/2016/10/08/fine-tuning-in-keras-part2.html)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 17,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "INFO:tensorflow:Using the Keras model provided.\n",
+      "INFO:tensorflow:Using default config.\n",
+      "WARNING:tensorflow:Using temporary folder as model directory: /tmp/tmplcozb2dq\n",
+      "INFO:tensorflow:Using config: {'_model_dir': '/tmp/tmplcozb2dq', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': None, '_save_checkpoints_secs': 600, '_session_config': None, '_keep_checkpoint_max': 5, '_keep_checkpoint_every_n_hours': 10000, '_log_step_count_steps': 100, '_train_distribute': None, '_service': None, '_cluster_spec': <tensorflow.python.training.server_lib.ClusterSpec object at 0x7f58bb6fcb38>, '_task_type': 'worker', '_task_id': 0, '_global_id_in_cluster': 0, '_master': '', '_evaluation_master': '', '_is_chief': True, '_num_ps_replicas': 0, '_num_worker_replicas': 1}\n"
+     ]
+    },
+    {
+     "data": {
+      "text/plain": [
+       "<tensorflow.python.estimator.estimator.Estimator at 0x7f58d86fff28>"
+      ]
+     },
+     "execution_count": 17,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "import sys\n",
+    "sys.path.append('../sentinel/')\n",
+    "sys.path.append('../')\n",
+    "\n",
+    "from importlib import reload\n",
+    "\n",
+    "import numpy as np\n",
+    "import tensorflow as tf\n",
+    "\n",
+    "import vgg_unet\n",
+    "reload(vgg_unet)\n",
+    "from vgg_unet import vgg16_unet, jaccard_coef_loss, jaccard_coef_int\n",
+    "\n",
+    "%matplotlib inline\n",
+    "%load_ext autoreload\n",
+    "%autoreload 2\n",
+    "\n",
+    "hparams = tf.contrib.training.HParams(\n",
+    "    image_size=224,\n",
+    "    num_classes=1,\n",
+    "    num_filters=32,\n",
+    "    is_deconv=False,\n",
+    "    use_bn_elu=False,\n",
+    "    freeze_encoder=True,\n",
+    ")\n",
+    "\n",
+    "ternaus_net = vgg16_unet(hparams)\n",
+    "Nadam = tf.keras.optimizers.Nadam\n",
+    "ternaus_net.compile(optimizer=Nadam(lr=1e-3), loss=jaccard_coef_loss,\n",
+    "                    metrics=['binary_crossentropy', jaccard_coef_int])\n",
+    "\n",
+    "estimator = tf.keras.estimator.model_to_estimator(ternaus_net)\n",
+    "estimator"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Fix the variables in the TEST"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 16,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# ternaus_net._make_train_function()\n",
+    "# train_op = ternaus_net.train_function.updates_op\n",
+    "\n",
+    "# var_name = 'block'\n",
+    "# variables = [x for x in tf.trainable_variables() if var_name in x.name]\n",
+    "# var_list = list(set(tf.trainable_variables()) - set(variables))\n",
+    "\n",
+    "# import numpy as np\n",
+    "# import mltest\n",
+    "# feed_dict = {\n",
+    "#         'input_1:0': np.random.normal(size=(10,)+input_shape),\n",
+    "#         'logit_target:0': np.random.randint(1, size=(10, 224, 224, 1)),\n",
+    "#         'logit_sample_weights:0': [0.] * 10\n",
+    "#     }\n",
+    "\n",
+    "# # Run the test suite!\n",
+    "# mltest.test_suite(\n",
+    "#     ternaus_net.output,\n",
+    "#     train_op,\n",
+    "#     feed_dict=feed_dict,\n",
+    "#     var_list=var_list)"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "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": "3.6.5"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}