Add files via upload

seismicclassguide.ipynb

@@ -0,0 +1,127 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Guide to run Tensorflow classifier"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Steps to follow"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "#### 1. Convert the image files to jpg and save in a convenient directory\n",
+    "\n",
+    "My file for this is seismicneural.ipynb saved in the repo\n",
+    "I ended up copying the jpg files over to a directory close to home `C:\\Users\\niall\\rguhack\\notebooks`"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Only save the train images in here - but you can use most of the set for training\n",
+    "\n",
+    "Save the images in separate folders according to the classifications you want.  I saved them  in two directories `final_train_jpg` and `raw_train_jpg` "
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "#### 2. Install tensorflow -hub"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "I assume you already have tensorflow installed using pip or conda. In addition you need to install a pre-trained classifier model called tensorflow -hub.\n",
+    "You can do this with pip install - the instructions at this link\n",
+    "https://www.tensorflow.org/hub/installation"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "#### 3. Run the classification\n",
+    "\n",
+    "In the terminal type the following all in a single line\n",
+    "\n",
+    "`python hub/examples/image_retraining/retrain.py --how_many_training_steps 100 --image_dir=C:/Users/niall/rguhack/notebooks/training_data`\n",
+    "\n",
+    "It took ages to figure out this command properly - make sure and save -hub in the directory you're opening the terminal in annd for the `image-dir` command make sure you type the full directory\n",
+    "\n",
+    "You can play around with the training steps parameter.  I ran one with just 50 and it still gave excellent results\n",
+    "\n",
+    "For 100 steps this took me around 15 minutes of training"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "#### 4. Use your classifier\n",
+    "\n",
+    "This also took a while to figure out.  I'm sure Mike will be able to do something really efficient in C++ to make this more user friendly.\n",
+    "\n",
+    "The instructions for use are here but aren't easy to follow:\n",
+    "\n",
+    "https://www.tensorflow.org/hub/tutorials/image_retraining#using_the_retrained_model\n",
+    "\n",
+    "I had to find the label_image.py file somehwere else but it's in the repo now for convenience.  Copy it into your main directiry so it can run more easily.\n",
+    "\n",
+    "I just tested 'test'  images individually with weirdly accurate results.  Here's the command I ran:\n",
+    "\n",
+    "`python label_image.py --graph=/tmp/output_graph.pb --labels=/tmp/output_labels.txt --input_layer=Placeholder --output_layer=final_result --image=c\\Users\\niall\\rguhack\\ESP2D_RAW_FULL_KPSTM_STACK_059B064_4900_5000.jpg`\n",
+    "\n",
+    "This is a raw image example and I returned the following result:\n",
+    "\n",
+    "raw train jpg 0.9285909\n",
+    "final_train jpg 0.0714\n",
+    "\n",
+    "So this is saying with 93% certainty this image i passed it was 'raw'.  I ran a few and got simiar results.  Quite spooky (or over fitted!?)\n",
+    "\n",
+    "\n",
+    "\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "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": "3.6.6"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}