onnx_tensorflow_example.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "import numpy as np\n",
    "from tensorflow import keras\n",
    "import onnxruntime"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "outputs": [],
   "source": [
    "model = tf.keras.Sequential([keras.layers.Dense(units=1, input_shape=[1])])\n",
    "# 1 layer 1 value"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_2\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_2 (Dense)              (None, 1)                 2         \n",
      "=================================================================\n",
      "Total params: 2\n",
      "Trainable params: 2\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.compile(optimizer='sgd',loss='mean_squared_error')\n",
    "model.summary()\n"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "outputs": [],
   "source": [
    "# Neural network to predict y = 3x + 1\n",
    "xs = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], dtype=float)\n",
    "ys = np.array([4.0, 7.0, 10.0, 13.0, 16.0, 19.0], dtype=float)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "outputs": [
    {
     "data": {
      "text/plain": "<tensorflow.python.keras.callbacks.History at 0x7f96ba8d43d0>"
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Train model\n",
    "model.fit(xs,ys,epochs=500,verbose=False)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "outputs": [
    {
     "data": {
      "text/plain": "array([[31.02069]], dtype=float32)"
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "original = model.predict([10.0])\n",
    "\n",
    "original"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Assets written to: tf/saved/assets\n"
     ]
    }
   ],
   "source": [
    "# save model architecture, weights, and training configuration in a single file/folder\n",
    "model.save('tf/saved')"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2020-12-27 12:44:43,732 - WARNING - '--tag' not specified for saved_model. Using --tag serve\r\n",
      "2020-12-27 12:44:43,800 - INFO - Signatures found in model: [serving_default].\r\n",
      "2020-12-27 12:44:43,800 - WARNING - '--signature_def' not specified, using first signature: serving_default\r\n",
      "WARNING:tensorflow:From /Users/vishwanath/opt/miniconda3/envs/onnx_virt/lib/python3.8/site-packages/tf2onnx/tf_loader.py:416: extract_sub_graph (from tensorflow.python.framework.graph_util_impl) is deprecated and will be removed in a future version.\r\n",
      "Instructions for updating:\r\n",
      "Use `tf.compat.v1.graph_util.extract_sub_graph`\r\n",
      "2020-12-27 12:44:43,816 - WARNING - From /Users/vishwanath/opt/miniconda3/envs/onnx_virt/lib/python3.8/site-packages/tf2onnx/tf_loader.py:416: extract_sub_graph (from tensorflow.python.framework.graph_util_impl) is deprecated and will be removed in a future version.\r\n",
      "Instructions for updating:\r\n",
      "Use `tf.compat.v1.graph_util.extract_sub_graph`\r\n",
      "2020-12-27 12:44:43,823 - INFO - Using tensorflow=2.4.0, onnx=1.8.0, tf2onnx=1.7.2/995bd6\r\n",
      "2020-12-27 12:44:43,823 - INFO - Using opset <onnx, 10>\r\n",
      "2020-12-27 12:44:43,823 - INFO - Computed 0 values for constant folding\r\n",
      "2020-12-27 12:44:43,829 - INFO - Optimizing ONNX model\r\n",
      "2020-12-27 12:44:43,836 - INFO - After optimization: Identity -5 (5->0)\r\n",
      "2020-12-27 12:44:43,836 - INFO - \r\n",
      "2020-12-27 12:44:43,837 - INFO - Successfully converted TensorFlow model tf/saved to ONNX\r\n",
      "2020-12-27 12:44:43,837 - INFO - ONNX model is saved at tf/output/simplemodel.onnx\r\n"
     ]
    }
   ],
   "source": [
    "#Convert to ONNX\n",
    "\n",
    "# opset 8 to generate the graph. By specifying --opset the user can override the default to generate a graph with the desired opset. For example --opset 5 would create a onnx graph that uses only ops available in opset 5. Because older opsets have in most cases fewer ops, some models might not convert on a older opset.\n",
    "!python -m tf2onnx.convert --opset 10  --saved-model tf/saved --output tf/output/simplemodel.onnx\n"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "input name dense_2_input:0\n",
      "input shape ['unk__6', 1]\n"
     ]
    }
   ],
   "source": [
    "path = 'tf/output/simplemodel.onnx'\n",
    "sess = onnxruntime.InferenceSession(path)\n",
    "# get name and shape\n",
    "input_name = sess.get_inputs()[0].name\n",
    "print(\"input name\", input_name)\n",
    "input_shape = sess.get_inputs()[0].shape\n",
    "print(\"input shape\", input_shape)\n",
    "\n"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Input values to ONNX model = \n",
      "[[10.]]\n",
      "\n",
      "Output value from ONNX model = \n",
      "[array([[31.02069]], dtype=float32)]\n"
     ]
    }
   ],
   "source": [
    "# predict\n",
    "x = np.array([[10.0]], dtype=np.float32)\n",
    "print(\"\\nInput values to ONNX model = \")\n",
    "print(x)\n",
    "res = sess.run(None, {input_name: x})\n",
    "print(\"\\nOutput value from ONNX model = \")\n",
    "print(res)\n"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "outputs": [],
   "source": [
    "# Check if equal\n",
    "assert res == original"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "outputs": [],
   "source": [],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}