{
  "cells": [
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "%matplotlib inline"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "\n\n# Common errors with onnxruntime\n\nThis example looks into several common situations\nin which *onnxruntime* does not return the model \nprediction but raises an exception instead.\nIt starts by loading the model trained in example\n`l-logreg-example` which produced a logistic regression\ntrained on *Iris* datasets. The model takes\na vector of dimension 2 and returns a class among three.\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "import onnxruntime as rt\nfrom onnxruntime.capi.onnxruntime_pybind11_state import InvalidArgument\nimport numpy\nfrom onnxruntime.datasets import get_example\n\nexample2 = get_example(\"logreg_iris.onnx\")\nsess = rt.InferenceSession(example2, providers=rt.get_available_providers())\n\ninput_name = sess.get_inputs()[0].name\noutput_name = sess.get_outputs()[0].name"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "The first example fails due to *bad types*.\n*onnxruntime* only expects single floats (4 bytes)\nand cannot handle any other kind of floats.\n\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "try:\n    x = numpy.array([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]], dtype=numpy.float64)\n    sess.run([output_name], {input_name: x})\nexcept Exception as e:\n    print(\"Unexpected type\")\n    print(\"{0}: {1}\".format(type(e), e))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "The model fails to return an output if the name\nis misspelled.\n\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "try:\n    x = numpy.array([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]], dtype=numpy.float32)\n    sess.run([\"misspelled\"], {input_name: x})\nexcept Exception as e:\n    print(\"Misspelled output name\")\n    print(\"{0}: {1}\".format(type(e), e))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "The output name is optional, it can be replaced by *None*\nand *onnxruntime* will then return all the outputs.\n\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "x = numpy.array([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]], dtype=numpy.float32)\ntry:\n    res = sess.run(None, {input_name: x})\n    print(\"All outputs\")\n    print(res)\nexcept (RuntimeError, InvalidArgument) as e:\n    print(e)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "The same goes if the input name is misspelled.\n\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "try:\n    x = numpy.array([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]], dtype=numpy.float32)\n    sess.run([output_name], {\"misspelled\": x})\nexcept Exception as e:\n    print(\"Misspelled input name\")\n    print(\"{0}: {1}\".format(type(e), e))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "*onnxruntime* does not necessarily fail if the input\ndimension is a multiple of the expected input dimension.\n\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "for x in [\n        numpy.array([1.0, 2.0, 3.0, 4.0], dtype=numpy.float32),\n        numpy.array([[1.0, 2.0, 3.0, 4.0]], dtype=numpy.float32),\n        numpy.array([[1.0, 2.0], [3.0, 4.0]], dtype=numpy.float32),\n        numpy.array([1.0, 2.0, 3.0], dtype=numpy.float32),\n        numpy.array([[1.0, 2.0, 3.0]], dtype=numpy.float32),\n        ]:\n    try:\n        r = sess.run([output_name], {input_name: x})\n        print(\"Shape={0} and predicted labels={1}\".format(x.shape, r))\n    except (RuntimeError, InvalidArgument) as e:\n        print(\"ERROR with Shape={0} - {1}\".format(x.shape, e))\n\nfor x in [\n        numpy.array([1.0, 2.0, 3.0, 4.0], dtype=numpy.float32),\n        numpy.array([[1.0, 2.0, 3.0, 4.0]], dtype=numpy.float32),\n        numpy.array([[1.0, 2.0], [3.0, 4.0]], dtype=numpy.float32),\n        numpy.array([1.0, 2.0, 3.0], dtype=numpy.float32),\n        numpy.array([[1.0, 2.0, 3.0]], dtype=numpy.float32),\n        ]:\n    try:\n        r = sess.run(None, {input_name: x})\n        print(\"Shape={0} and predicted probabilities={1}\".format(x.shape, r[1]))\n    except (RuntimeError, InvalidArgument) as e:\n        print(\"ERROR with Shape={0} - {1}\".format(x.shape, e))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "It does not fail either if the number of dimension\nis higher than expects but produces a warning.\n\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "for x in [\n        numpy.array([[[1.0, 2.0], [3.0, 4.0]]], dtype=numpy.float32),\n        numpy.array([[[1.0, 2.0, 3.0]]], dtype=numpy.float32),\n        numpy.array([[[1.0, 2.0]], [[3.0, 4.0]]], dtype=numpy.float32),\n        ]:\n    try:\n        r = sess.run([output_name], {input_name: x})\n        print(\"Shape={0} and predicted labels={1}\".format(x.shape, r))\n    except (RuntimeError, InvalidArgument) as e:\n        print(\"ERROR with Shape={0} - {1}\".format(x.shape, e))"
      ]
    }
  ],
  "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": "3.8.10"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}