takotab/mnist-data-size-example-of-beter-generaization.ipynb

## mnist-data-size-example-of-beter-generaization.ipynb
{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "MNIST data size example of beter generaization.ipynb",
      "version": "0.3.2",
      "provenance": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/gist/takotab/5e44e4f229e9f0acc19ca6926585f380/mnist-data-size-example-of-beter-generaization.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "metadata": {
        "id": "GD7Jw1uY26Xq",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Other data formats"
      ]
    },
    {
      "metadata": {
        "id": "RTSzZyYF26Xq",
        "colab_type": "code",
        "outputId": "6681bdc9-a5ff-4722-feb6-2524f76a6ed9",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "cell_type": "code",
      "source": [
        "path = untar_data(URLs.MNIST_SAMPLE); path"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "PosixPath('/root/.fastai/data/mnist_sample')"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 4
        }
      ]
    },
    {
      "metadata": {
        "id": "3XrV6g_f26Xu",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "tfms = get_transforms(do_flip=False)\n",
        "data = ImageDataBunch.from_folder(path, ds_tfms=tfms, size=20)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "P7LXe7z-26Xv",
        "colab_type": "code",
        "outputId": "97e9abcc-510c-4de2-80bb-be5d8fe5d3c3",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 365
        }
      },
      "cell_type": "code",
      "source": [
        "data.show_batch(rows=3, figsize=(5,5))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAU4AAAFcCAYAAACnTVf5AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nTkSuQmCC\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7f0489b81b00>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "prItFOSc26X3",
        "colab_type": "code",
        "outputId": "8e014c54-ba5d-49dc-cbc3-a283e3ee89fe",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 85
        }
      },
      "cell_type": "code",
      "source": [
        "data = ImageDataBunch.from_folder(path, ds_tfms=tfms, size=20)\n",
        "learn = create_cnn(data, models.resnet18, metrics=accuracy)\n",
        "learn.fit(1)\n",
        "learn.save('save-20')\n",
        "data = ImageDataBunch.from_folder(path, ds_tfms=tfms, size=25)\n",
        "learn = create_cnn(data, models.resnet18, metrics=accuracy)\n",
        "learn.load('save-20')\n",
        "learn.fit(1)\n",
        "learn.save('save-25')\n",
        "data = ImageDataBunch.from_folder(path, ds_tfms=tfms, size=28)\n",
        "learn = create_cnn(data, models.resnet18, metrics=accuracy)\n",
        "learn.load('save-25')\n",
        "learn.fit(1)\n",
        "learn.save('save-28')"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Total time: 03:37\n",
            "epoch  train_loss  valid_loss  accuracy\n",
            "1      0.040100    0.008342    0.997547  (03:37)\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "LHnt21Sh9WZC",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "interp = ClassificationInterpretation.from_learner(learn)\n",
        "interp.plot_top_losses(9, figsize=(15,11))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "NrtgF53O8ZUk",
        "colab_type": "code",
        "outputId": "bffb6ba2-e8b8-49f0-8dca-ccfc30d02db3",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 119
        }
      },
      "cell_type": "code",
      "source": [
        "# benchmark 28\n",
        "data = ImageDataBunch.from_folder(path, ds_tfms=tfms, size=28)\n",
        "learn = create_cnn(data, models.resnet18, metrics=accuracy)\n",
        "learn.fit(3)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Total time: 10:28\n",
            "epoch  train_loss  valid_loss  accuracy\n",
            "1      0.107596    0.026979    0.990677  (03:29)\n",
            "2      0.056810    0.015477    0.994112  (03:28)\n",
            "3      0.037681    0.009312    0.998037  (03:29)\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    }
  ]
}
	{
	"nbformat": 4,
	"nbformat_minor": 0,
	"metadata": {
	"colab": {
	"name": "MNIST data size example of beter generaization.ipynb",
	"version": "0.3.2",
	"provenance": [],
	"include_colab_link": true
	},
	"kernelspec": {
	"display_name": "Python 3",
	"language": "python",
	"name": "python3"
	}
	},
	"cells": [
	{
	"cell_type": "markdown",
	"metadata": {
	"id": "view-in-github",
	"colab_type": "text"
	},
	"source": [
	"<a href=\"https://colab.research.google.com/gist/takotab/5e44e4f229e9f0acc19ca6926585f380/mnist-data-size-example-of-beter-generaization.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
	]
	},
	{
	"metadata": {
	"id": "GD7Jw1uY26Xq",
	"colab_type": "text"
	},
	"cell_type": "markdown",
	"source": [
	"## Other data formats"
	]
	},
	{
	"metadata": {
	"id": "RTSzZyYF26Xq",
	"colab_type": "code",
	"outputId": "6681bdc9-a5ff-4722-feb6-2524f76a6ed9",
	"colab": {
	"base_uri": "https://localhost:8080/",
	"height": 34
	}
	},
	"cell_type": "code",
	"source": [
	"path = untar_data(URLs.MNIST_SAMPLE); path"
	],
	"execution_count": 0,
	"outputs": [
	{
	"output_type": "execute_result",
	"data": {
	"text/plain": [
	"PosixPath('/root/.fastai/data/mnist_sample')"
	]
	},
	"metadata": {
	"tags": []
	},
	"execution_count": 4
	}
	]
	},
	{
	"metadata": {
	"id": "3XrV6g_f26Xu",
	"colab_type": "code",
	"colab": {}
	},
	"cell_type": "code",
	"source": [
	"tfms = get_transforms(do_flip=False)\n",
	"data = ImageDataBunch.from_folder(path, ds_tfms=tfms, size=20)"
	],
	"execution_count": 0,
	"outputs": []
	},
	{
	"metadata": {
	"id": "P7LXe7z-26Xv",
	"colab_type": "code",
	"outputId": "97e9abcc-510c-4de2-80bb-be5d8fe5d3c3",
	"colab": {
	"base_uri": "https://localhost:8080/",
	"height": 365
	}
	},
	"cell_type": "code",
	"source": [
	"data.show_batch(rows=3, figsize=(5,5))"
	],
	"execution_count": 0,
	"outputs": [
	{
	"output_type": "display_data",
	"data": {
	"image/png": "iVBORw0KGgoAAAANSUhEUgAAAU4AAAFcCAYAAACnTVf5AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nTkSuQmCC\n",
	"text/plain": [
	"<matplotlib.figure.Figure at 0x7f0489b81b00>"
	]
	},
	"metadata": {
	"tags": []
	}
	}
	]
	},
	{
	"metadata": {
	"id": "prItFOSc26X3",
	"colab_type": "code",
	"outputId": "8e014c54-ba5d-49dc-cbc3-a283e3ee89fe",
	"colab": {
	"base_uri": "https://localhost:8080/",
	"height": 85
	}
	},
	"cell_type": "code",
	"source": [
	"data = ImageDataBunch.from_folder(path, ds_tfms=tfms, size=20)\n",
	"learn = create_cnn(data, models.resnet18, metrics=accuracy)\n",
	"learn.fit(1)\n",
	"learn.save('save-20')\n",
	"data = ImageDataBunch.from_folder(path, ds_tfms=tfms, size=25)\n",
	"learn = create_cnn(data, models.resnet18, metrics=accuracy)\n",
	"learn.load('save-20')\n",
	"learn.fit(1)\n",
	"learn.save('save-25')\n",
	"data = ImageDataBunch.from_folder(path, ds_tfms=tfms, size=28)\n",
	"learn = create_cnn(data, models.resnet18, metrics=accuracy)\n",
	"learn.load('save-25')\n",
	"learn.fit(1)\n",
	"learn.save('save-28')"
	],
	"execution_count": 0,
	"outputs": [
	{
	"output_type": "stream",
	"text": [
	"Total time: 03:37\n",
	"epoch train_loss valid_loss accuracy\n",
	"1 0.040100 0.008342 0.997547 (03:37)\n",
	"\n"
	],
	"name": "stdout"
	}
	]
	},
	{
	"metadata": {
	"id": "LHnt21Sh9WZC",
	"colab_type": "code",
	"colab": {}
	},
	"cell_type": "code",
	"source": [
	"interp = ClassificationInterpretation.from_learner(learn)\n",
	"interp.plot_top_losses(9, figsize=(15,11))"
	],
	"execution_count": 0,
	"outputs": []
	},
	{
	"metadata": {
	"id": "NrtgF53O8ZUk",
	"colab_type": "code",
	"outputId": "bffb6ba2-e8b8-49f0-8dca-ccfc30d02db3",
	"colab": {
	"base_uri": "https://localhost:8080/",
	"height": 119
	}
	},
	"cell_type": "code",
	"source": [
	"# benchmark 28\n",
	"data = ImageDataBunch.from_folder(path, ds_tfms=tfms, size=28)\n",
	"learn = create_cnn(data, models.resnet18, metrics=accuracy)\n",
	"learn.fit(3)"
	],
	"execution_count": 0,
	"outputs": [
	{
	"output_type": "stream",
	"text": [
	"Total time: 10:28\n",
	"epoch train_loss valid_loss accuracy\n",
	"1 0.107596 0.026979 0.990677 (03:29)\n",
	"2 0.056810 0.015477 0.994112 (03:28)\n",
	"3 0.037681 0.009312 0.998037 (03:29)\n",
	"\n"
	],
	"name": "stdout"
	}
	]
	}
	]
	}