Skip to content

Instantly share code, notes, and snippets.

@ababino

ababino/tabular-export-learner-testing.ipynb

Last active November 13, 2020 08:28
Show Gist options
  • Save ababino/2a2c67ac264e2ed8c95144377b9be2b4 to your computer and use it in GitHub Desktop.
Save ababino/2a2c67ac264e2ed8c95144377b9be2b4 to your computer and use it in GitHub Desktop.
Download ZIP
tabular-export-learner-testing.ipynb
Display the source blob
Display the rendered blob
Raw
tabular-export-learner-testing.ipynb
{
"nbformat": 4,
"nbformat_minor": 0,
"metadata": {
"accelerator": "GPU",
"colab": {
"name": "tabular-export-learner-testing.ipynb",
"provenance": [],
"collapsed_sections": [],
"include_colab_link": true
},
"jupytext": {
"split_at_heading": 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/ababino/2a2c67ac264e2ed8c95144377b9be2b4/tabular-export-learner-testing.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "XrJzkfJ-1bKp"
},
"source": [
"# Tabular models"
]
},
{
"cell_type": "code",
"metadata": {
"id": "Lu4U4bzjv0oH",
"outputId": "5e4e9767-2525-447b-803e-7877cce896eb",
"colab": {
"base_uri": "https://localhost:8080/"
}
},
"source": [
"!pip install fastai --upgrade -qqq\n",
"!pip install pympler"
],
"execution_count": 1,
"outputs": [
{
"output_type": "stream",
"text": [
"Requirement already satisfied: pympler in /usr/local/lib/python3.6/dist-packages (0.9)\n"
],
"name": "stdout"
}
]
},
{
"cell_type": "code",
"metadata": {
"id": "_kw8-Dju_6J5",
"outputId": "3b2ec4af-7fa7-4a72-f542-366f39ea0a33",
"colab": {
"base_uri": "https://localhost:8080/",
"height": 35
}
},
"source": [
"import fastai\n",
"fastai.__version__"
],
"execution_count": 2,
"outputs": [
{
"output_type": "execute_result",
"data": {
"application/vnd.google.colaboratory.intrinsic+json": {
"type": "string"
},
"text/plain": [
"'2.1.5'"
]
},
"metadata": {
"tags": []
},
"execution_count": 2
}
]
},
{
"cell_type": "code",
"metadata": {
"id": "K-NLPMGM1bKr"
},
"source": [
"from fastai.tabular.all import *\n",
"from fastai.callback.all import *\n",
"from pympler import muppy\n",
"import warnings\n",
"warnings.filterwarnings('ignore')"
],
"execution_count": 3,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "7fsTZYn5a18b"
},
"source": [
"def print_dfs_lens():\n",
" all_objects = muppy.get_objects()\n",
" my_types = muppy.filter(all_objects, Type=pd.DataFrame)\n",
" my_types_lens = [len(x) for x in my_types]\n",
" print(my_types_lens[:-3], my_types_lens[-3:])"
],
"execution_count": 4,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "dzHkM_t41bKx"
},
"source": [
"Tabular data should be in a Pandas `DataFrame`."
]
},
{
"cell_type": "code",
"metadata": {
"id": "xLLnzxw_1bKy"
},
"source": [
"path = untar_data(URLs.ADULT_SAMPLE)\n",
"df = pd.read_csv(path/'adult.csv')\n",
"for i in range(4):\n",
" df = df.append(df.copy())"
],
"execution_count": 5,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "TBoo6c761bK2"
},
"source": [
"dep_var = 'salary'\n",
"cat_names = ['workclass', 'education', 'marital-status', 'occupation', 'relationship', 'race']\n",
"cont_names = ['age', 'fnlwgt', 'education-num']\n",
"procs = [Categorify, FillMissing, Normalize]"
],
"execution_count": 6,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "ZNlr9V4H1bK_"
},
"source": [
"splits = RandomSplitter()(range_of(df))"
],
"execution_count": 7,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "2ERqyHkxE7sq",
"outputId": "7e5f0ce5-b21c-4990-ff15-6cee2a8b0f94",
"colab": {
"base_uri": "https://localhost:8080/"
}
},
"source": [
"splits"
],
"execution_count": 8,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/plain": [
"((#416781) [500824,451294,253829,307438,386053,451314,171109,189741,68905,477899...],\n",
" (#104195) [7873,188230,219329,201724,405331,315616,162262,14314,369423,412459...])"
]
},
"metadata": {
"tags": []
},
"execution_count": 8
}
]
},
{
"cell_type": "code",
"metadata": {
"id": "cTd_Ckt61bLH"
},
"source": [
"to = TabularPandas(df, procs, cat_names, cont_names, y_names=\"salary\", splits=splits)"
],
"execution_count": 9,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "7UvARddj1bLN"
},
"source": [
"dls = to.dataloaders(bs=128)"
],
"execution_count": 10,
"outputs": []
},
{
"cell_type": "markdown",
"metadata": {
"id": "Ucg9ao3dEpbF"
},
"source": [
"First we'll check before fit:"
]
},
{
"cell_type": "code",
"metadata": {
"id": "bRYiL1gZ1bLZ",
"outputId": "429d9059-504e-4e7b-d870-eb80f0ca046b",
"colab": {
"base_uri": "https://localhost:8080/"
}
},
"source": [
"print_dfs_lens()\n",
"learn = tabular_learner(dls, layers=[200,100], metrics=accuracy)\n",
"learn.export('t1')\n",
"l1 = load_learner('t1')\n",
"print_dfs_lens()"
],
"execution_count": 11,
"outputs": [
{
"output_type": "stream",
"text": [
"[520976, 0] [520976, 104195, 416781]\n",
"[520976, 0, 520976, 104195, 416781] [0, 0, 0]\n"
],
"name": "stdout"
}
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "dexvA6_4q8i6"
},
"source": [
"Now exporting after fit"
]
},
{
"cell_type": "code",
"metadata": {
"id": "n3VQpLnjjV2D",
"outputId": "6b79d14c-0972-4c44-c9e4-ec24bd305b31",
"colab": {
"base_uri": "https://localhost:8080/",
"height": 114
}
},
"source": [
"print_dfs_lens()\n",
"learn = tabular_learner(dls, layers=[200,100], metrics=accuracy)\n",
"learn.fit(1, cbs=ShortEpochCallback())\n",
"learn.export('t2')\n",
"l2 = load_learner('t2')\n",
"print_dfs_lens()"
],
"execution_count": 12,
"outputs": [
{
"output_type": "stream",
"text": [
"[520976, 0, 520976, 104195, 416781] [0, 0, 0]\n"
],
"name": "stdout"
},
{
"output_type": "display_data",
"data": {
"text/html": [
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: left;\">\n",
" <th>epoch</th>\n",
" <th>train_loss</th>\n",
" <th>valid_loss</th>\n",
" <th>accuracy</th>\n",
" <th>time</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <td>0</td>\n",
" <td>00:00</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {
"tags": []
}
},
{
"output_type": "stream",
"text": [
"[520976, 0, 520976, 104195, 416781, 0, 0, 0, 0] [0, 104195, 0]\n"
],
"name": "stdout"
}
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "bSpqmQc5rF21"
},
"source": [
"We see that the loaded learner experted after fitting has holds the valid ds somewhere"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "XtvccFvdpwk1"
},
"source": [
"## Fix"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "Fz0-WjWVozli"
},
"source": [
"The problem is that `ProgressCallback` holds the last dl (valid) in the `pbar` attribute as you can see [here](https://github.com/fastai/fastai/blob/ef5c3609b8da03fb25953a7f4f49aaa1e69197d2/fastai/callback/progress.py#L34).\n",
"If we patch ProgressCallback to delete the `pbar` attribute in the last line in `after_fit` we fix the problem."
]
},
{
"cell_type": "code",
"metadata": {
"id": "0Klq5ig7oqWh"
},
"source": [
"@patch\n",
"def after_fit(self:ProgressCallback):\n",
" if getattr(self, 'mbar', False):\n",
" self.mbar.on_iter_end()\n",
" delattr(self, 'mbar')\n",
" if hasattr(self, 'old_logger'): self.learn.logger = self.old_logger\n",
" ###### this is the new line #############\n",
" if getattr(self, 'pbar', False): delattr(self, 'pbar')"
],
"execution_count": 13,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "b5RQn0X7p78J",
"outputId": "6c986bce-ee4b-4100-f56d-593a9388581b",
"colab": {
"base_uri": "https://localhost:8080/",
"height": 114
}
},
"source": [
"print_dfs_lens()\n",
"learn = tabular_learner(dls, layers=[200,100], metrics=accuracy)\n",
"learn.fit(1, cbs=ShortEpochCallback())\n",
"learn.export('t3')\n",
"l3 = load_learner('t3')\n",
"print_dfs_lens()"
],
"execution_count": 14,
"outputs": [
{
"output_type": "stream",
"text": [
"[520976, 0, 520976, 104195, 416781, 0, 0, 0, 0] [0, 104195, 0]\n"
],
"name": "stdout"
},
{
"output_type": "display_data",
"data": {
"text/html": [
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: left;\">\n",
" <th>epoch</th>\n",
" <th>train_loss</th>\n",
" <th>valid_loss</th>\n",
" <th>accuracy</th>\n",
" <th>time</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <td>0</td>\n",
" <td>00:00</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {
"tags": []
}
},
{
"output_type": "stream",
"text": [
"[520976, 0, 520976, 104195, 416781, 0, 0, 0, 0, 0, 104195, 0] [0, 0, 0]\n"
],
"name": "stdout"
}
]
}
]
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment