{
"cells": [
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"## Inżynieria uczenia maszynowego\n",
"### 17 kwietnia 2024\n",
"# 7. Sacred"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"## Sacred\n",
"> Every experiment is sacred
\n",
"> Every experiment is great
\n",
"> If an experiment is wasted
\n",
"> God gets quite irate\n",
">\n",
" —https://github.com/IDSIA/sacred / [Sens życia według Monty Pythona](https://en.wikipedia.org/wiki/Every_Sperm_Is_Sacred) "
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"- https://sacred.readthedocs.io/ - dokumentacja\n",
"- https://github.com/IDSIA/sacred - Github\n",
"- Open source\n",
"- Prosty w użyciu\n",
"- Wiele webowych frontendów"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"- Przeprowadzanie eksperymentów (zmiana parametrów, trenowanie, ewaluacja) uczenia maszynowego jest kosztowne i czasochłonne\n",
"- Dlatego warto przeprowadzać je w zorganizowany sposób\n",
"- I tak, żebyśmy mogli powtórzyć / odtworzyć raz uzyskane wyniki"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"> Sacred is a tool to help you:\n",
"> - configure\n",
"> - organize\n",
"> - log \n",
"> - reproduce \n",
"> experiments. \n",
"> \n",
">It is designed to do all the tedious overhead work that you need to do around your actual experiment in order to:\n",
"> - keep track of all the parameters of your experiment\n",
"> - easily run your experiment for different settings\n",
"> - save configurations for individual runs in a database\n",
"> - reproduce your results\n",
" \n",
" —https://github.com/IDSIA/sacred"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"- **ConfigScopes** A very convenient way of the local variables in a function to define the parameters your experiment uses.\n",
"- **Config Injection** You can access all parameters of your configuration from every function. They are automatically injected by name.\n",
"- **Command-line interface** You get a powerful command-line interface for each experiment that you can use to change parameters and run different variants.\n",
"- **Observers** Sacred provides Observers that log all kinds of information about your experiment, its dependencies, the configuration you used, the machine it is run on, and of course the result. These can be saved to a MongoDB, for easy access later.\n",
"- **Automatic seeding** helps controlling the randomness in your experiments, such that the results remain reproducible."
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"## Instalacja"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Requirement already satisfied: sacred in /home/tomek/miniconda3/lib/python3.9/site-packages (0.8.4)\n",
"Requirement already satisfied: jsonpickle>=1.2 in /home/tomek/miniconda3/lib/python3.9/site-packages (from sacred) (3.0.1)\n",
"Requirement already satisfied: colorama>=0.4 in /home/tomek/miniconda3/lib/python3.9/site-packages (from sacred) (0.4.6)\n",
"Requirement already satisfied: GitPython in /home/tomek/miniconda3/lib/python3.9/site-packages (from sacred) (3.1.31)\n",
"Requirement already satisfied: py-cpuinfo>=4.0 in /home/tomek/miniconda3/lib/python3.9/site-packages (from sacred) (9.0.0)\n",
"Requirement already satisfied: wrapt<2.0,>=1.0 in /home/tomek/miniconda3/lib/python3.9/site-packages (from sacred) (1.15.0)\n",
"Requirement already satisfied: munch<3.0,>=2.5 in /home/tomek/miniconda3/lib/python3.9/site-packages (from sacred) (2.5.0)\n",
"Requirement already satisfied: docopt<1.0,>=0.3 in /home/tomek/miniconda3/lib/python3.9/site-packages (from sacred) (0.6.2)\n",
"Requirement already satisfied: packaging>=18.0 in /home/tomek/miniconda3/lib/python3.9/site-packages (from sacred) (23.0)\n",
"Requirement already satisfied: six in /home/tomek/miniconda3/lib/python3.9/site-packages (from munch<3.0,>=2.5->sacred) (1.16.0)\n",
"Requirement already satisfied: gitdb<5,>=4.0.1 in /home/tomek/miniconda3/lib/python3.9/site-packages (from GitPython->sacred) (4.0.10)\n",
"Requirement already satisfied: smmap<6,>=3.0.1 in /home/tomek/miniconda3/lib/python3.9/site-packages (from gitdb<5,>=4.0.1->GitPython->sacred) (5.0.0)\n"
]
}
],
"source": [
"!pip3 install sacred"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"## Funkcja main"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Overwriting sacred_hello.py\n"
]
}
],
"source": [
"%%writefile sacred_hello.py\n",
"from sacred import Experiment\n",
"\n",
"ex = Experiment()\n",
"\n",
"@ex.automain\n",
"def my_main():\n",
" print('Witaj świecie!')\n"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {
"slideshow": {
"slide_type": "fragment"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"WARNING - sacred_hello - No observers have been added to this run\r\n",
"INFO - sacred_hello - Running command 'my_main'\r\n",
"INFO - sacred_hello - Started\r\n",
"Witaj świecie!\r\n",
"INFO - sacred_hello - Completed after 0:00:00\r\n"
]
}
],
"source": [
"!python3 IUM_07/sacred_hello.py"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"##### Co się dzieje w kodzie powyżej?\n",
"1. Tworzymy obiekt klasy Experiment\n",
"2. Dekorujemy funkcję \"my_main\" dekoratorem [automain](https://sacred.readthedocs.io/en/stable/apidoc.html#sacred.Experiment.automain)\n",
" Dzięki temu:\n",
" - otrzymujemy interfejs CLI, m.in. do kontrolowania poziomu logowania, przekazywania parametrów itp.\n",
" - oznaczamy funkcję \"my_main\" jako główną funkcję, która będzie wywoływana podczas wykonywania eksperymentu\n",
" - funkcja oznaczona jako główna musi być ostatnią funkcją zdefiniowaną w pliku!\n",
" "
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"##### Co nam daje interejs CLI:"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {
"slideshow": {
"slide_type": "fragment"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Usage:\r\n",
" sacred_hello.py [(with UPDATE...)] [options]\r\n",
" sacred_hello.py help [COMMAND]\r\n",
" sacred_hello.py (-h | --help)\r\n",
" sacred_hello.py COMMAND [(with UPDATE...)] [options]\r\n",
"\r\n",
"\r\n",
"\r\n",
"Options:\r\n",
" -b VALUE --beat-interval=VALUE Set the heart-beat interval for this run. Time\r\n",
" between two heartbeat events is measured in\r\n",
" seconds.\r\n",
" -C VALUE --capture=VALUE Control the way stdout and stderr are captured.\r\n",
" The argument value must be one of [no, sys, fd]\r\n",
" -c VALUE --comment=VALUE Add a comment to this run.\r\n",
" -d --debug Set this run to debug mode. Suppress warnings\r\n",
" about missing observers and don't filter the\r\n",
" stacktrace. Also enables usage with ipython\r\n",
" `--pdb`.\r\n",
" -e --enforce_clean Fail if any version control repository is\r\n",
" dirty.\r\n",
" -F VALUE --file_storage=VALUE Add a file-storage observer to the experiment.\r\n",
" The value of the arguement should be the base-\r\n",
" directory to write the runs to\r\n",
" -f --force Disable warnings about suspicious changes for\r\n",
" this run.\r\n",
" -h --help Print this help message and exit.\r\n",
" -i VALUE --id=VALUE Set the id for this run.\r\n",
" -l VALUE --loglevel=VALUE Set the LogLevel. Loglevel either as 0 - 50 or\r\n",
" as string: DEBUG(10), INFO(20), WARNING(30),\r\n",
" ERROR(40), CRITICAL(50)\r\n",
" -m VALUE --mongo_db=VALUE Add a MongoDB Observer to the experiment. The\r\n",
" argument value is the database specification.\r\n",
" Should be in the form: `[host:port:]db_name[.c\r\n",
" ollection[:id]][!priority]`\r\n",
" -n VALUE --name=VALUE Set the name for this run.\r\n",
" -D --pdb Automatically enter post-mortem debugging with\r\n",
" pdb on failure.\r\n",
" -p --print-config Always print the configuration first.\r\n",
" -P VALUE --priority=VALUE Sets the priority for a queued up experiment.\r\n",
" `--priority=NUMBER` The number represent the\r\n",
" priority for this run.\r\n",
" -q --queue Only queue this run, do not start it.\r\n",
" -S VALUE --s3=VALUE Add a S3 File observer to the experiment. The\r\n",
" argument value should be\r\n",
" `s3:///path/to/exp`.\r\n",
" -s VALUE --sql=VALUE Add a SQL Observer to the experiment. The\r\n",
" typical form is:\r\n",
" dialect://username:password@host:port/database\r\n",
" -t VALUE --tiny_db=VALUE Add a TinyDB Observer to the experiment. The\r\n",
" argument is the path to be given to the\r\n",
" TinyDbObserver.\r\n",
" -u --unobserved Ignore all observers for this run.\r\n",
"\r\n",
"\r\n",
"Arguments:\r\n",
" COMMAND Name of command to run (see below for list of commands)\r\n",
" UPDATE Configuration assignments of the form foo.bar=17\r\n",
"\r\n",
"\r\n",
"Commands:\r\n",
" print_config Print the updated configuration and exit.\r\n",
" print_dependencies Print the detected source-files and dependencies.\r\n",
" save_config Store the updated configuration in a file.\r\n",
" print_named_configs Print the available named configs and exit.\r\n",
" my_main \r\n",
"\r\n"
]
}
],
"source": [
"!python3 IUM_07/sacred_hello.py -h"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"## Konfiguracje\n",
" - Konfiguracje pozwalają nam sparametryzować wywołania eksperymentu.\n",
" - Ułatwiają przekazywanie parametrów - zmienne z konfiguracji są wstrzykiwane do funkcji wywoływanych \n",
" - Mogą być automatycznie zapisywane (dzięki czemu możemy śledzić jak zmieniały się parametry i jaki miały wpływ na wyniki)\n",
" - Konfigurację można stworzyć w jeden z 3 sposobów:\n",
" - używając config scopes (z dekoratorem `@config`)\n",
" - jako słownik\n",
" - wczytując ją z pliku"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"### Konfiguracje - config scopes\n",
"Jeśli oznaczymy jakąś funkcję dekoratorem `@config`, to zostanie ona uruchoniona przed wywołaniem eksperymentu i wszystkie jej lokalne zmienne, które da się zserializować jako json, zostaną dodane do konfiguracji. Potem ich wartości zostaną wstrzyknięte do innych funkcji wywoływanych w eksperymencie. "
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [],
"source": [
"from sacred import Experiment\n",
"\n",
"exint = Experiment(\"sacred_scopes\", interactive=True) #Jeśli wykonujemy interaktywnie (w konsoli Pythona albo w Jupyter):\n",
"# - musimy podać nazwę eksperymentu (domyślnie jako nazwa używana jest nazwa pliku źródłowego)\n",
"# - musimy dodać parametr \"interactive=True\"\n",
"# - zamiast dekoratora \"@ex.automain\" używamy \"@ex.main\"\n",
"\n",
"@exint.config\n",
"def my_config():\n",
" recipient = \"Świecie\"\n",
" greeting = \"Witaj\"\n",
" message = \"{0} {1}!\".format(greeting, recipient)\n",
"\n",
"\n",
"@exint.main\n",
"def my_main(message):\n",
" print(message)"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {
"slideshow": {
"slide_type": "fragment"
}
},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"WARNING - sacred_scopes - No observers have been added to this run\n",
"INFO - sacred_scopes - Running command 'my_main'\n",
"INFO - sacred_scopes - Started\n",
"INFO - sacred_scopes - Completed after 0:00:00\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Witaj Świecie!\n"
]
},
{
"data": {
"text/plain": [
""
]
},
"execution_count": 9,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"exint.run()"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"##### Możemy podejrzeć wartości zmiennych w konfiguracji:"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {
"slideshow": {
"slide_type": "fragment"
}
},
"outputs": [
{
"data": {
"text/plain": [
"{'recipient': 'Świecie', 'greeting': 'Witaj', 'message': 'Witaj Świecie!'}"
]
},
"execution_count": 10,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"my_config()"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"#### Parametry możemy podejrzeć i modyfikować z poziomu CLI\n",
" - wartości podane w CLI nadpiszą te podane w kodzie"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"slideshow": {
"slide_type": "fragment"
}
},
"outputs": [],
"source": [
"# %load IUM_07/sacred_scopes.py\n",
"from sacred import Experiment\n",
"\n",
"ex = Experiment()\n",
"\n",
"@ex.config\n",
"def my_config():\n",
" recipient = \"Świecie\"\n",
" greeting = \"Witaj\"\n",
" message = \"{0} {1}!\".format(greeting, recipient)\n",
"\n",
"@ex.automain\n",
"def my_main(message):\n",
" print(message)"
]
},
{
"cell_type": "code",
"execution_count": 14,
"metadata": {
"slideshow": {
"slide_type": "fragment"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"WARNING - sacred_scopes - No observers have been added to this run\r\n",
"INFO - sacred_scopes - Running command 'my_main'\r\n",
"INFO - sacred_scopes - Started\r\n",
"Witaj Przygodo!\r\n",
"INFO - sacred_scopes - Completed after 0:00:00\r\n"
]
}
],
"source": [
"!python3 IUM_07/sacred_scopes.py with 'recipient=Przygodo'"
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {
"slideshow": {
"slide_type": "fragment"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"INFO - sacred_scopes - Running command 'print_config'\r\n",
"INFO - sacred_scopes - Started\r\n",
"Configuration (\u001b[34mmodified\u001b[0m, \u001b[32madded\u001b[0m, \u001b[31mtypechanged\u001b[0m, \u001b[2mdoc\u001b[0m):\r\n",
" greeting = 'Witaj'\r\n",
" message = 'Witaj Świecie!'\r\n",
" recipient = 'Świecie'\r\n",
" seed = 269258424 \u001b[2m# the random seed for this experiment\u001b[0m\r\n",
"INFO - sacred_scopes - Completed after 0:00:00\r\n"
]
}
],
"source": [
"!python3 IUM_07/sacred_scopes.py print_config"
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {
"slideshow": {
"slide_type": "fragment"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"INFO - sacred_scopes - Running command 'print_config'\r\n",
"INFO - sacred_scopes - Started\r\n",
"Configuration (\u001b[34mmodified\u001b[0m, \u001b[32madded\u001b[0m, \u001b[31mtypechanged\u001b[0m, \u001b[2mdoc\u001b[0m):\r\n",
" greeting = 'Witaj'\r\n",
" message = 'Witaj Przygodo!'\r\n",
"\u001b[34m recipient = 'Przygodo'\u001b[0m\r\n",
" seed = 667939214 \u001b[2m# the random seed for this experiment\u001b[0m\r\n",
"INFO - sacred_scopes - Completed after 0:00:00\r\n"
]
}
],
"source": [
"!python IUM_07/sacred_scopes.py print_config with 'recipient=Przygodo'"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"### Wczytywanie konfiguracji z pliku"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"slideshow": {
"slide_type": "fragment"
}
},
"outputs": [],
"source": [
"# %load IUM_07/config.json\n",
"{\n",
" \"recipient\": \"samotności\",\n",
" \"greeting\": \"Żegnaj\"\n",
"}"
]
},
{
"cell_type": "code",
"execution_count": 20,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [],
"source": [
"from sacred import Experiment\n",
"\n",
"ex = Experiment(\"sacred_scopes\", interactive=True) #Jeśli wykonujemy interaktywnie (w konsoli Pythona albo w Jupyter):\n",
"# - musimy podać nazwę eksperymentu (domyślnie jako nazwa używana jest nazwa pliku źródłowego)\n",
"# - musimy dodać parametr \"interactive=True\"\n",
"# - zamiast \"automain\" używamy parametru \"main\"\n",
"\n",
"@ex.config\n",
"def my_config():\n",
" recipient = \"Świecie\"\n",
" greeting = \"Witaj\"\n",
"\n",
"ex.add_config(\"IUM_07/config.json\")\n",
"\n",
"\n",
"@ex.main\n",
"def my_main(recipient, greeting):\n",
" print(\"{0} {1}!\".format(greeting, recipient))"
]
},
{
"cell_type": "code",
"execution_count": 21,
"metadata": {
"slideshow": {
"slide_type": "fragment"
}
},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"WARNING - sacred_scopes - No observers have been added to this run\n",
"INFO - sacred_scopes - Running command 'my_main'\n",
"INFO - sacred_scopes - Started\n",
"INFO - sacred_scopes - Completed after 0:00:00\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Żegnaj samotności!\n"
]
}
],
"source": [
"r = ex.run()"
]
},
{
"cell_type": "code",
"execution_count": 22,
"metadata": {
"slideshow": {
"slide_type": "fragment"
}
},
"outputs": [
{
"data": {
"text/plain": [
"{'recipient': 'samotności', 'greeting': 'Żegnaj', 'seed': 877272352}"
]
},
"execution_count": 22,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"r.config"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"#### Możemy modyfikować części konfiguracji bezpośrednio przed wywołaniem"
]
},
{
"cell_type": "code",
"execution_count": 23,
"metadata": {
"slideshow": {
"slide_type": "fragment"
}
},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"WARNING - sacred_scopes - No observers have been added to this run\n",
"INFO - sacred_scopes - Running command 'my_main'\n",
"INFO - sacred_scopes - Started\n",
"INFO - sacred_scopes - Completed after 0:00:00\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Żegnaj nudo!\n"
]
}
],
"source": [
"r = ex.run(config_updates={\"recipient\":\"nudo\"})"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"#### Wstrzykiwanie zależności\n",
" - Oprócz funkcji głównej, wartości z konfiguracji są też wstrzykiwane do funkcji udekorowanych dekoratorem `@ex.capture`\n",
" - Możemy korzystać w nich ze specjalnych parametrów, np.:\n",
" - `_log` - daje nam dostęp do obiektu logera (więcej: [logowanie](https://sacred.readthedocs.io/en/stable/logging.html))\n",
" - `_run` - daje dostęp do obiektu reprezentującego aktualne wywołanie eksperymentu (przykład później)"
]
},
{
"cell_type": "code",
"execution_count": 24,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"WARNING - sacred_scopes - No observers have been added to this run\n",
"INFO - sacred_scopes - Running command 'my_main'\n",
"INFO - sacred_scopes - Started\n",
"INFO - prepare_message - Enterred prepare_message\n",
"INFO - sacred_scopes - Completed after 0:00:00\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Witaj Świecie!\n"
]
},
{
"data": {
"text/plain": [
""
]
},
"execution_count": 24,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"from sacred import Experiment\n",
"\n",
"ex = Experiment(\"sacred_scopes\", interactive=True)\n",
"\n",
"@ex.config\n",
"def my_config():\n",
" recipient = \"Świecie\"\n",
" greeting = \"Witaj\"\n",
"\n",
"@ex.capture\n",
"def prepare_message(recipient, greeting, _log):\n",
" _log.info(\"Enterred prepare_message\")\n",
" return \"{0} {1}!\".format(greeting, recipient)\n",
"\n",
"@ex.main\n",
"def my_main():\n",
" print(prepare_message()) ## Nie musimy przekazywać wartości\n",
" \n",
"ex.run()"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"### Obserwowanie eksperymentów\n",
"Sacred zapisuje szereg informacji na temat każdego eksperymentu:\n",
" - czas wykonania\n",
" - konfigurację\n",
" - tekst zwrócony na stdout/stderr\n",
" - błędy, jeśli wystąpiły\n",
" - podstawowe informacje o środowisku (maszynie), na której przeprowadzono eksperyment\n",
" - użyte pliki źródłowe\n",
" - użyte zależności i ich wersje\n",
" - pliki otwarte za pomocą [ex.open_resource()](https://sacred.readthedocs.io/en/stable/apidoc.html#sacred.Experiment.open_resource) albo [ex.add_resource()](https://sacred.readthedocs.io/en/stable/apidoc.html#sacred.Experiment.add_resource)\n",
" - pliki dodane za pomocą [ex.add_artifact()](https://sacred.readthedocs.io/en/stable/apidoc.html#sacred.run.Run.add_artifact)"
]
},
{
"cell_type": "code",
"execution_count": 28,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"total 20\r\n",
"drwsrwsr-t 2 tomek tomek 4096 May 28 2022 1\r\n",
"drwsrwsr-t 2 tomek tomek 4096 May 28 2022 2\r\n",
"drwxr-sr-x 2 tomek tomek 4096 Apr 12 15:11 3\r\n",
"drwxr-sr-x 2 tomek tomek 4096 Apr 12 15:11 _resources\r\n",
"drwsrwsr-t 2 tomek tomek 4096 May 28 2022 _sources\r\n"
]
}
],
"source": [
"!ls -l my_runs"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"Obserwowane infromacje mogą zostać zapisane za pomocą jednego z [obserwatorów](https://sacred.readthedocs.io/en/stable/observers.html):\n",
" - Mongo Observer - zapisuje dane w MongoDB\n",
" - File Storage Observer - zapisuje dane lokalnie w pliku\n",
" - TinyDB Observer - korzysta z lokalnej bazy zapisanej w pliku JSON\n",
" - SQL Observer - przechowuje informacje w bazie SQL\n",
" - S3 Observer - korzysta z AWS S3\n",
" - gcs_observer - korzysta z Google Cloud Storage\n",
" - Queue Observer - rodzaj lokalnego bufora nakładanego na jeden z powyższych\n",
" - Slack Observer - używany do powiadomień wysyłanych na komunikator Slack\n",
" - Telegram Observer - używany do powiadomień wysyłanych na komunikator Telegram"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"### File storage observer\n",
"- zapisuje informacje o eksperymencie w lokalnych plikach \n",
"- można go dodać tak: `ex.observers.append(FileStorageObserver('my_runs_directory'))`, gdzie `my_runs_directory` to ścieżka, gdzie będą zapisywane informacje o eksperymentach"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [],
"source": [
"%%writefile IUM_07/file_observer.py\n",
"from sacred.observers import FileStorageObserver\n",
"from sacred import Experiment\n",
"\n",
"ex = Experiment(\"file_observer\")\n",
"\n",
"ex.observers.append(FileStorageObserver('my_runs'))\n",
"\n",
"@ex.config\n",
"def my_config():\n",
" recipient = \"Świecie\"\n",
" greeting = \"Witaj\"\n",
"\n",
"@ex.capture\n",
"def prepare_message(recipient, greeting):\n",
" return \"{0} {1}!\".format(greeting, recipient)\n",
"\n",
"@ex.automain\n",
"def my_main(recipient, greeting):\n",
" print(prepare_message()) ## Nie musimy przekazywać wartości"
]
},
{
"cell_type": "code",
"execution_count": 29,
"metadata": {
"slideshow": {
"slide_type": "fragment"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"INFO - file_observer - Running command 'my_main'\r\n",
"INFO - file_observer - Started run with ID \"4\"\r\n",
"Witaj Świecie!\r\n",
"INFO - file_observer - Completed after 0:00:00\r\n"
]
}
],
"source": [
"!python3 IUM_07/file_observer.py"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"#### Zobaczmy jakie informacje zostały zapisane"
]
},
{
"cell_type": "code",
"execution_count": 30,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"total 24\r\n",
"drwsrwsr-t 2 tomek tomek 4096 May 28 2022 1\r\n",
"drwsrwsr-t 2 tomek tomek 4096 May 28 2022 2\r\n",
"drwxr-sr-x 2 tomek tomek 4096 Apr 12 15:11 3\r\n",
"drwxr-sr-x 2 tomek tomek 4096 Apr 20 12:09 4\r\n",
"drwxr-sr-x 2 tomek tomek 4096 Apr 12 15:11 _resources\r\n",
"drwsrwsr-t 2 tomek tomek 4096 May 28 2022 _sources\r\n"
]
}
],
"source": [
"!ls -l my_runs"
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"total 16\r\n",
"-rw-r--r-- 1 tomek tomek 77 May 28 2022 config.json\r\n",
"-rw-r--r-- 1 tomek tomek 159 May 28 2022 cout.txt\r\n",
"-rw-r--r-- 1 tomek tomek 2 May 28 2022 metrics.json\r\n",
"-rw-r--r-- 1 tomek tomek 1659 May 28 2022 run.json\r\n"
]
}
],
"source": [
"!ls -l my_runs/1"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [],
"source": [
"# %load my_runs/1/config.json\n",
"{\n",
" \"greeting\": \"Witaj\",\n",
" \"recipient\": \"\\u015awiecie\",\n",
" \"seed\": 805857632\n",
"}"
]
},
{
"cell_type": "code",
"execution_count": 31,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"INFO - file_observer - Running command 'my_main'\r\n",
"INFO - file_observer - Started run with ID \"1\"\r\n",
"Witaj Świecie!\r\n",
"INFO - file_observer - Completed after 0:00:00\r\n"
]
}
],
"source": [
"!cat my_runs/1/cout.txt"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [],
"source": [
"# %load my_runs/1/run.json\n",
"{\n",
" \"artifacts\": [],\n",
" \"command\": \"my_main\",\n",
" \"experiment\": {\n",
" \"base_dir\": \"/home/tomek/repos/aitech/aitech-ium/IUM_07\",\n",
" \"dependencies\": [\n",
" \"sacred==0.8.2\"\n",
" ],\n",
" \"mainfile\": \"file_observer.py\",\n",
" \"name\": \"file_observer\",\n",
" \"repositories\": [\n",
" {\n",
" \"commit\": \"3055a4f1c2ef06ea1c29e3d41d862827cede7e2a\",\n",
" \"dirty\": true,\n",
" \"url\": \"git@git.wmi.amu.edu.pl:tzietkiewicz/aitech-ium.git\"\n",
" }\n",
" ],\n",
" \"sources\": [\n",
" [\n",
" \"file_observer.py\",\n",
" \"_sources/file_observer_cd34a0ef4a32fb0a966eaa01ea6371ad.py\"\n",
" ]\n",
" ]\n",
" },\n",
" \"heartbeat\": \"2022-04-25T07:51:37.853633\",\n",
" \"host\": {\n",
" \"ENV\": {},\n",
" \"cpu\": \"Intel(R) Core(TM) i5-4200H CPU @ 2.80GHz\",\n",
" \"hostname\": \"ASUSEK\",\n",
" \"os\": [\n",
" \"Linux\",\n",
" \"Linux-4.4.0-19041-Microsoft-x86_64-with-Ubuntu-18.04-bionic\"\n",
" ],\n",
" \"python_version\": \"3.6.9\"\n",
" },\n",
" \"meta\": {\n",
" \"command\": \"my_main\",\n",
" \"options\": {\n",
" \"--beat-interval\": null,\n",
" \"--capture\": null,\n",
" \"--comment\": null,\n",
" \"--debug\": false,\n",
" \"--enforce_clean\": false,\n",
" \"--file_storage\": null,\n",
" \"--force\": false,\n",
" \"--help\": false,\n",
" \"--loglevel\": null,\n",
" \"--mongo_db\": null,\n",
" \"--name\": null,\n",
" \"--pdb\": false,\n",
" \"--print-config\": false,\n",
" \"--priority\": null,\n",
" \"--queue\": false,\n",
" \"--s3\": null,\n",
" \"--sql\": null,\n",
" \"--tiny_db\": null,\n",
" \"--unobserved\": false,\n",
" \"COMMAND\": null,\n",
" \"UPDATE\": [],\n",
" \"help\": false,\n",
" \"with\": false\n",
" }\n",
" },\n",
" \"resources\": [],\n",
" \"result\": null,\n",
" \"start_time\": \"2022-04-25T07:51:37.831461\",\n",
" \"status\": \"COMPLETED\",\n",
" \"stop_time\": \"2022-04-25T07:51:37.849334\"\n",
"}"
]
},
{
"cell_type": "code",
"execution_count": 32,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"total 4\r\n",
"-rw-r--r-- 1 tomek tomek 464 May 28 2022 file_observer_cd34a0ef4a32fb0a966eaa01ea6371ad.py\r\n"
]
}
],
"source": [
"! ls -l my_runs/_sources\n",
"## W run.json możemy znaleźć ścieżkę do pliku z źródłami: \"_sources/file_observer_bb0a5c4720d1072b641d23da080696b6.py\"\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [],
"source": [
"# %load my_runs/_sources/file_observer_cd34a0ef4a32fb0a966eaa01ea6371ad.py\n",
"from sacred.observers import FileStorageObserver\n",
"from sacred import Experiment\n",
"\n",
"ex = Experiment(\"file_observer\")\n",
"\n",
"ex.observers.append(FileStorageObserver('my_runs'))\n",
"\n",
"@ex.config\n",
"def my_config():\n",
" recipient = \"Świecie\"\n",
" greeting = \"Witaj\"\n",
"\n",
"@ex.capture\n",
"def prepare_message(recipient, greeting):\n",
" return \"{0} {1}!\".format(greeting, recipient)\n",
"\n",
"@ex.automain\n",
"def my_main(recipient, greeting):\n",
" print(prepare_message()) ## Nie musimy przekazywać wartości\n"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"### Dodawanie własnych informacji\n"
]
},
{
"cell_type": "code",
"execution_count": 36,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"INFO - file_observer - Running command 'my_main'\n",
"INFO - file_observer - Started run with ID \"5\"\n",
"INFO - file_observer - Completed after 0:00:00\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Witaj Świecie!\n"
]
}
],
"source": [
"from sacred.observers import FileStorageObserver\n",
"from sacred import Experiment\n",
"from datetime import datetime\n",
"\n",
"ex = Experiment(\"file_observer\", interactive=True)\n",
"\n",
"ex.observers.append(FileStorageObserver('my_runs'))\n",
"\n",
"@ex.config\n",
"def my_config():\n",
" recipient = \"Świecie\"\n",
" greeting = \"Witaj\"\n",
"\n",
"### - Do \"przechwyconej\" przez @ex.capture funkcji prepare_message dodaliśmy specjalny parametr _run\n",
"### - Daje on dostęp do obiektu wywołania eksperymentu w trakcie jego wywołania\n",
"### - umożliwia m.in. zapisywanie dodatkowych informacji w słowniku info\n",
"@ex.capture\n",
"def prepare_message(recipient, greeting, _run):\n",
" _run.info[\"prepare_message_ts\"] = str(datetime.now())\n",
" return \"{0} {1}!\".format(greeting, recipient)\n",
"\n",
"@ex.main\n",
"def my_main(recipient, greeting):\n",
" print(prepare_message()) ## Nie musimy przekazywać wartości\n",
" \n",
"r = ex.run()"
]
},
{
"cell_type": "code",
"execution_count": 37,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{\r\n",
" \"prepare_message_ts\": \"2023-04-20 12:10:28.197315\"\r\n",
"}"
]
}
],
"source": [
"cat my_runs/5/info.json"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"### Artefakty\n",
"\n",
"- Artefakty służą do zapisywania plików, np. z wytrenowanym modelem\n",
"- Plik można zapisać jako artefakt korzystając z : [ex.add_artifact()](https://sacred.readthedocs.io/en/stable/apidoc.html?highlight=artifact#sacred.Experiment.add_artifact)\n",
"```python\n",
"ex.add_artifact(\"model.pb\")\n",
"```"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Otwieranie zasobów\n",
"- Zmiana danych wejściowych wpłwa w oczywisty sposób na wyniki\n",
"- Dlatego warto śledzić te zmiany za pomocą:\n",
" - [ex.open_resource()](https://sacred.readthedocs.io/en/stable/apidoc.html#sacred.Experiment.open_resource)\n",
" - [ex.add_resource()](https://sacred.readthedocs.io/en/stable/apidoc.html#sacred.Experiment.add_resource)"
]
},
{
"cell_type": "code",
"execution_count": 38,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"INFO - resources - Running command 'my_main'\n",
"INFO - resources - Started run with ID \"6\"\n",
"INFO - resources - Completed after 0:00:00\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Id,SepalLengthCm,SepalWidthCm,PetalLengthCm,PetalWidthCm,Species\n",
"\n"
]
},
{
"data": {
"text/plain": [
""
]
},
"execution_count": 38,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"from sacred import Experiment\n",
"from sacred.observers import FileStorageObserver\n",
"\n",
"\n",
"ex = Experiment(\"resources\", interactive=True)\n",
"ex.observers.append(FileStorageObserver('my_runs'))\n",
"\n",
"@ex.main\n",
"def my_main():\n",
" f = ex.open_resource(\"Iris.csv\", \"r\")\n",
" print(f.readline())\n",
" \n",
"ex.run()"
]
},
{
"cell_type": "code",
"execution_count": 39,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"total 8\r\n",
"-rw-r--r-- 1 tomek tomek 5107 Apr 12 15:11 Iris_717820ef0af287ff346c5cabfb4c612c.csv\r\n"
]
}
],
"source": [
"!ls -l my_runs/_resources"
]
},
{
"cell_type": "code",
"execution_count": 41,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"my_runs/6/run.json: \"name\": \"resources\",\r\n",
"my_runs/6/run.json: \"resources\": [\r\n",
"my_runs/6/run.json: \"my_runs/_resources/Iris_717820ef0af287ff346c5cabfb4c612c.csv\"\r\n"
]
}
],
"source": [
"!grep -e \"resources\" -R my_runs/6"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
" ## Obserwator mongo\n",
" - Żeby skorzystać z obserwatora Mongo, musimy mieć dostęp do bazy Mongo.\n",
" - Można ją łatwo \"postawić\" za pomocą [docker-compose ](https://docs.docker.com/compose/).\n",
" - W tym celu wystarczy skopiować katalog [examples/docker](https://github.com/IDSIA/sacred/tree/master/examples/docker) z repozytorium SACRED i uruchomić `docker-compose up` - dostaniemy uruchomioną bazę MongoDB i dodatkowo [Omniboard ](https://vivekratnavel.github.io/omniboard/#/). Więcej informacji w [dokumentacji](https://sacred.readthedocs.io/en/stable/examples.html#docker-setup)\n",
" - Baza taka została już postawiona na serwerze Jenkins, więc pracując na Jenkinsie można skorzystać z lokalnej bazy (`localhost:27017`)"
]
},
{
"cell_type": "code",
"execution_count": 42,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Requirement already satisfied: pymongo in /home/tomek/miniconda3/lib/python3.9/site-packages (4.3.3)\r\n",
"Requirement already satisfied: dnspython<3.0.0,>=1.16.0 in /home/tomek/miniconda3/lib/python3.9/site-packages (from pymongo) (2.3.0)\r\n"
]
}
],
"source": [
"!pip3 install pymongo"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"INFO - sacred_scopes - Running command 'my_main'\n",
"ERROR - sacred_scopes - Failed after 0:00:30!\n"
]
}
],
"source": [
"from sacred.observers import MongoObserver\n",
"from sacred import Experiment\n",
"\n",
"ex = Experiment(\"sacred_scopes\", interactive=True)\n",
"ex.observers.append(MongoObserver(url='mongodb://admin:IUM_2021@172.17.0.1:27017',\n",
" db_name='sacred')) # Tutaj podajemy dane uwierzytelniające i nazwę bazy skonfigurowane w pliku .env podczas uruchamiania bazy.\n",
"# W przypadku instancji na Jenkinsie url będzie wyglądał następująco: mongodb://admin:IUM_2021@172.17.0.1:27017\n",
"@ex.config\n",
"def my_config():\n",
" recipient = \"Świecie\"\n",
" greeting = \"Witaj\"\n",
"\n",
"@ex.capture\n",
"def prepare_message(recipient, greeting):\n",
" return \"{0} {1}!\".format(greeting, recipient)\n",
"\n",
"@ex.main\n",
"def my_main(recipient, greeting):\n",
" print(prepare_message()) ## Nie musimy przekazywać wartości\n",
" \n",
"ex.run()"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"- Informacje o eksperymencie można obejrzeć na Omniboard: http://127.0.0.1:9000/sacred\n",
"- Instancja na Jenkinsie: http://tzietkiewicz.vm.wmi.amu.edu.pl:9000/sacred\n",
"
"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"## Metryki\n",
"\n",
"- W trakcie eksperymentu możemy śledzić [metryki](https://sacred.readthedocs.io/en/stable/collected_information.html#metrics-api), np. aktualny loss\n",
"- W tym celu wystarczy:\n",
" - dodać do funkcji udekorowanej `@ex.main` albo `@ex.capure` parametr `_run`\n",
" - potem wywołać np. `_run.log_scalar()`"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"outputs": [],
"source": [
"from sacred.observers import MongoObserver\n",
"from sacred import Experiment\n",
"import random\n",
"import time\n",
"\n",
"ex = Experiment(\"sacred_scopes\", interactive=True)\n",
"ex.observers.append(MongoObserver(url='mongodb://admin:IUM_2021@172.17.0.1:27017',\n",
" db_name='sacred')) # Tutaj podajemy dane uwierzytelniające i nazwę bazy skonfigurowane w pliku .env podczas uruchamiania bazy.\n",
"# W przypadku instancji na Jenkinsie url będzie wyglądał następująco: mongodb://admin:IUM_2021@172.17.0.1:27017\n",
"@ex.config\n",
"def my_config():\n",
" recipient = \"Świecie\"\n",
" greeting = \"Witaj\"\n",
"\n",
"@ex.capture\n",
"def prepare_message(recipient, greeting):\n",
" return \"{0} {1}!\".format(greeting, recipient)\n",
"\n",
"@ex.main\n",
"def my_main(recipient, greeting, _run):\n",
" print(prepare_message()) ## Nie musimy przekazywać wartości \n",
" counter = 0\n",
" while counter < 20:\n",
" counter+=1\n",
" value = counter\n",
" ms_to_wait = random.randint(5, 5000)\n",
" time.sleep(ms_to_wait/1000)\n",
" noise = 1.0 + 0.1 * (random.randint(0, 10) - 5)\n",
" # This will add an entry for training.loss metric in every second iteration.\n",
" # The resulting sequence of steps for training.loss will be 0, 2, 4, ...\n",
" if counter % 2 == 0:\n",
" _run.log_scalar(\"training.loss\", value * 1.5 * noise, counter)\n",
" # Implicit step counter (0, 1, 2, 3, ...)\n",
" # incremented with each call for training.accuracy:\n",
" _run.log_scalar(\"training.accuracy\", value * 2 * noise)\n",
"\n",
"ex.run() "
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"#### Wartości metryk możemy na żywo śledzić w Omniboard\n",
"
"
]
},
{
"cell_type": "markdown",
"metadata": {
"slideshow": {
"slide_type": "slide"
}
},
"source": [
"## Zadanie [15 pkt] (do 2023-06-?)\n",
"1. \"Owiń\" wywołanie swojego eksperymentu za pomocą Sacred, w ten sposób, żeby zapisane zostały [10pkt]:\n",
" - parametry, z którymi wywołany był trening\n",
" - powstały plik z modelem (jako artefakt)\n",
" - kod źródłowy użyty do przeprowadzenia treningu\n",
" - pliki wejściowe otwarte za pomocą open_resource\n",
" - metryki\n",
" \n",
"Jako nazwę eksperymentu użyj swojego numeru indeksu tak, żebyś mogła/mógł je odnaleźć w Omniboard\n",
"\n",
"2. Wykorzystaj 2 obserwatory [5pkt]: \n",
" - MongoObserver, skorzytaj nastęþującego URL: `mongodb://admin:IUM_2021@172.17.0.1:27017` (będziesz mógł przeglądać wyniki na http://tzietkiewicz.vm.wmi.amu.edu.pl:9000/sacred)\n",
" - FileObserver - zapisane pliki zarchiwizuj na Jenkinsie jako jego artefakty\n"
]
}
],
"metadata": {
"author": "Tomasz Ziętkiewicz",
"celltoolbar": "Slideshow",
"email": "tomasz.zietkiewicz@amu.edu.pl",
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"lang": "pl",
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.9.12"
},
"slideshow": {
"slide_type": "slide"
},
"subtitle": "7.Sacred[laboratoria]",
"title": "Inżynieria uczenia maszynowego",
"toc": {
"base_numbering": 1,
"nav_menu": {},
"number_sections": false,
"sideBar": false,
"skip_h1_title": false,
"title_cell": "Table of Contents",
"title_sidebar": "Contents",
"toc_cell": false,
"toc_position": {},
"toc_section_display": false,
"toc_window_display": false
},
"year": "2021"
},
"nbformat": 4,
"nbformat_minor": 4
}