{ "cells": [ { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "## ![Logo 1](https://git.wmi.amu.edu.pl/AITech/Szablon/raw/branch/master/Logotyp_AITech1.jpg)\n", "
\n", "

Inżynieria uczenia maszynowego

\n", "

8. MLFlow [laboratoria]

\n", "

Tomasz Ziętkiewicz (2023)

\n", "
\n", "\n", "![Logo 2](https://git.wmi.amu.edu.pl/AITech/Szablon/raw/branch/master/Logotyp_AITech2.jpg)" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ " ## MLflow\n", "
\n", "\n", " - https://mlflow.org/\n", " - Narzędzie podobne do omawianego na poprzednich zajęciach Sacred\n", " - Nieco inne podejście: mniej ingerencji w istniejący kod\n", " - Bardziej kompleksowe rozwiązanie: 4 komponenty, pierwszy z nich ma funkcjonalność podobną do Sacred\n", " - Działa \"z każdym\" językiem. A tak naprawdę: Python, R, Java + CLI API + REST API\n", " - Popularna wśród pracodawców - wyniki wyszukiwania ofert pracy: 20 ofert (https://pl.indeed.com/), 36 ofert (linkedin). Sacred: 0\n", " - Integracja z licznymi bibliotekami / chmurami\n", " - Rozwiązanie OpenSource, stworzone przez firmę Databricks\n", " - Dostępna [odpłatna wersja \"Managed\"](https://databricks.com/product/managed-mlflow) (w ordóżnieniu od \"self-hosted\")\n" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "## Komponenty\n", "\n", "MLflow składa się z czterech niezależnych komponentów:\n", " - **MLflow Tracking** - pozwala śledzić zmiany parametrów, kodu, środowiska i ich wpływ na metryki. Jest to funkcjonalność bardzo zbliżona do tej, którą zapewnia Sacred\n", " - **MLflow Projects** - umożliwia \"pakowanie\" kodu ekserymentów w taki sposób, żeby mogłby być w łatwy sposób zreprodukowane przez innych\n", " - **MLflow Models** - ułatwia \"pakowanie\" modeli uczenia maszynowego\n", " - **MLflow Registry** - zapewnia centralne miejsce do przechowywania i współdzielenia modeli. Zapewnia narzędzia do wersjonowania i śledzenia pochodzenia tych modeli.\n", " \n", "Komponenty te mogą być używane razem bądź oddzielnie." ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "## MLflow Tracking - przykład\n", "(poniższe przykłady kodu trenującego pochodzą z tutoriala MLflow: https://mlflow.org/docs/latest/tutorials-and-examples/tutorial.html)" ] }, { "cell_type": "code", "execution_count": 2, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [], "source": [ "%%capture null\n", "!pip install mlflow\n", "!pip install sklearn" ] }, { "cell_type": "code", "execution_count": 2, "metadata": {}, "outputs": [], "source": [ "!mkdir -p IUM_08/examples/sklearn_elasticnet_wine/" ] }, { "cell_type": "code", "execution_count": 4, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Overwriting IUM_08/examples/sklearn_elasticnet_wine/train.py\n" ] } ], "source": [ "%%writefile IUM_08/examples/sklearn_elasticnet_wine/train.py\n", "# The data set used in this example is from http://archive.ics.uci.edu/ml/datasets/Wine+Quality\n", "# P. Cortez, A. Cerdeira, F. Almeida, T. Matos and J. Reis.\n", "# Modeling wine preferences by data mining from physicochemical properties. In Decision Support Systems, Elsevier, 47(4):547-553, 2009.\n", "\n", "import os\n", "import warnings\n", "import sys\n", "\n", "import pandas as pd\n", "import numpy as np\n", "from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score\n", "from sklearn.model_selection import train_test_split\n", "from sklearn.linear_model import ElasticNet\n", "from urllib.parse import urlparse\n", "import mlflow\n", "import mlflow.sklearn\n", "\n", "import logging\n", "\n", "logging.basicConfig(level=logging.WARN)\n", "logger = logging.getLogger(__name__)\n", "\n", "mlflow.set_tracking_uri(\"http://localhost:5000\")\n", "mlflow.set_experiment(\"s123456\")\n", "\n", "def eval_metrics(actual, pred):\n", " rmse = np.sqrt(mean_squared_error(actual, pred))\n", " mae = mean_absolute_error(actual, pred)\n", " r2 = r2_score(actual, pred)\n", " return rmse, mae, r2\n", "\n", "\n", "if __name__ == \"__main__\":\n", " warnings.filterwarnings(\"ignore\")\n", " np.random.seed(40)\n", "\n", " # Read the wine-quality csv file from the URL\n", " csv_url = (\n", " \"http://archive.ics.uci.edu/ml/machine-learning-databases/wine-quality/winequality-red.csv\"\n", " )\n", " try:\n", " data = pd.read_csv(csv_url, sep=\";\")\n", " except Exception as e:\n", " logger.exception(\n", " \"Unable to download training & test CSV, check your internet connection. Error: %s\", e\n", " )\n", "\n", " # Split the data into training and test sets. (0.75, 0.25) split.\n", " train, test = train_test_split(data)\n", "\n", " # The predicted column is \"quality\" which is a scalar from [3, 9]\n", " train_x = train.drop([\"quality\"], axis=1)\n", " test_x = test.drop([\"quality\"], axis=1)\n", " train_y = train[[\"quality\"]]\n", " test_y = test[[\"quality\"]]\n", "\n", " \n", " alpha = float(sys.argv[1]) if len(sys.argv) > 1 else 0.5\n", " #alpha = 0.5\n", " l1_ratio = float(sys.argv[2]) if len(sys.argv) > 2 else 0.5\n", " #l1_ratio = 0.5\n", "\n", " with mlflow.start_run() as run:\n", " print(\"MLflow run experiment_id: {0}\".format(run.info.experiment_id))\n", " print(\"MLflow run artifact_uri: {0}\".format(run.info.artifact_uri))\n", "\n", " lr = ElasticNet(alpha=alpha, l1_ratio=l1_ratio, random_state=42)\n", " lr.fit(train_x, train_y)\n", "\n", " predicted_qualities = lr.predict(test_x)\n", "\n", " (rmse, mae, r2) = eval_metrics(test_y, predicted_qualities)\n", "\n", " print(\"Elasticnet model (alpha=%f, l1_ratio=%f):\" % (alpha, l1_ratio))\n", " print(\" RMSE: %s\" % rmse)\n", " print(\" MAE: %s\" % mae)\n", " print(\" R2: %s\" % r2)\n", "\n", " mlflow.log_param(\"alpha\", alpha)\n", " mlflow.log_param(\"l1_ratio\", l1_ratio)\n", " mlflow.log_metric(\"rmse\", rmse)\n", " mlflow.log_metric(\"r2\", r2)\n", " mlflow.log_metric(\"mae\", mae)\n", " \n", " # Infer model signature to log it\n", " # Więcej o sygnaturach: https://mlflow.org/docs/latest/models.html?highlight=signature#model-signature\n", " signature = mlflow.models.signature.infer_signature(train_x, lr.predict(train_x))\n", "\n", " tracking_url_type_store = urlparse(mlflow.get_tracking_uri()).scheme\n", "\n", " # Model registry does not work with file store\n", " if tracking_url_type_store != \"file\":\n", "\n", " # Register the model\n", " # There are other ways to use the Model Registry, which depends on the use case,\n", " # please refer to the doc for more information:\n", " # https://mlflow.org/docs/latest/model-registry.html#api-workflow\n", " mlflow.sklearn.log_model(lr, \"wines-model\", registered_model_name=\"ElasticnetWineModel\", signature=signature)\n", " else:\n", " mlflow.sklearn.log_model(lr, \"model\", signature=signature)" ] }, { "cell_type": "code", "execution_count": 81, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "total 4\n", "drwxrwxr-x 3 tomek tomek 4096 maj 19 21:31 1\n", "INFO: 's123456' does not exist. Creating a new experiment\n", "MLflow run experiment_id: 2\n", "MLflow run artifact_uri: /tmp/mlruns/2/c15feb5df335490ba990ddd4dd977c1b/artifacts\n", "Elasticnet model (alpha=0.500000, l1_ratio=0.500000):\n", " RMSE: 0.7931640229276851\n", " MAE: 0.6271946374319586\n", " R2: 0.10862644997792614\n", "Registered model 'ElasticnetWineModel' already exists. Creating a new version of this model...\n", "2021/05/19 22:34:48 INFO mlflow.tracking._model_registry.client: Waiting up to 300 seconds for model version to finish creation. Model name: ElasticnetWineModel, version 2\n", "Created version '2' of model 'ElasticnetWineModel'.\n" ] } ], "source": [ "! ls -l /tmp/mlruns\n", "### Wtyrenujmy model z domyślnymi wartościami parametrów\n", "! cd ./IUM_08/examples/; python sklearn_elasticnet_wine/train.py" ] }, { "cell_type": "code", "execution_count": 4, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Elasticnet model (alpha=0.100000, l1_ratio=0.100000):\n", " RMSE: 0.7128829045893679\n", " MAE: 0.5462202174984664\n", " R2: 0.2799376066653344\n", "Elasticnet model (alpha=0.200000, l1_ratio=0.100000):\n", " RMSE: 0.7268133518615142\n", " MAE: 0.5586842416161892\n", " R2: 0.251521166881557\n", "Elasticnet model (alpha=0.300000, l1_ratio=0.100000):\n", " RMSE: 0.7347397539240514\n", " MAE: 0.5657315547549873\n", " R2: 0.23510678899596094\n", "Elasticnet model (alpha=0.400000, l1_ratio=0.100000):\n", " RMSE: 0.7410782793160982\n", " MAE: 0.5712718681984227\n", " R2: 0.22185255063708875\n", "Elasticnet model (alpha=0.500000, l1_ratio=0.100000):\n", " RMSE: 0.7460550348172179\n", " MAE: 0.576381895873763\n", " R2: 0.21136606570632266\n", "Elasticnet model (alpha=0.600000, l1_ratio=0.100000):\n", " RMSE: 0.7510866447955419\n", " MAE: 0.5815681289333974\n", " R2: 0.20069264568704714\n", "Elasticnet model (alpha=0.700000, l1_ratio=0.100000):\n", " RMSE: 0.7560654760040749\n", " MAE: 0.5868129921328281\n", " R2: 0.19006056603695476\n", "Elasticnet model (alpha=0.800000, l1_ratio=0.100000):\n", " RMSE: 0.7609263702116827\n", " MAE: 0.5919470003487062\n", " R2: 0.17961256649282442\n", "Elasticnet model (alpha=0.900000, l1_ratio=0.100000):\n", " RMSE: 0.7656313758553691\n", " MAE: 0.5969367233859049\n", " R2: 0.16943586313742276\n", "Elasticnet model (alpha=0.100000, l1_ratio=0.200000):\n", " RMSE: 0.7201489594275661\n", " MAE: 0.5525324524014098\n", " R2: 0.26518433811823017\n", "Elasticnet model (alpha=0.200000, l1_ratio=0.200000):\n", " RMSE: 0.7336400911821402\n", " MAE: 0.5643841279275428\n", " R2: 0.23739466063584158\n", "Elasticnet model (alpha=0.300000, l1_ratio=0.200000):\n", " RMSE: 0.7397486012946922\n", " MAE: 0.5704931175017443\n", " R2: 0.22464242411894242\n", "Elasticnet model (alpha=0.400000, l1_ratio=0.200000):\n", " RMSE: 0.7468093030485085\n", " MAE: 0.5777243300021722\n", " R2: 0.2097706278632726\n", "Elasticnet model (alpha=0.500000, l1_ratio=0.200000):\n", " RMSE: 0.7543919979968401\n", " MAE: 0.5857669727382302\n", " R2: 0.19364204365178095\n", "Elasticnet model (alpha=0.600000, l1_ratio=0.200000):\n", " RMSE: 0.7622123676513404\n", " MAE: 0.5938629318868578\n", " R2: 0.17683724501340814\n", "Elasticnet model (alpha=0.700000, l1_ratio=0.200000):\n", " RMSE: 0.7700845840888665\n", " MAE: 0.6024685725504659\n", " R2: 0.15974600028150265\n", "Elasticnet model (alpha=0.800000, l1_ratio=0.200000):\n", " RMSE: 0.7778880968569085\n", " MAE: 0.6105907461474273\n", " R2: 0.14263059582492588\n", "Elasticnet model (alpha=0.900000, l1_ratio=0.200000):\n", " RMSE: 0.7855450337039626\n", " MAE: 0.6182359127922239\n", " R2: 0.1256689455181047\n", "Elasticnet model (alpha=0.100000, l1_ratio=0.300000):\n", " RMSE: 0.7260299544064643\n", " MAE: 0.5571534327625295\n", " R2: 0.2531337966130104\n", "Elasticnet model (alpha=0.200000, l1_ratio=0.300000):\n", " RMSE: 0.7357092639331829\n", " MAE: 0.5667609266233857\n", " R2: 0.23308686049079996\n", "Elasticnet model (alpha=0.300000, l1_ratio=0.300000):\n", " RMSE: 0.7443224557281489\n", " MAE: 0.5754825491733004\n", " R2: 0.2150247343683439\n", "Elasticnet model (alpha=0.400000, l1_ratio=0.300000):\n", " RMSE: 0.7545302211047864\n", " MAE: 0.5862255018460154\n", " R2: 0.19334652749043568\n", "Elasticnet model (alpha=0.500000, l1_ratio=0.300000):\n", " RMSE: 0.7657094552843393\n", " MAE: 0.597876674089536\n", " R2: 0.16926645189778677\n", "Elasticnet model (alpha=0.600000, l1_ratio=0.300000):\n", " RMSE: 0.7774287676055035\n", " MAE: 0.6102458961382884\n", " R2: 0.14364282001967787\n", "Elasticnet model (alpha=0.700000, l1_ratio=0.300000):\n", " RMSE: 0.7876149030178985\n", " MAE: 0.6208628759605734\n", " R2: 0.12105524358911324\n", "Elasticnet model (alpha=0.800000, l1_ratio=0.300000):\n", " RMSE: 0.7972426725990548\n", " MAE: 0.6310633254738363\n", " R2: 0.09943554388738107\n", "Elasticnet model (alpha=0.900000, l1_ratio=0.300000):\n", " RMSE: 0.806653553139972\n", " MAE: 0.6407940021176486\n", " R2: 0.07804901733081859\n", "Elasticnet model (alpha=0.100000, l1_ratio=0.400000):\n", " RMSE: 0.7301757756825391\n", " MAE: 0.5603782497631705\n", " R2: 0.24457984004307665\n", "Elasticnet model (alpha=0.200000, l1_ratio=0.400000):\n", " RMSE: 0.7383379454127179\n", " MAE: 0.5696920200435643\n", " R2: 0.22759672468382497\n", "Elasticnet model (alpha=0.300000, l1_ratio=0.400000):\n", " RMSE: 0.7501603725852\n", " MAE: 0.5818749078280213\n", " R2: 0.2026629101382652\n", "Elasticnet model (alpha=0.400000, l1_ratio=0.400000):\n", " RMSE: 0.7644619587468349\n", " MAE: 0.5966303605775048\n", " R2: 0.17197111491474282\n", "Elasticnet model (alpha=0.500000, l1_ratio=0.400000):\n", " RMSE: 0.7794144864140182\n", " MAE: 0.6125287339702588\n", " R2: 0.1392625955410326\n", "Elasticnet model (alpha=0.600000, l1_ratio=0.400000):\n", " RMSE: 0.7928446872861473\n", " MAE: 0.626666444473971\n", " R2: 0.10934405701835759\n", "Elasticnet model (alpha=0.700000, l1_ratio=0.400000):\n", " RMSE: 0.8064523157995205\n", " MAE: 0.6407990295001776\n", " R2: 0.07850896155515663\n", "Elasticnet model (alpha=0.800000, l1_ratio=0.400000):\n", " RMSE: 0.8200264141399415\n", " MAE: 0.6539313398770489\n", " R2: 0.04722706260889009\n", "Elasticnet model (alpha=0.900000, l1_ratio=0.400000):\n", " RMSE: 0.8317936823364004\n", " MAE: 0.6647839366878934\n", " R2: 0.01968654319755092\n", "Elasticnet model (alpha=0.100000, l1_ratio=0.500000):\n", " RMSE: 0.7308996187375898\n", " MAE: 0.5615486628017713\n", " R2: 0.2430813606733676\n", "Elasticnet model (alpha=0.200000, l1_ratio=0.500000):\n", " RMSE: 0.7415652207304311\n", " MAE: 0.573067857646195\n", " R2: 0.22082961765864062\n", "Elasticnet model (alpha=0.300000, l1_ratio=0.500000):\n", " RMSE: 0.7573787958793151\n", " MAE: 0.5893143148791096\n", " R2: 0.18724431943947983\n", "Elasticnet model (alpha=0.400000, l1_ratio=0.500000):\n", " RMSE: 0.7759342885655987\n", " MAE: 0.6090076377075831\n", " R2: 0.14693206734185604\n", "Elasticnet model (alpha=0.500000, l1_ratio=0.500000):\n", " RMSE: 0.7931640229276851\n", " MAE: 0.6271946374319586\n", " R2: 0.10862644997792614\n", "Elasticnet model (alpha=0.600000, l1_ratio=0.500000):\n", " RMSE: 0.8112953030727291\n", " MAE: 0.645693705089251\n", " R2: 0.06740807086129252\n", "Elasticnet model (alpha=0.700000, l1_ratio=0.500000):\n", " RMSE: 0.8298921852578498\n", " MAE: 0.6629780128961713\n", " R2: 0.024163452726365775\n", "Elasticnet model (alpha=0.800000, l1_ratio=0.500000):\n", " RMSE: 0.8320198635059106\n", " MAE: 0.6657357030427604\n", " R2: 0.019153337439844154\n", "Elasticnet model (alpha=0.900000, l1_ratio=0.500000):\n", " RMSE: 0.8323808561832262\n", " MAE: 0.6669472047761406\n", " R2: 0.0183020229672054\n", "Elasticnet model (alpha=0.100000, l1_ratio=0.600000):\n", " RMSE: 0.7317723392279818\n", " MAE: 0.5627373693033669\n", " R2: 0.24127270524006605\n", "Elasticnet model (alpha=0.200000, l1_ratio=0.600000):\n", " RMSE: 0.7454324777911233\n", " MAE: 0.5772117261484206\n", " R2: 0.21268169183406394\n", "Elasticnet model (alpha=0.300000, l1_ratio=0.600000):\n", " RMSE: 0.7661028672396263\n", " MAE: 0.5984406933733759\n", " R2: 0.16841259155853305\n", "Elasticnet model (alpha=0.400000, l1_ratio=0.600000):\n", " RMSE: 0.787179486885359\n", " MAE: 0.6210967388389844\n", " R2: 0.12202678676193257\n", "Elasticnet model (alpha=0.500000, l1_ratio=0.600000):\n", " RMSE: 0.809739471626647\n", " MAE: 0.6442565454817458\n", " R2: 0.07098152823463388\n", "Elasticnet model (alpha=0.600000, l1_ratio=0.600000):\n", " RMSE: 0.8317884179944764\n", " MAE: 0.6647524814105722\n", " R2: 0.019698951776764728\n", "Elasticnet model (alpha=0.700000, l1_ratio=0.600000):\n", " RMSE: 0.8321519738036909\n", " MAE: 0.6662086037874676\n", " R2: 0.018841829895677176\n", "Elasticnet model (alpha=0.800000, l1_ratio=0.600000):\n", " RMSE: 0.8326350511178233\n", " MAE: 0.6676630843299566\n", " R2: 0.01770234373563795\n", "Elasticnet model (alpha=0.900000, l1_ratio=0.600000):\n", " RMSE: 0.8332048101440411\n", " MAE: 0.6690717294644856\n", " R2: 0.016357542209390563\n", "Elasticnet model (alpha=0.100000, l1_ratio=0.700000):\n", " RMSE: 0.7327938109945942\n", " MAE: 0.5640101718105491\n", " R2: 0.23915303116151632\n", "Elasticnet model (alpha=0.200000, l1_ratio=0.700000):\n", " RMSE: 0.7499835110445395\n", " MAE: 0.5819389930665501\n", " R2: 0.20303883413454027\n", "Elasticnet model (alpha=0.300000, l1_ratio=0.700000):\n", " RMSE: 0.7747136483567111\n", " MAE: 0.6079678532556209\n", " R2: 0.14961391810397695\n", "Elasticnet model (alpha=0.400000, l1_ratio=0.700000):\n", " RMSE: 0.8004478857657858\n", " MAE: 0.6350378679245181\n", " R2: 0.09217977708630032\n", "Elasticnet model (alpha=0.500000, l1_ratio=0.700000):\n", " RMSE: 0.829586285479097\n", " MAE: 0.6627028304266674\n", " R2: 0.024882710417618137\n", "Elasticnet model (alpha=0.600000, l1_ratio=0.700000):\n", " RMSE: 0.8321502650365332\n", " MAE: 0.6662000872414003\n", " R2: 0.018845859373919027\n", "Elasticnet model (alpha=0.700000, l1_ratio=0.700000):\n", " RMSE: 0.832725785743381\n", " MAE: 0.667898097502809\n", " R2: 0.017488244494447747\n", "Elasticnet model (alpha=0.800000, l1_ratio=0.700000):\n", " RMSE: 0.8331825395236181\n", " MAE: 0.6692175076829847\n", " R2: 0.016410124803194592\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Elasticnet model (alpha=0.900000, l1_ratio=0.700000):\n", " RMSE: 0.8331069437643933\n", " MAE: 0.6697424890266508\n", " R2: 0.016588601539516357\n", "Elasticnet model (alpha=0.100000, l1_ratio=0.800000):\n", " RMSE: 0.7339712501091269\n", " MAE: 0.5654097809725043\n", " R2: 0.23670603806205326\n", "Elasticnet model (alpha=0.200000, l1_ratio=0.800000):\n", " RMSE: 0.7552646505492441\n", " MAE: 0.5873472009739388\n", " R2: 0.19177543499093674\n", "Elasticnet model (alpha=0.300000, l1_ratio=0.800000):\n", " RMSE: 0.7836957692333741\n", " MAE: 0.6176788505535867\n", " R2: 0.12978065429593022\n", "Elasticnet model (alpha=0.400000, l1_ratio=0.800000):\n", " RMSE: 0.8160164529135189\n", " MAE: 0.650349905850893\n", " R2: 0.05652247327326554\n", "Elasticnet model (alpha=0.500000, l1_ratio=0.800000):\n", " RMSE: 0.8320145539945119\n", " MAE: 0.6657081587004348\n", " R2: 0.019165855890777572\n", "Elasticnet model (alpha=0.600000, l1_ratio=0.800000):\n", " RMSE: 0.8326325509502465\n", " MAE: 0.6676500690618903\n", " R2: 0.01770824285088779\n", "Elasticnet model (alpha=0.700000, l1_ratio=0.800000):\n", " RMSE: 0.8331830329685253\n", " MAE: 0.6692142378162035\n", " R2: 0.016408959758236752\n", "Elasticnet model (alpha=0.800000, l1_ratio=0.800000):\n", " RMSE: 0.8330972295348316\n", " MAE: 0.669813814205792\n", " R2: 0.016611535037920344\n", "Elasticnet model (alpha=0.900000, l1_ratio=0.800000):\n", " RMSE: 0.8330208354420413\n", " MAE: 0.6704133670619602\n", " R2: 0.016791878033996177\n", "Elasticnet model (alpha=0.100000, l1_ratio=0.900000):\n", " RMSE: 0.735314956888905\n", " MAE: 0.566974647785579\n", " R2: 0.23390870203034675\n", "Elasticnet model (alpha=0.200000, l1_ratio=0.900000):\n", " RMSE: 0.7613249071370938\n", " MAE: 0.593613372674502\n", " R2: 0.1787529818606436\n", "Elasticnet model (alpha=0.300000, l1_ratio=0.900000):\n", " RMSE: 0.7940027723712206\n", " MAE: 0.6284316436541582\n", " R2: 0.10674024649047587\n", "Elasticnet model (alpha=0.400000, l1_ratio=0.900000):\n", " RMSE: 0.831784893250733\n", " MAE: 0.6647313794016759\n", " R2: 0.019707259905588637\n", "Elasticnet model (alpha=0.500000, l1_ratio=0.900000):\n", " RMSE: 0.8323747376136406\n", " MAE: 0.6669171677143245\n", " R2: 0.018316455219614114\n", "Elasticnet model (alpha=0.600000, l1_ratio=0.900000):\n", " RMSE: 0.8332063354920289\n", " MAE: 0.6690618761753936\n", " R2: 0.01635394069773599\n", "Elasticnet model (alpha=0.700000, l1_ratio=0.900000):\n", " RMSE: 0.8331078270287657\n", " MAE: 0.6697360518827573\n", " R2: 0.016586516302516174\n", "Elasticnet model (alpha=0.800000, l1_ratio=0.900000):\n", " RMSE: 0.8330212125502486\n", " MAE: 0.6704102143580977\n", " R2: 0.016790987837928095\n", "Elasticnet model (alpha=0.900000, l1_ratio=0.900000):\n", " RMSE: 0.8329464950658837\n", " MAE: 0.6710843636018047\n", " R2: 0.01696735695860563\n" ] } ], "source": [ "### I jeszcze raz, tym razem ze zmienionymi wartościami parametrów\n", "! cd ./IUM_08/examples/; for l in {1..9}; do for a in {1..9}; do python sklearn_elasticnet_wine/train.py 0.$a 0.$l; done; done" ] }, { "cell_type": "code", "execution_count": 32, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "total 16\r\n", "drwxrwxr-x 6 tomek tomek 4096 maj 17 08:43 375cde31bdd44a45a91fd7cee92ebcda\r\n", "drwxrwxr-x 6 tomek tomek 4096 maj 17 10:38 b395b55b47fc43de876b67f5a4a5dae9\r\n", "drwxrwxr-x 6 tomek tomek 4096 maj 17 09:15 b3ead42eca964113b29e7e5f8bcb7bb7\r\n", "-rw-rw-r-- 1 tomek tomek 151 maj 17 08:43 meta.yaml\r\n" ] } ], "source": [ "### Informacje o przebieagach eksperymentu zostały zapisane w katalogu mlruns\n", "! ls -l IUM_08/examples/mlruns/0 | head" ] }, { "cell_type": "code", "execution_count": 33, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "total 20\r\n", "drwxrwxr-x 3 tomek tomek 4096 maj 17 08:43 artifacts\r\n", "-rw-rw-r-- 1 tomek tomek 423 maj 17 08:43 meta.yaml\r\n", "drwxrwxr-x 2 tomek tomek 4096 maj 17 08:43 metrics\r\n", "drwxrwxr-x 2 tomek tomek 4096 maj 17 08:43 params\r\n", "drwxrwxr-x 2 tomek tomek 4096 maj 17 08:43 tags\r\n" ] } ], "source": [ "! ls -l IUM_08/examples/mlruns/0/375cde31bdd44a45a91fd7cee92ebcda" ] }, { "cell_type": "code", "execution_count": 6, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "[2021-05-16 17:58:43 +0200] [118029] [INFO] Starting gunicorn 20.1.0\n", "[2021-05-16 17:58:43 +0200] [118029] [ERROR] Connection in use: ('127.0.0.1', 5000)\n", "[2021-05-16 17:58:43 +0200] [118029] [ERROR] Retrying in 1 second.\n", "[2021-05-16 17:58:44 +0200] [118029] [ERROR] Connection in use: ('127.0.0.1', 5000)\n", "[2021-05-16 17:58:44 +0200] [118029] [ERROR] Retrying in 1 second.\n", "[2021-05-16 17:58:45 +0200] [118029] [ERROR] Connection in use: ('127.0.0.1', 5000)\n", "[2021-05-16 17:58:45 +0200] [118029] [ERROR] Retrying in 1 second.\n", "[2021-05-16 17:58:46 +0200] [118029] [ERROR] Connection in use: ('127.0.0.1', 5000)\n", "[2021-05-16 17:58:46 +0200] [118029] [ERROR] Retrying in 1 second.\n", "[2021-05-16 17:58:47 +0200] [118029] [ERROR] Connection in use: ('127.0.0.1', 5000)\n", "[2021-05-16 17:58:47 +0200] [118029] [ERROR] Retrying in 1 second.\n", "[2021-05-16 17:58:48 +0200] [118029] [ERROR] Can't connect to ('127.0.0.1', 5000)\n", "Running the mlflow server failed. Please see the logs above for details.\n" ] } ], "source": [ "### Możemy je obejrzeć w przeglądarce uruchamiając interfejs webowy:\n", "### (powinniśmy to wywołać w normalnej konsoli, w jupyter będziemy mieli zablokowany kernel)\n", "! cd IUM_08/examples/; mlflow ui" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Instancja na naszym serwerze: http://tzietkiewicz.vm.wmi.amu.edu.pl:5000/#/" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "### Wygląd interfejsu webowego\n", "" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "### Porównywanie wyników\n", "" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "## Logowanie\n", " - logowania metryk i parametrów można dokonać m.in. poprzez wywołania Python-owego API: `mlflow.log_param()` i `mlflow.log_metric()`. Więcej dostępnych funkcji: [link](https://mlflow.org/docs/latest/tracking.html#logging-functions)\n", " - wywołania te muszą nastąpić po wykonaniu [`mlflow.start_run()`](https://mlflow.org/docs/latest/python_api/mlflow.html#mlflow.start_run), najlepiej wewnątrz bloku:\n", "```python\n", " with mlflow.start_run():\n", " \n", " #[...]\n", "\n", " mlflow.log_param(\"alpha\", alpha)\n", " mlflow.log_param(\"l1_ratio\", l1_ratio)\n", "```\n", " - jest też możliwość automatycznego logwania dla wybranych bibliotek: https://mlflow.org/docs/latest/tracking.html#automatic-logging" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "# MLflow Projects\n", " - MLflow projects to zestaw konwencji i kilku narzędzi\n", " - ułatwiają one uruchamianie eskperymentów" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "### Konfiguracja projektu\n", " - W pliku `MLproject` zapisuje się konfigurację projektu ([specyfikacja](https://mlflow.org/docs/latest/projects.html))\n", " - Zawiera ona:\n", " - odnośnik do środowiska, w którym ma być wywołany eksperyment [szczegóły](https://mlflow.org/docs/latest/projects.html#specifying-an-environment):\n", " - nazwa obrazu Docker\n", " - albo ścieżka do pliku conda.yaml definiującego środowisko wykonania Conda\n", " - parametry, z którymi można wywołać eksperyment\n", " - polecenia służące do wywołania eksperymentu" ] }, { "cell_type": "code", "execution_count": 9, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Overwriting IUM_08/examples/sklearn_elasticnet_wine/MLproject\n" ] } ], "source": [ "%%writefile IUM_08/examples/sklearn_elasticnet_wine/MLproject\n", "name: tutorial\n", "\n", "conda_env: conda.yaml #ścieżka do pliku conda.yaml z definicją środowiska\n", " \n", "#docker_env:\n", "# image: mlflow-docker-example-environment\n", "\n", "entry_points:\n", " main:\n", " parameters:\n", " alpha: {type: float, default: 0.5}\n", " l1_ratio: {type: float, default: 0.1}\n", " command: \"python train.py {alpha} {l1_ratio}\"\n", " test:\n", " parameters:\n", " alpha: {type: cutoff, default: 0}\n", " command: \"python test.py {cutoff}\"" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "### Środowisko Conda\n", "
\n", " - https://docs.conda.io\n", " - Składnia plików conda.yaml definiujących środowisko: https://docs.conda.io/projects/conda/en/4.6.1/user-guide/tasks/manage-environments.html#create-env-file-manually\n", " - Składnia YAML: [przystępnie](https://learnxinyminutes.com/docs/yaml/), [oficjalnie](https://yaml.org/spec/1.2/spec.html)" ] }, { "cell_type": "code", "execution_count": 10, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Overwriting IUM_08/examples/sklearn_elasticnet_wine/conda.yaml\n" ] } ], "source": [ "%%writefile IUM_08/examples/sklearn_elasticnet_wine/conda.yaml\n", "name: tutorial\n", "channels:\n", " - defaults\n", "dependencies:\n", " - python=3.6 #Te zależności będą zainstalowane za pomocą conda isntall\n", " - pip\n", " - pip: #Te ząś za pomocą pip install\n", " - scikit-learn==0.23.2\n", " - mlflow>=1.0" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "### Środowisko docker\n", "- zamiast środowiska Conda możemy również podać nazwę obrazu docker, w którym ma być wywołany eksperyment.\n", "- obraz będzie szukany lokalnie a następnie na DockerHub, lub w innym repozytorium dockera\n", "- składnia specyfikacji ścieżki jest taka sama jak w przypadki poleceń dockera, np. docker pull [link](https://docs.docker.com/engine/reference/commandline/pull/#pull-from-a-different-registry)\n", "- Można również podać katalogi do podmontowania wewnątrz kontenera oraz wartości zmiennych środowiskowych do ustawienia w kontenerze:\n", "```yaml\n", "docker_env:\n", " image: mlflow-docker-example-environment\n", " volumes: [\"/local/path:/container/mount/path\"]\n", " environment: [[\"NEW_ENV_VAR\", \"new_var_value\"], \"VAR_TO_COPY_FROM_HOST_ENVIRONMENT\"]\n", "```" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "### Parametry\n", " - Specyfikacja parametrów w pliku MLproject pozwala na ich walidację i używanie wartości domyślnych\n", " - Dostępne typy:\n", " - String\n", " - Float - dowolna liczba (MLflow waliduje, czy podana wartość jest liczbą)\n", " - Path - pozwala podawać ścieżki względne (przekształca je na bezwzlędne) do plików lokalnych albo do plików zdalnych (np. do s3://) - zostaną wtedy ściągnięte lokalnie\n", " - URI - podobnie jak path, ale do rozproszonych systemów plików\n", "\n", "- [Składnia](https://mlflow.org/docs/latest/projects.html#specifying-parameters)\n", " \n", "```yml:\n", " parameter_name: {type: data_type, default: value} # Short syntax\n", "\n", " parameter_name: # Long syntax\n", " type: data_type\n", " default: value\n", "```" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "### Uruchamianie projektu\n", " - Projekt możemy uruchomić przy pomocy polecenia `mlflow run` ([dokumentacja](https://mlflow.org/docs/latest/cli.html#mlflow-run))\n", " - Spowoduje to przygotowanie środowiska i uruchomienie eksperymentu wewnątrz środowiska\n", " - domyślnie zostanie uruchomione polecenie zdefiniowane w \"entry point\" `main`. Żeby uruchomić inny \"entry point\", możemy użyć parametru `-e`, np:\n", " ```bash\n", " mlflow run sklearn_elasticnet_wine -e test\n", " ```\n", " - Parametry do naszego polecenia możemy przekazywać przy pomocy flagi `-P`" ] }, { "cell_type": "code", "execution_count": 11, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "2021/05/16 17:59:10 INFO mlflow.projects.utils: === Created directory /tmp/tmprq4mdosv for downloading remote URIs passed to arguments of type 'path' ===\n", "2021/05/16 17:59:10 INFO mlflow.projects.backend.local: === Running command 'source /home/tomek/miniconda3/bin/../etc/profile.d/conda.sh && conda activate mlflow-5987e03d4dbaa5faa1a697bb113be9b9bdc39b29 1>&2 && python train.py 0.42 0.1' in run with ID '1860d321ea1545ff8866e4ba199d1712' === \n", "Elasticnet model (alpha=0.420000, l1_ratio=0.100000):\n", " RMSE: 0.7420620899060748\n", " MAE: 0.5722846717246247\n", " R2: 0.21978513651550236\n", "2021/05/16 17:59:19 INFO mlflow.projects: === Run (ID '1860d321ea1545ff8866e4ba199d1712') succeeded ===\n" ] } ], "source": [ "!cd IUM_08/examples/; mlflow run sklearn_elasticnet_wine -P alpha=0.42" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "# Zadania [10p pkt]\n", "1. Dodaj do swojego projektu logowanie parametrów i metryk za pomocą MLflow (polecenia `mlflow.log_param` i `mlflow.log_metric`\n", "2. Dodaj plik MLProject definiujący polecenia do trenowania i testowania, ich parametry wywołania oraz środowisko (Conda albo Docker)" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "## MLflow Models\n", "\n", "MLflow Models to konwencja zapisu modeli, która ułatwia potem ich załadowanie i użycie" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "Rodzaje modeli (\"flavors\") wspierane przez MLflow:\n", "\n", " - Python Function (python_function)\n", " - PyTorch (pytorch)\n", " - TensorFlow (tensorflow)\n", " - Keras (keras)\n", " - Scikit-learn (sklearn)\n", " - Spacy(spaCy)\n", " - ONNX (onnx)\n", " - R Function (crate)\n", " - H2O (h2o)\n", " - MLeap (mleap)\n", " - Spark MLlib (spark)\n", " - MXNet Gluon (gluon)\n", " - XGBoost (xgboost)\n", " - LightGBM (lightgbm)\n", " - CatBoost (catboost)\n", " - Fastai(fastai)\n", " - Statsmodels (statsmodels)" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "### Zapisywanie modelu\n", "Model ML można zapisać w MLflow przy pomocy jednej z dwóch funkcji z pakietu odpowiadającego używanej przez nas bibliotece:\n", " - `save_model()` - zapisuje model na dysku\n", " - `log_model()` - zapisuje model razem z innymi informacjami (metrykami, parametrami). W zależności od ustawień [\"tracking_uri\"](https://mlflow.org/docs/latest/python_api/mlflow.html#mlflow.set_tracking_uri) może być to lokalny folder w `mlruns/ ` lub ścieżka na zdalnym serwerze MLflow\n", "\n", "```Python\n", " mlflow.sklearn.save_model(lr, \"my_model\")\n", "```\n", "\n", "```Python\n", " mlflow.keras.save_model(lr, \"my_model\")\n", "```\n", "\n", "Wywołanie tej funkcji spowoduje stworzenie katalogu \"my_model\" zawierającego:\n", " - plik *MLmodel* zawierający informacje o sposobach, w jaki model można załadować (\"flavors\") oraz ścieżki do plików związanych z modelem, takich jak:\n", " - *conda.yaml* - opis środowiska potrzebnego do załadowania modelu\n", " - *model.pkl* - plik z zserializowanym modelem\n", "\n", "Tylko plik *MLmodel* jest specjalnym plikiem MLflow - reszta zależy od konkrentego \"flavour\"\n" ] }, { "cell_type": "code", "execution_count": 16, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "conda.yaml MLmodel model.pkl\r\n" ] } ], "source": [ "ls IUM_08/examples/my_model" ] }, { "cell_type": "code", "execution_count": 39, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "total 12\r\n", "-rw-rw-r-- 1 tomek tomek 153 maj 17 10:38 conda.yaml\r\n", "-rw-rw-r-- 1 tomek tomek 958 maj 17 10:38 MLmodel\r\n", "-rw-rw-r-- 1 tomek tomek 641 maj 17 10:38 model.pkl\r\n" ] } ], "source": [ "! ls -l IUM_08/examples/mlruns/0/b395b55b47fc43de876b67f5a4a5dae9/artifacts/model" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [], "source": [ "# %load IUM_08/examples/mlruns/0/b395b55b47fc43de876b67f5a4a5dae9/artifacts/model/MLmodel\n", "artifact_path: model\n", "flavors:\n", " python_function:\n", " env: conda.yaml\n", " loader_module: mlflow.sklearn\n", " model_path: model.pkl\n", " python_version: 3.9.1\n", " sklearn:\n", " pickled_model: model.pkl\n", " serialization_format: cloudpickle\n", " sklearn_version: 0.24.2\n", "run_id: b395b55b47fc43de876b67f5a4a5dae9\n", "signature:\n", " inputs: '[{\"name\": \"fixed acidity\", \"type\": \"double\"}, {\"name\": \"volatile acidity\",\n", " \"type\": \"double\"}, {\"name\": \"citric acid\", \"type\": \"double\"}, {\"name\": \"residual\n", " sugar\", \"type\": \"double\"}, {\"name\": \"chlorides\", \"type\": \"double\"}, {\"name\": \"free\n", " sulfur dioxide\", \"type\": \"double\"}, {\"name\": \"total sulfur dioxide\", \"type\": \"double\"},\n", " {\"name\": \"density\", \"type\": \"double\"}, {\"name\": \"pH\", \"type\": \"double\"}, {\"name\":\n", " \"sulphates\", \"type\": \"double\"}, {\"name\": \"alcohol\", \"type\": \"double\"}]'\n", " outputs: '[{\"type\": \"tensor\", \"tensor-spec\": {\"dtype\": \"float64\", \"shape\": [-1]}}]'\n", "utc_time_created: '2021-05-17 08:38:41.749670'\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [], "source": [ "# %load IUM_08/examples/my_model/conda.yaml\n", "channels:\n", "- defaults\n", "- conda-forge\n", "dependencies:\n", "- python=3.9.1\n", "- pip\n", "- pip:\n", " - mlflow\n", " - scikit-learn==0.24.2\n", " - cloudpickle==1.6.0\n", "name: mlflow-env" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "### Dodatkowe pola w MLmodel\n", "\n", "\n", "- *utc_time_created* - timestamp z czasem stworzenia modelu\n", "- *run_id* - ID uruchomienia (\"run\"), które stworzyło ten model, jeśli model był zapisany za pomocą MLflow Tracking.\n", "- *signature* - opisa danych wejściowych i wyjściowych w formacie JSON\n", "- *input_example* przykładowe wejście przyjmowane przez model. Można je podać poprzez parametr `input_example` funkcji [log_model](https://mlflow.org/docs/latest/python_api/mlflow.sklearn.html#mlflow.sklearn.log_model)\n", "\n" ] }, { "cell_type": "code", "execution_count": 50, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [ { "data": { "text/plain": [ "array([5.57688397, 5.50664777, 5.52550482, 5.50431125, 5.57688397])" ] }, "execution_count": 50, "metadata": {}, "output_type": "execute_result" } ], "source": [ "import mlflow\n", "import pandas as pd\n", "model = mlflow.sklearn.load_model(\"IUM_08/examples/mlruns/0/b395b55b47fc43de876b67f5a4a5dae9/artifacts/model\")\n", "csv_url = \"http://archive.ics.uci.edu/ml/machine-learning-databases/wine-quality/winequality-red.csv\"\n", "data = pd.read_csv(csv_url, sep=\";\")\n", "model.predict(data.drop([\"quality\"], axis=1).head())" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "### Serwowanie modeli" ] }, { "cell_type": "code", "execution_count": 7, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Usage: mlflow models [OPTIONS] COMMAND [ARGS]...\r\n", "\r\n", " Deploy MLflow models locally.\r\n", "\r\n", " To deploy a model associated with a run on a tracking server, set the\r\n", " MLFLOW_TRACKING_URI environment variable to the URL of the desired server.\r\n", "\r\n", "Options:\r\n", " --help Show this message and exit.\r\n", "\r\n", "Commands:\r\n", " build-docker **EXPERIMENTAL**: Builds a Docker image whose default...\r\n", " predict Generate predictions in json format using a saved MLflow...\r\n", " prepare-env **EXPERIMENTAL**: Performs any preparation necessary to...\r\n", " serve Serve a model saved with MLflow by launching a webserver on...\r\n" ] } ], "source": [ "!cd IUM_08/examples/; mlflow models --help" ] }, { "cell_type": "code", "execution_count": 8, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Usage: mlflow models serve [OPTIONS]\r\n", "\r\n", " Serve a model saved with MLflow by launching a webserver on the specified\r\n", " host and port. The command supports models with the ``python_function`` or\r\n", " ``crate`` (R Function) flavor. For information about the input data\r\n", " formats accepted by the webserver, see the following documentation:\r\n", " https://www.mlflow.org/docs/latest/models.html#built-in-deployment-tools.\r\n", "\r\n", " You can make requests to ``POST /invocations`` in pandas split- or record-\r\n", " oriented formats.\r\n", "\r\n", " Example:\r\n", "\r\n", " .. code-block:: bash\r\n", "\r\n", " $ mlflow models serve -m runs:/my-run-id/model-path &\r\n", "\r\n", " $ curl http://127.0.0.1:5000/invocations -H 'Content-Type:\r\n", " application/json' -d '{ \"columns\": [\"a\", \"b\", \"c\"],\r\n", " \"data\": [[1, 2, 3], [4, 5, 6]] }'\r\n", "\r\n", "Options:\r\n", " -m, --model-uri URI URI to the model. A local path, a 'runs:/' URI, or a\r\n", " remote storage URI (e.g., an 's3://' URI). For more\r\n", " information about supported remote URIs for model\r\n", " artifacts, see\r\n", " https://mlflow.org/docs/latest/tracking.html#artifact-\r\n", " stores [required]\r\n", "\r\n", " -p, --port INTEGER The port to listen on (default: 5000).\r\n", " -h, --host HOST The network address to listen on (default: 127.0.0.1).\r\n", " Use 0.0.0.0 to bind to all addresses if you want to\r\n", " access the tracking server from other machines.\r\n", "\r\n", " -w, --workers TEXT Number of gunicorn worker processes to handle requests\r\n", " (default: 4).\r\n", "\r\n", " --no-conda If specified, will assume that MLmodel/MLproject is\r\n", " running within a Conda environment with the necessary\r\n", " dependencies for the current project instead of\r\n", " attempting to create a new conda environment.\r\n", "\r\n", " --install-mlflow If specified and there is a conda environment to be\r\n", " activated mlflow will be installed into the environment\r\n", " after it has been activated. The version of installed\r\n", " mlflow will be the same asthe one used to invoke this\r\n", " command.\r\n", "\r\n", " --help Show this message and exit.\r\n" ] } ], "source": [ "!cd IUM_08/examples/; mlflow models serve --help" ] }, { "cell_type": "code", "execution_count": 54, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "{\"columns\":[\"fixed acidity\",\"volatile acidity\",\"citric acid\",\"residual sugar\",\"chlorides\",\"free sulfur dioxide\",\"total sulfur dioxide\",\"density\",\"pH\",\"sulphates\",\"alcohol\"],\"index\":[0],\"data\":[[7.4,0.7,0.0,1.9,0.076,11.0,34.0,0.9978,3.51,0.56,9.4]]}\n" ] } ], "source": [ "import pandas as pd\n", "csv_url = \"http://archive.ics.uci.edu/ml/machine-learning-databases/wine-quality/winequality-red.csv\"\n", "data = pd.read_csv(csv_url, sep=\";\").drop([\"quality\"], axis=1).head(1).to_json(orient='split')\n", "print(data)" ] }, { "cell_type": "code", "execution_count": 57, "metadata": { "slideshow": { "slide_type": "slide" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "[5.576883967129615]" ] } ], "source": [ "!curl http://127.0.0.1:5003/invocations -H 'Content-Type: application/json' -d '{\\\n", " \"columns\":[\\\n", " \"fixed acidity\",\"volatile acidity\",\"citric acid\",\"residual sugar\",\"chlorides\",\"free sulfur dioxide\",\"total sulfur dioxide\",\"density\",\"pH\",\"sulphates\",\"alcohol\"],\\\n", " \"index\":[0],\\\n", " \"data\":[[7.4,0.7,0.0,1.9,0.076,11.0,34.0,0.9978,3.51,0.56,9.4]]}'" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "```\n", "$ cd IUM_08/examples/\n", "$ mlflow models serve -m my_model\n", "2021/05/17 08:52:07 INFO mlflow.models.cli: Selected backend for flavor 'python_function'\n", "2021/05/17 08:52:07 INFO mlflow.pyfunc.backend: === Running command 'source /home/tomek/miniconda3/bin/../etc/profile.d/conda.sh && conda activate mlflow-503f0c7520a32f054a9d168bd099584a9439de9d 1>&2 && gunicorn --timeout=60 -b 127.0.0.1:5003 -w 1 ${GUNICORN_CMD_ARGS} -- mlflow.pyfunc.scoring_server.wsgi:app'\n", "[2021-05-17 08:52:07 +0200] [291217] [INFO] Starting gunicorn 20.1.0\n", "[2021-05-17 08:52:07 +0200] [291217] [INFO] Listening at: http://127.0.0.1:5003 (291217)\n", "[2021-05-17 08:52:07 +0200] [291217] [INFO] Using worker: sync\n", "[2021-05-17 08:52:07 +0200] [291221] [INFO] Booting worker with pid: 291221\n", "```" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "## MLflow Registry\n", " - umożliwia [zapisywanie](https://mlflow.org/docs/latest/model-registry.html#adding-an-mlflow-model-to-the-model-registry) i [ładowanie](https://mlflow.org/docs/latest/model-registry.html#fetching-an-mlflow-model-from-the-model-registry) modeli z centralnego rejestru\n", " - Modele można też serwować bezpośrednio z rejestru:\n", "\n", "```bash\n", "#!/usr/bin/env sh\n", "\n", "# Set environment variable for the tracking URL where the Model Registry resides\n", "export MLFLOW_TRACKING_URI=http://localhost:5000\n", "\n", "# Serve the production model from the model registry\n", "mlflow models serve -m \"models:/sk-learn-random-forest-reg-model/Production\"\n", "```\n", "\n", "- Żeby było to możliwe, musimy mieć uruchomiony [serwer MLflow](https://mlflow.org/docs/latest/tracking.html#tracking-server)\n", "- Umożliwia zarządzanie wersjami modeli i oznaczanie ich różnymi fazami, np. \"Staging\", \"Production\"" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "## Zadania (termin: 14.05 EOD) \n", "1. [2 pkt] Dodaj do joba treningowego wywołania MLflow, tak, żeby przy każdym uruchomieniu stworzyć i zarchiwizować katalog z modelem. Plik MLmodel powinien zawierać pola:\n", " - signature\n", " - input_example\n", "\n", " Folder powinien również zawierać definicję środowiska - conda lub docker, umożliwiającego uruchomienie projektu.\n", "\n", "2. [6 pkt] Wybierz jedną osobę z grupy. Załóżmy, że Twoje ID to s123456 a jej s654321. Stwórz na Jenkinsie projekt `s123456-predict-s654321`, w którym:\n", " - pobierzesz artefakt z zapisanym modelem z joba osoby s654321\n", " - dokonasz na nim predykcji danych wejściowych podanych w formacie json jako parametr zadania Jenkinsowego. Domyślną wartością tego parametru niech będą przykładowe dane wejściowe z `input_example`\n", " \n", "3. [1 pkt] Zarejestruj swój model w MLflow registry\n", "\n", "4. [6 pkt] Stwórz na Jenkinsie projekt `s123456-predict-s654321-from-registry`, który zrealizuje to samo zadanie co `s123456-predict-s654321`, ale tym razem pobierze model z rejestru MLflow zamiast z artefaktów Jenkinsa" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "Dane do konfiguracji MLflow registry\n", "\n", "- Podgląd w przeglądarce: http://tzietkiewicz.vm.wmi.amu.edu.pl/#/\n", " - user: `student`\n", " - hasło: IUM@2021\n", "\n", "- Tracking URI:\n", " - Python: `mlflow.set_tracking_uri(\"http://172.17.0.1:5000\")`\n", " - CLI: `export MLFLOW_TRACKING_URI=http://172.17.0.1:5000`\n", " \n", "- Żeby klient MLflow działający w kontenerze docker mógł zapisywać i pdczytywać artefakty, muszą Państwo podmonotwać katalog `/tmp/mlruns` i `/mlruns` (ten drugi po to, żeby po restarcie serwera katalog nie został wyczyszczony)\n", " - jak podmontować: ```docker.image('my-image').inside('-v /tmp/mlruns:/tmp/mlruns' -v /mlruns:/mlruns')```\n", " - Przykład działąjącego joba na Jenkinsie zapisującego wyniki do MLFlow: https://tzietkiewicz.vm.wmi.amu.edu.pl:8081/job/gitea-test/job/ium-helloworld/job/master/\n", " - Repo: https://git.wmi.amu.edu.pl/tzietkiewicz/ium-helloworld/src/branch/master\n", "\n", "- Proszę ustawić nazwę eksperymentu na numer indeksu, dzięki temu każdy z Państwa będzie widział swoje eksperymenty oddzielnie:\n", "`mlflow.set_experiment(\"s123456\")`" ] } ], "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": "8.MLFlow[laboratoria]", "title": "Inżynieria uczenia maszynowego", "year": "2021" }, "nbformat": 4, "nbformat_minor": 4 }