Azure Automated Machine Learningを使ってみた。自動MLの可能性について

ノーコードで予測分析を実施してみました。

ノーコードと言いつつ、作成したモデルをデプロイしてサービス化したあとにサービスを使用するアプリケーションはコーディングが必要とあなります。

モデルのランキングは、RMSEがNO1は、StackEnsemble

自動MLで分析を実施して、約80のアルゴリズムで分析をかけます。

使用したアルゴリズムです。

AutoArima

MaxAbsScaler, DecisionTree

MaxAbsScaler, SGD

MinMaxScaler, DecisionTree

MinMaxScaler, ElasticNet

MinMaxScaler, LightGBM

MinMaxScaler, SGD

ProphetModel

RobustScaler, DecisionTree

RobustScaler, ElasticNet

RobustScaler, LassoLars

RobustScaler, LightGBM

RobustScaler, RandomForest

StackEnsemble

StandardScalerWrapper, DecisionTree

StandardScalerWrapper, ElasticNet

StandardScalerWrapper, ExtremeRandomTrees

StandardScalerWrapper, LassoLars

StandardScalerWrapper, RandomForest

StandardScalerWrapper, SGD

TruncatedSVDWrapper, LassoLars

VotingEnsemble

実験を

type algorithm.txt | sort | uniaue > algorithm_list.txt

で出力。

中身を見てみるとどの特徴がどのくらい影響を与えているかがわかります。

デプロイしてSwaggerでI/Fを確認

ヘルスがあるんですね。

クライントロジックの作成

using System;
using System.Collections.Generic;
using System.Data;
using System.IO;
using System.Net.Http;
using System.Net.Http.Headers;
using Newtonsoft.Json;
using Newtonsoft.Json.Converters;

namespace MLWebServiceClient
{
	internal class InputData
	{

		[JsonProperty("data")]
		internal Request[] Data;
	}
	// The data structure expected by the service
	internal class Request
	{
		[JsonProperty("instant")]
		// The service used by this example expects an array containing
		//   one or more arrays of doubles
		internal int Instant;
		[JsonProperty("date"), JsonConverter(typeof(CustomDateTimeConverter))]
		internal DateTime Date;
		[JsonProperty("season")]
		internal int Season;
		[JsonProperty("yr")]
		internal int Year;
		[JsonProperty("mnth")]
		internal int Month;
		[JsonProperty("weekday")]
		internal int Weekday;
		[JsonProperty("weathersit")]
		internal int Weathersit;
		[JsonProperty("temp")]
		internal double Temp;
		[JsonProperty("atemp")]
		internal double ATemp;
		[JsonProperty("hum")]
		internal double Hum;
		[JsonProperty("windspeed")]
		internal double Windspeed;
	}
	class CustomDateTimeConverter : IsoDateTimeConverter
	{
		public CustomDateTimeConverter()
		{
			base.DateTimeFormat = "yyyy-MM-dd HH:mm:ss,ffffff";
		}
	}

	class Program
	{
		static void Main(string[] args)
		{
			Health();

			Score();
			Console.ReadKey();
		}
		private static void Health()
		{
			// Set the scoring URI and authentication key or token
			string scoringUri = "http://baeab633-xxxx-xxxx-xxxx-6b4317bfbbdf.westus.azurecontainer.io/";
			HttpClient client = new HttpClient();
			try
			{
				var request = new HttpRequestMessage(HttpMethod.Get, new Uri(scoringUri));
				request.Content = new StringContent("");
				request.Content.Headers.ContentType = new MediaTypeHeaderValue("application/json");
				var response = client.GetAsync(new Uri(scoringUri));
				Console.WriteLine(response.Result);
			}
			catch (Exception e)
			{
				Console.Out.WriteLine(e.ToString());
			}
		}
		private static void Score()
		{

			// Set the scoring URI and authentication key or token
			string scoringUri = "http://baeab633-xxxx-xxxx-xxxx-6b4317bfbbdf.westus.azurecontainer.io/score";
			string authKey = "<your key or token>";

			// Set the data to be sent to the service.
			// In this case, we are sending two sets of data to be scored.
			InputData payload = new InputData
			{
				Data = new[] {
			new Request
			{
				Instant = 732,
				Date = new DateTime(2013, 1, 1),
				Season = 1,
				Year = 2,
				Month = 1,
				Weekday = 5,
				Weathersit = 2,
				Temp = 0.215833,
				ATemp = 0.223487,
				Hum = 0.5775,
				Windspeed = 0.154846,
			}}
			};


			// Create the HTTP client
			HttpClient client = new HttpClient();
			// Set the auth header. Only needed if the web service requires authentication.
			//client.DefaultRequestHeaders.Authorization = new AuthenticationHeaderValue("Bearer", authKey);

			// Make the request
			try
			{
				Console.WriteLine(JsonConvert.SerializeObject(payload));
				var content = new StringContent(JsonConvert.SerializeObject(payload));
				content.Headers.ContentType = new MediaTypeHeaderValue("application/json");
				var response = client.PostAsync(new Uri(scoringUri), content).Result;
				// Display the response from the web service
				Console.WriteLine(response.Content.ReadAsStringAsync().Result);
			}
			catch (Exception e)
			{
				Console.Out.WriteLine(e.ToString());
			}


		}
	}
}

処理した結果

うまくコード２００が返ってきて予測値が取得できました。