RecommendationApiParamSearch.py

import requests
import time
import json

class AzureReco:
    BASE_URL = 'https://westus.api.cognitive.microsoft.com/recommendations/v4.0'

    def __init__(self, subscription_key):
        self._subscription_key = subscription_key

    def _common_header(self):
        return {'Ocp-Apim-Subscription-Key': self._subscription_key}

    def get_last_build(self, model_id):
        r = requests.get(
            self.BASE_URL+'/models/{model_id}/builds'.format(**locals()),
            params={'onlyLastRequestedBuild': True},
            headers=self._common_header(),
        )
        builds = r.json()['builds']
        return builds[0] if builds else None

    def delete_build(self, model_id, build_id):
        r = requests.delete(
            self.BASE_URL+'/models/{model_id}/builds/{build_id}'.format(**locals()),
            headers=self._common_header(),
        )

    def start_recommendation_build(self, model_id, recom_model_params):
        headers = self._common_header()
        headers['content-type'] = 'application/json'
        params = {
            "description": "",
            "buildType": "recommendation",
            "buildParameters": {
                "recommendation": recom_model_params,
            },
        }
        r = requests.post(
            self.BASE_URL+'/models/{model_id}/builds'.format(**locals()),
            data=json.dumps(params),
            headers=headers,
        )
        ret = r.json()
        print(ret)
        return ret

    def get_build_metrics(self, model_id, build_id):
        r = requests.get(
            self.BASE_URL+'/models/{model_id}/builds/{build_id}/metrics'.format(**locals()),
            headers=self._common_header(),
        )
        return r.json()

API_KEY = '__SET__YOUR__API__KEY__'
MODEL_ID = '__SET__YOUR__MODEL__ID__'

def main():
    model_params_base = {
        "numberOfModelIterations": 10,
        "numberOfModelDimensions": 10,
        "itemCutOffLowerBound": 2,
        "userCutOffLowerBound": 2,
        "enableModelingInsights": True,
        "enableU2I": True,
        "splitterStrategy": "LastEventSplitter",
    }

    recoapi = AzureReco(API_KEY)

    for num_dim in (10,20,30,40):
        for num_iter in (10,20,30,40):
            print('build (%d,%d)' % (num_dim, num_iter))
            model_params = model_params_base.copy()
            # パラメータをセット
            model_params.update({
                "numberOfModelIterations": num_iter,
                "numberOfModelDimensions": num_dim,
            })
            # ビルド開始
            build_id = recoapi.start_recommendation_build(MODEL_ID, model_params)['buildId']
            print('build_id = %d' % build_id)
            
            # ビルドが完了するまで待つ
            while True:
                status = recoapi.get_last_build(MODEL_ID)['status']
                print('status = %s' % status)
                if status in ('NotStarted', 'Running',):
                    time.sleep(30) # sleepを短くしすぎるとAPI Rate Limitでエラーになるので注意
                    continue
                elif status == 'Succeeded':
                    break
                else:
                    break

            # 評価値を取得
            metrics = recoapi.get_build_metrics(MODEL_ID, build_id)
            with open('metrics/result-%d-%d.json'%(num_dim,num_iter), 'w+') as f:
                f.write(json.dumps(metrics))
            
            # 保持できるBuildは1モデルあたり20個が上限なので、用済みになったBuildはすぐに消す。
            recoapi.delete_build(MODEL_ID, build_id)


if __name__ == '__main__':
    main()

jqコマンドで評価値(k=1..5)の平均を算出できる

cat metrics/result-*.json | jq '[.precisionUserRecommend.precisionMetrics[].percentage]|add/5'
cat metrics/result-*.json | jq '[.precisionItemRecommend.precisionMetrics[].percentage]|add/5'