BigQuery、Google Kubernetes Engine など
整備工場など、自動車アフター マーケットを中心に、携帯電話販売代理店、旅行代理店など、幅広い業種・業界の現場業務を支援するソフトウエア・ITソリューション＆各種サービスを提供（約 3 万社が利用）。近年は中国、フィリピンなど、アジア地域へも進出している。2005 年 12 月創業。開発拠点は東京・札幌・福岡の 3 か所。従業員数は 921 人（2018 年 12 月末時点）。
message RolloutRequest { // Index of rollout int32 rollout_index = 1; // Random seed from the master int32 env_seed = 2; // The weights of the neural network policy Repeated double policy_weights = 3; // Whether this request is evaluation bool evaluate = 4; } message RolloutResponse { // Index of rollout int32 rollout_index = 1; // The reward collected in the rollout double rollout_rewards = 2; } service Rollout { // The RPC service for policy evaluation rpc RolloutWithParameter(RolloutRequest) returns (RolloutResponse) {} }
message RolloutRequest {
// Index of rollout
int32 rollout_index = 1;
// Random seed from the master
int32 env_seed = 2;
// The weights of the neural network policy
Repeated double policy_weights = 3;
// Whether this request is evaluation
bool evaluate = 4;
}
message RolloutResponse {
// The reward collected in the rollout
double rollout_rewards = 2;
service Rollout {
// The RPC service for policy evaluation
rpc RolloutWithParameter(RolloutRequest) returns (RolloutResponse) {}
GCLOUD_PROJECT={your-project-id} # Create a cluster. gcloud container clusters create es-cluster \ --zone=us-central1-a \ --machine-type=n1-standard-64 \ --max-nodes=20 \ --min-nodes=16 \ --num-nodes=17 \ --enable-autoscaling \ --project ${GCLOUD_PROJECT}
GCLOUD_PROJECT={your-project-id}
# Create a cluster.
gcloud container clusters create es-cluster \
--zone=us-central1-a \
--machine-type=n1-standard-64 \
--max-nodes=20 \
--min-nodes=16 \
--num-nodes=17 \
--enable-autoscaling \
--project ${GCLOUD_PROJECT}
# Step 1: Build a container image for the controller and the workers. gcloud builds submit \ --tag gcr.io/${GCLOUD_PROJECT}/es-on-gke:1.0 . \ --timeout 3600 \ --project ${GCLOUD_PROJECT} # Step 2: Replace the ${GCLOUD_PROJECT} in the YAML file and run the command. kubectl apply -f deploy_workers.yaml # Step 3: When all the workers are started and running, # replace the ${GCLOUD_PROJECT} in the YAML file and run the command. kubectl apply -f deploy_master.yaml
# Step 1: Build a container image for the controller and the workers.
gcloud builds submit \
--tag gcr.io/${GCLOUD_PROJECT}/es-on-gke:1.0 . \
--timeout 3600 \
# Step 2: Replace the ${GCLOUD_PROJECT} in the YAML file and run the command.
kubectl apply -f deploy_workers.yaml
# Step 3: When all the workers are started and running,
# replace the ${GCLOUD_PROJECT} in the YAML file and run the command.
kubectl apply -f deploy_master.yaml
POD_NAME=$(kubectl get pod | grep es-master | awk '{print $1}') # Download reward vs time plot. kubectl cp $POD_NAME:/var/log/es/log/reward_vs_time.png $HOME/ # Download reward vs iteration plot. kubectl cp $POD_NAME:/var/log/es/log/reward_vs_iteration.png $HOME/ # Download best model so far. kubectl cp $POD_NAME:/var/log/es/log/best_model.npz $HOME/ # Download model at iteration 1000. kubectl cp $POD_NAME:/var/log/es/log/model_1000.npz $HOME/ # Download all test scores. kubectl cp $POD_NAME:/var/log/es/log/scores.csv $HOME/
POD_NAME=$(kubectl get pod | grep es-master | awk '{print $1}')
# Download reward vs time plot.
kubectl cp $POD_NAME:/var/log/es/log/reward_vs_time.png $HOME/
# Download reward vs iteration plot.
kubectl cp $POD_NAME:/var/log/es/log/reward_vs_iteration.png $HOME/
# Download best model so far.
kubectl cp $POD_NAME:/var/log/es/log/best_model.npz $HOME/
# Download model at iteration 1000.
kubectl cp $POD_NAME:/var/log/es/log/model_1000.npz $HOME/
# Download all test scores.
kubectl cp $POD_NAME:/var/log/es/log/scores.csv $HOME/
# Train locally. # E.g. bash ./train_local.sh -c configs/BipedalWalkerHardcore.gin -n 10 bash ./train_local.sh -c {path-to-config-file} -n {number-of-workers} # Test locally. # E.g. python test.py --checkpoint=1000 --logdir=./log python test.py --checkpoint={checkpoint} --logdir={path-to-log-directory}
# Train locally.
# E.g. bash ./train_local.sh -c configs/BipedalWalkerHardcore.gin -n 10
bash ./train_local.sh -c {path-to-config-file} -n {number-of-workers}
# Test locally.
# E.g. python test.py --checkpoint=1000 --logdir=./log
python test.py --checkpoint={checkpoint} --logdir={path-to-log-directory}
model1 = """ CREATE OR REPLACE MODEL `my-example-housing-dataset.ameshousing.linearmodel` OPTIONS(model_type='linear_reg', ls_init_learn_rate=.15, l1_reg=1, max_iterations=5) AS SELECT IFNULL(SalePrice, 0) AS label, IFNULL(GrLivArea, 0) AS LivingAreaSize, YearBuilt, OverallCond, OverallQual FROM `my-example-housing-dataset.ameshousing.train` """
model1 = """
CREATE OR REPLACE MODEL
`my-example-housing-dataset.ameshousing.linearmodel`
OPTIONS(model_type='linear_reg', ls_init_learn_rate=.15, l1_reg=1, max_iterations=5) AS
SELECT
IFNULL(SalePrice, 0) AS label,
IFNULL(GrLivArea, 0) AS LivingAreaSize,
YearBuilt,
OverallCond,
OverallQual
FROM
`my-example-housing-dataset.ameshousing.train`
"""
