singtown_ai 1.3.0

SingTown AI Python Client

SingTown AI Python SDK

This SDK is designed to interact with SingTown AI Cloud Service or SingTown AI Standalone(self-hosted).

Installation

pip install singtown_ai

Usage

Login Configuration

• SingTown AI Cloud Service: The host is "https://ai.singtown.com".
• SingTown AI Standalone (self-hosted): The host is something like "http://127.0.0.1:8000".

You can obtain the token and task_id from Project -> Information.

Environment Variables:

export SINGTOWN_AI_HOST="https://ai.singtown.com"  # Or the cloud service URL
export SINGTOWN_AI_TOKEN="your token"            # Your token
export SINGTOWN_AI_TASK_ID="your id"             # Your task ID

Alternatively, set them directly in code:

SingTownAiClient(
  host="https://ai.singtown.com",  # Or the cloud service URL
  token="your token",            # Your token
  task_id="your id"              # Your task ID
)

Dry Run

python -m singtown_ai.dryrun --host=http://127.0.0.1:8000 --token=012345 --task_id=1

• This command will simulate 10s train task.

Basic Usage

from singtown_ai import SingTownAiClient

with SingTownAiClient() as client:
    pass  # Insert your code here

• This will periodically update the running status. After finished, it will post a "training succeeded" status. If an error occurs, it will post a "training failed" status.

Mock Usage

from singtown_ai import SingTownAiClient
mock_data = {
    "task": {
        "project": {
            "labels": ["cat", "dog"],
            "type": "CLASSIFICATION",
        },
        "type": "TRAIN",
        "status": "PENDING",
        "params": {
            "model": "MobileNetV2",
            "weight": "imagenet",
            "alpha": 0.35,
            "imgw": 96,
            "imgh": 96,
            "epochs": 1,
            "learning_rate": 0.001,
        },
    },
    "dataset": [
        {
            "url": "https://ai.singtown.com/media/cat.0.jpg",
            "subset": "TRAIN",
            "classification": "cat",
        },
        {
            "url": "https://ai.singtown.com/media/cat.1.jpg",
            "subset": "VALID",
            "classification": "cat",
        },
        {
            "url": "https://ai.singtown.com/media/cat.2.jpg",
            "subset": "TEST",
            "classification": "cat",
        },
        {
            "url": "https://ai.singtown.com/media/dog.0.jpg",
            "subset": "TRAIN",
            "classification": "dog",
        },
        {
            "url": "https://ai.singtown.com/media/dog.1.jpg",
            "subset": "VALID",
            "classification": "dog",
        },
        {
            "url": "https://ai.singtown.com/media/dog.2.jpg",
            "subset": "TEST",
            "classification": "dog",
        },
    ],
}
with SingTownAiClient(mock=True) as client:
    pass  # Insert your code here

• Set mock_data, Will mock demo task and dataset, this is useful for debugging.

Uploading Metrics

metrics = [
    {"epoch": 0, "accuracy": 0.8, "loss": 0.2},
    {"epoch": 1, "accuracy": 0.9, "loss": 0.1},
]
with SingTownAiClient() as client:
    client.upload_metrics(metrics)

• The field names in metrics are not restricted, and they will appear on the Metrics page in SingTown AI.

Watching metrics.csv

with SingTownAiClient(metrics_file="metrics.csv") as client:
    pass  # Insert your code here

• Every 3 seconds, the SDK will parse the metrics.csv and upload data.

Posting Logs

with SingTownAiClient() as client:
    import time
    for i in range(100):
        client.log(f"epoch: {i}")
        time.sleep(0.1)

• This will upload log strings, posting them every 3 seconds.

Download Trained Files from Server

with SingTownAiClient() as client:
    client.download_trained_file("folder")

• This method will download the trained file and automatically extract it into the specified folder.

Uploading Result Files

with SingTownAiClient() as client:
    client.upload_results_zip("your.zip")

• This method uploads a .zip result file.

Run Subprocess Command

with SingTownAiClient() as client:
    client.run_subprocess("echo hello world!")
    client.run_subprocess("python3 train.py", ignore_stdout=True)

• This method will run subprocess and log stdout and stderr.
• If ignore_stdout=True , will not log stdout.