Using multiple loggers in the config file when training

[farrandi]

I made a custom logger which I added to the config.cfg file:
@loggers = "mlflow_logger.v1"

It works and logs the values to mlflow but it does not show the results in the console. I want spacy to both log it to mlflow and print it to the console. Is there a way I could do this?

I tried logging 2 loggers by:
@loggers = "spacy.ConsoleLogger.v1", "mlflow_logger.v1"

but it did not work, gave me this error:
catalogue.RegistryError: [E893] Could not find function '"spacy.ConsoleLogger.v1", "mlflow_logger.v1"' in function registry 'loggers'.

[polm]

I don't believe there's a way to specify multiple loggers at the moment. What you can do is write a custom logger that sends data to both the console and the mlflow logger.

[farrandi]

Sorry for the late reply, I tried doing it and it works However, I do have some formatting problems:

=========================== Initializing pipeline ===========================
[2021-11-12 12:02:52,862] [INFO] Set up nlp object from config
[2021-11-12 12:02:52,870] [INFO] Pipeline: ['sentencizer', 'tok2vec', 'ner']
[2021-11-12 12:02:53,145] [INFO] Created vocabulary
[2021-11-12 12:02:53,146] [INFO] Finished initializing nlp object
[2021-11-12 12:02:58,148] [INFO] Initialized pipeline components: ['tok2vec', 'ner']
✔ Initialized pipeline

============================= Training pipeline =============================
ℹ Pipeline: ['sentencizer', 'tok2vec', 'ner']
ℹ Frozen components: ['sentencizer']
ℹ Initial learn rate: 0.001
step
score
loss ner


0
0.0
61.69229990243912


200
0.46699015617934536
8290.14936476677

It prints something like above in the console, when I want it to be in a table-like format (usual spacy)

This is my current code:

@spacy.registry.loggers("mlflow_logger.v1")
def mlflow_logger():
    def setup_logger(
        nlp: Language,
        stdout: IO=sys.stdout,
        stderr: IO=sys.stderr
    )-> Tuple[Callable, Callable]:
        print("step\t")
        print("score\t")
        print("loss ner\t")
        print("\n")
        def log_step(info: Optional[Dict[str, Any]]):
            if info:
                mlflow.log_metric("step", info['step'])
                mlflow.log_metric("score", info['score'])
                mlflow.log_metric(f"loss_ner", info['losses']['ner'])

                print(f"{info['step']}\t")
                print(f"{info['score']}\t")
                print(f"{info['losses']['ner']}\t")
                print("\n")
        def finalize():
            # mlflow.set_terminated(mlflow.active_run().info.run_id)
             print("Model saved in run %s" % mlflow.active_run().info.run_uuid)

        return log_step, finalize

    return setup_logger

Is there a way I could see the code in which spacy normally does it?

[polm]

Code for the default ConsoleLogger is here.

                print(f"{info['step']}\t")
                print(f"{info['score']}\t")
                print(f"{info['losses']['ner']}\t")
                print("\n")

This code is your issue - print doesn't just write to stdout, it automatically adds a newline at the end of each call. If you combine all your info there into one call to print you should get the output you want. You also don't have to add a newline.

[farrandi]

I see, thanks!