dvc: prevent printing html path

[sirily]

I'm using dvclive with pytorch lightning in dvc pipeline. While training there is a string with the path to metrics html, and it breaks progress bar and fills the console with useless messages. Is there a way to disable printing this line?

Epoch 10: 8%|█▊ | 103/1361 [00:11<02:13, 9.42it/s, loss=1, v_num=_run, train_positive_step=0.988]
file:///path/index.html
Epoch 10: 15%|███▋ | 209/1361 [00:22<02:01, 9.51it/s, loss=1, v_num=_run, train_positive_step=0.985]
file:///path/index.html
Epoch 10: 22%|█████▍ | 305/1361 [00:32<01:50, 9.53it/s, loss=1, v_num=_run, train_positive_step=0.991]
file:///path/index.html
Epoch 10: 29%|███████ | 400/1361 [00:42<01:40, 9.53it/s, loss=1, v_num=_run, train_positive_step=0.984]
file:///path/index.html
Epoch 10: 37%|████████▉ | 505/1361 [00:53<01:29, 9.53it/s, loss=1, v_num=_run, train_positive_step=0.990]
file:///path/index.html

Inside train stage logger is defined like this:

logger = DvcLiveLogger()
model = Trainer(hparams, model_path)
trainer = pl.Trainer(
                    gpus=1,
                    logger=logger,
                    max_epochs=hparams['num_epochs'],
                    log_every_n_steps=hparams['log_every_n_steps'],
                    callbacks=[checkpoint_callback, early_stopping],
                    stochastic_weight_avg=True)

in dvc.yaml stage looks like this:

cmd: python3 pipeline/train.py --task ${item.task} --inp_dir ${item.in} --out_dir ${item.out}
deps:
- data/processed/train_${item.task}_df.par
- data/processed/val_${item.task}_df.par
- models/${item.in}
- pipeline/train.py
- src/trainer.py
params:
- ${item.task}_train
outs:
- models/${item.out}
live:
  metrics/${item.out}_train:
    cache: false
    summary: true
    html: true

It seems like nothing bad happens in dvclive, maybe it prints because of summary path here. But @shcheklein asked me to open an issue here.

dvc doctor:

DVC version: 2.7.4 (pip)

---

Platform: Python 3.8.10 on Linux-5.4.0-72-generic-x86_64-with-glibc2.29
Supports:
hdfs (pyarrow = 5.0.0),
http (aiohttp = 3.7.4.post0, aiohttp-retry = 2.4.5),
https (aiohttp = 3.7.4.post0, aiohttp-retry = 2.4.5),
s3 (s3fs = 2021.8.1, boto3 = 1.17.106)
Cache types: hardlink, symlink
Cache directory: ext4 on /dev/mapper/vg0-root
Caches: local
Remotes: s3
Workspace directory: ext4 on /dev/mapper/vg0-root
Repo: dvc (subdir), git

dvclive version: 0.4.1

[daavoo]

Thanks for the report @sirily !

You are right, the html path is being printed by DVC. Maybe we can lower the logging level from info to debug and/or only print the html path once, at the beginning of training, instead of at the end of each epoch.

[daavoo]

Hei @sirily the fix will be shipped in the next DVC release. It has been already merged so you could install from pre-release version for now.