As you can see dvc repro will run 1 stage which creats m.json with same value as input param.

On the first step I initialized repo, ran dvc repro, and commited changes:

- git init
- dvc init
- dvc remote add -d rm_storage data
- dvc repro
- git add params.yaml dvc.yaml train.py data/.gitignore dvc.lock .dvc/config
- git commit -am c1

On the second step I changed params.yaml, ran dvc repro:

At this step my cache/runs dir was filled with 2 repros:

After I ran two experiments

(dvc exp run --set-param train.metric_input=0.81
dvc exp run --set-param train.metric_input=0.82)

my cache/runs dir looked like:

Then I commited changes and again ran dvc repro and 1 exp. On picture below you can see my cache/runs dir as last and output of dvc exp:

BUT when I ran dvc gc -w, nothing changed:

Additionally I can say that dvc gc -w --dry doesn-t work too:

Способы создания тензоров в PyTorch

Пример считывания csv файла с указанием колонки индекса и преобразования ее к типу даты и времени. Для указания колонки индекса используется параметр index_col, а в аргументе parse_dates можно перечислить колонки с датами.