Aim.md

+[Aim](https://aimstack.io/#aim) is an interactive tool that helps to visualize and compare several metrics (losses, learning rates, etc.) between different runs.
+
+![aimui](uploads/630b5dcb8b615e7da5a7de3b6e12fef6/aimui.png)
+
+The main goal of Aim is to help developers during different iterations of model structure or data processing.
+
+## How it works
+Aim is a very powerful tool but at the same time is very simple to use. It defines three main interfaces:
+
+### Session
+Session class is the core of Aim. During the creation of a session, it is possible to define an experiment name and/or a run name: experiments are useful to group together runs related to the same model.
+
+Inside this template, experiment and run names are handled by the [config file](https://gitlab.fbk.eu/dsip/templates/dl_setup/-/wikis/Configuration-file) and run script helper.
+```python
+from aim import Session
+
+aim_session = Session(
+    repo='/aim/repo',
+    experiment='experiment_name',
+    run='run_name',
+)
+``` 
+
+Example:
+
+![aim_session](uploads/dbcc86e07952e54837a644b669bc2490/aim_session.png)
+
+The red arrow points to a "filter by experiment", and the green arrow points to runs related to the selected experiment.
+
+### set_params
+It is a session method used to set hyper-parameters associated with the run.
+
+Inside dl_setup, set_params is called automatically: all configs and hyper-paramters related to the run are saved. This approach allows to match a specific run (from [config file](https://gitlab.fbk.eu/dsip/templates/dl_setup/-/wikis/Configuration-file)) to related results (losses tracked from aim for example).
+```python
+aim_session.set_params(
+    { 'lr': 0.01, 'optimizer': 'Adam', 'batch_size': 64 },
+    name="hparams"
+)
+```
+
+Example:
+
+![aim_set_params](uploads/eafeee99fd773d15c87d7016eb9516cf/aim_set_params.png)
+
+### track
+It is a session method used to track a specific metric over time.
+
+```python
+aim_session.track(loss_value, name="loss",
+                  epoch=epoch_number, subset='train_phase')
+```
+
+It accepts any argument defined in the form *keyword=value*.
+
+Inside dl_setup, train and validation losses are already being tracked. If the user wants to add another metric (accuracy for example), they should call the `track` method of the session instance.
+
+Example:
+
+![aim_track](uploads/92368e641d6e07a60ae3aacc0c5e437b/aim_track.png)
+
+The X-axis represents epochs.
+The Y-axis represents loss value.
+The red line is validation loss.
+The green line is train loss.
+
+### Aim on cluster
+At the moment there is no option to be able to visualize on a dashboard the aim logs that are stored in the cluster. We are currently developing a feature that enables aim to send the tracked data to an exposed online webserver. Additional info will be provided as soon as possible.
+
+For the moment, in order to visualize aim logs, it is necessary to copy the `.aim` folder on your local machine.