Just load an experiment to start plotting your metrics or try with some examples.
Experiments data have to be in the json format and collected in an inner dictionary having the performances names as keys and the performances values written in a list associated to the relative key. This dictionary has then to be wrapped in an outer dictionary that has the runs names as keys and the aforesaid inner dictionaries as corresponding values.
Example: for the recordings of a RUN_1 and a RUN_2 keeping track of a metric_a, a metric_b and a metric_c, the resulting json file has to be
{
"RUN_1": {
"metric_a": [metric_a_of_run_1_at_epoch_1, metric_a_of_run_1_at_epoch_2, ...],
"metric_b": [metric_b_of_run_1_at_epoch_1, metric_b_of_run_1_at_epoch_2, ...],
"metric_c": [metric_c_of_run_1_at_epoch_1, metric_c_of_run_1_at_epoch_2, ...],
},
"RUN_2": {
"metric_a": [metric_a_of_run_2_at_epoch_1, metric_a_of_run_2_at_epoch_2, ...],
"metric_b": [metric_b_of_run_2_at_epoch_1, metric_b_of_run_2_at_epoch_2, ...],
"metric_c": [metric_c_of_run_2_at_epoch_1, metric_c_of_run_2_at_epoch_2, ...],
},
}