Current Requirement:
Currently, Metrax metrics require the user to manually apply the softmax function to their model's output logits first. The user must then pass the resulting probabilities (predictions) and the ground-truth labels into the metric for calculation.
Proposed Change:
To improve usability, we will introduce a new boolean flag, from_logits.
When from_logits=False (the default), the behavior will remain the same, expecting pre-computed probabilities.
When from_logits=True, users can pass the raw logits directly to the metric. Metrax will then be responsible for applying the softmax function internally before calculating the metric.
This change simplifies the user's workflow and reduces the chance of errors.
Current Requirement:
Currently, Metrax metrics require the user to manually apply the softmax function to their model's output logits first. The user must then pass the resulting probabilities (predictions) and the ground-truth labels into the metric for calculation.
Proposed Change:
To improve usability, we will introduce a new boolean flag, from_logits.
When from_logits=False (the default), the behavior will remain the same, expecting pre-computed probabilities.
When from_logits=True, users can pass the raw logits directly to the metric. Metrax will then be responsible for applying the softmax function internally before calculating the metric.
This change simplifies the user's workflow and reduces the chance of errors.