Monitor

class tinder.monitor.Stat(alpha: float = 0.6)

A small class that calculates the statistics for a scalar metric. This is useful to calcaulate accuracies and losses across minibatches.

Parameters

alpha (float) – Defaults to 1. Weight(0~1) used for exponential moving average. A smaller weight reprsents more smooth average.

count

the number of samples.

Type

int

sum

Type

float

average

the average of samples. read-only.

Type

float

ema

exponential moving average.

Type

float

alpha

coefficient for exponential moving average.

Type

float

average

Average of all samples.

Returns

average

clear()

Clear the history and reset stats to zero.

update(value, count: int = 1, is_mean=False)

Update stats with new samples.

Parameters

• value (int or float) – new sample, or average of multiple samples.

• count (int) – Defaults to 1. This is useful when batch size is not uniform.

class tinder.monitor.Stats(**stats)

A class that holds named stats.

clear()

Clear all stats.