Monitors

Monitors

Monitor([mutual_info, normalize_inverse])

Generic Monitor that provides meaningful statistics in interactive Visdom plots throughout training.

MonitorEmbedding([mutual_info, ...])

A monitor for mighty.trainer.TrainerEmbedding.

MonitorAutoencoder([mutual_info, ...])

A monitor for mighty.trainer.TrainerAutoencoder.

Monitor Parameter Records

ParamRecord(param[, monitor_level])

A parameter record, created by a monitor, that tracks parameter statistics like gradient variance, sign flips on update step, etc.

ParamsDict()

A dictionary that holds named `ParamRecord`s.

Mutual Information Estimates

MutualInfoNeuralEstimation(data_loader[, ...])

Mutual Information Neural Estimation [Rd32d51429fda-1], followed by PCA dimensionality reduction.

MutualInfoGCMI(data_loader[, pca_size, debug])

Gaussian-Copula Mutual Information Estimator [R07d168308d56-1], followed by PCA dimensionality reduction.

MutualInfoNPEET(data_loader[, pca_size, debug])

Non-parametric Kraskov-like Mutual Information Estimator [R9b99107211de-1], followed by PCA dimensionality reduction.

MutualInfoKMeans(data_loader[, n_bins, debug])

Classical binning approach to estimate the mutual information.

MutualInfoStub()

Mutual Info stub that does nothing.

Accuracy measures

AccuracyArgmax()

Softmax accuracy.

AccuracyEmbedding([metric, cache])

Calculates the accuracy of embedding vectors.

Timer and schedulers

BatchTimer()

A global batch timer.

ScheduleStep([epoch_step, batch_step])

Performs an update each epoch_step * timer.epoch_size + batch_step batches.

ScheduleExp()

Schedule updates at batches that are powers of two: 1, 2, 4, 8, 16, .

Visdom server

VisdomMighty([env, offline])

A Visdom server that updates measures in online fashion.