Query the stop metric.

Helper function for input validation.

Given predictions and labels, compute a few metrics.

Given a 1D numpy array label_vec, return num_sel indices chosen at random from all indices where label_vec equals label_value. Note that

Store data from a batch for later use in computing metrics.

Compute and store end-of-phase metrics.

Initialize, run the training process, and save the results. Parameters P: Dictionary of parameters, which completely specify the trainin

Computes indices for a randomized split of num_ex objects into two parts, so we return two index vectors: idx_1 and idx_2. Note that idx_1 ha

Returns standard ImageNet statistics.

Return a copy of all log data.

Returns image transforms.

Set up for model training. Parameters P: Dictionary of parameters, which completely specify the training procedure. feature_extracto

label_matrix: binary (-1/+1) label matrix with shape num_items x num_classes num_pos: number of positive labels to observe for each item

Reset for a new phase.

Run one evaluation phase. Parameters model: Model to train. P: Dictionary of parameters, which completely specify the training proce

Run one training phase. Parameters model: Model to train. P: Dictionary of parameters, which completely specify the training procedu

Train the model. Parameters P: Dictionary of parameters, which completely specify the training procedure. Z: Dictionary of temporary

Update the current best epoch info if applicable.

Compute average precision. Parameters targs: Binary targets. preds: Predicted probability scores.

Compute precision@k. Parameters targs: Binary targets. preds: Predicted probability scores. k: Number of predictions to con

Compute recall@k. Parameters targs: Binary targets. preds: Predicted probability scores. k: Number of predictions to consid

Given a parameter dictionary P, initialize and return the specified dataset.