Next, go to Terminal and make sure that the present working directory is the same as where you ran your Python code. Created directory which contains the event file Now if you run this code, it creates a directory inside your current directory (beside your Python code) which contains the event file.įig. Learning to use TensorBoard early and often will make working with TensorFlow much more enjoyable and productive. We'll cover this two main usages of TensorBoard in this tutorial. It is generally used for two main purposes:Ģ. TensorBoard was created as a way to help you understand the flow of tensors in your model so that you can debug and optimize it. When fully configured, TensorBoard window will look something like this: As explained in the previous tutorials, the idea is that you create a model that consists of a set of operations, feed data in to the model and the tensors will flow between the operations until you get an output tensor, your result. TensorFlow programs can range from very simple to super complex problems (using thousands of computations), and they all have two basic components, Operations and Tensors. To make it easier to understand, debug, and optimize TensorFlow programs, we've included a suite of visualization tools called TensorBoard.” In Google’s words: “The computations you'll use TensorFlow for (like training a massive deep neural network) can be complex and confusing.
Tensorflow board view values software#
Users can trace their pointers across a chart to make a tooltip appear.TensorBoard is a visualization software that comes with any standard TensorFlow installation. Hovering over the icon on the top right reveals the description.
Only runs with precision-recall data have step selectors. UsersĬan specify different steps for each run. Users can change the step at which to view PR curves via a step selector. Num_thresholds = 11) The Dashboard UI The Sidebar This method directly returns a tf.Summary proto.įrom tensorboard import summary as summary_lib import numpy as np labels = np. pr_curve_raw_data_pb accepts the analogous lists or numpy arrays. May be used outside of a TensorFlow environment to collect precision-recallĭata.
The pr_curve_raw_data_pb method is an analog of pr_curve_raw_data_op that run( merged_summary), global_step = step) pr_curve_raw_data_pb # We can also compute metrics such as F1 max to be shown in the scalar # dashboard. pr_curve_raw_data_op(ĭisplay_name = 'foo (really some random data)',ĭescription = 'Predictions are generated from a uniform distribution.') # Write the data to disk for visualization within the PR curve dashboard. Num_thresholds = 11 # `data` is a `PrecisionRecallData` named tuple, which contains several ops for # precision-recall related data (precision, recall, TP, FP, TN, FN). Use of precision-recall data to compute other values such as anįrom tensorboard import summary as summary_lib import tensorflow as tf labels = tf. The precision_recall_at_equal_thresholds metric also lets users further make May prefer using precision_recall_at_equal_thresholds. Hence, for instance, use cases that assign a prediction per pixel in a big image Predictions tensor (unlike pr_curve_streaming_op, which scales quadratically). The run time and space of this metric scales linearly with the size of the Tf._recall_at_equal_thresholds streaming metric toĭisk. The values within each tensor should correspond to threshold values (spanning 0Īs shown below, one effective use case is writing the output of the The pr_curve_raw_data_op method accepts a num_thresholds int as well as 6 Precision-recall data with this plugin's logic). Like to merely visualize that data within TensorBoard (instead of computing Sometimes, a project computes precision-recall data using custom logic and would Return metric_fn( predictions = predictions, labels = labels, weights = weights) """ # The streaming op accepts boolean labels, so we cast. Returns: A metric value that conforms to tflearn's API.
"""A metric for a binary classification problem. """Wrapper method that makes a metric out of the PR curve streaming op.""" def metric( predictions, labels, weights = None): From tensorboard import summary as summary_lib