{"id":345,"date":"2020-09-21T15:12:44","date_gmt":"2020-09-21T23:12:44","guid":{"rendered":"https:\/\/live-cometml.pantheonsite.io\/?p=345"},"modified":"2025-04-24T17:30:39","modified_gmt":"2025-04-24T17:30:39","slug":"logging-histograms-gradients-and-activations-with-comet","status":"publish","type":"post","link":"https:\/\/www.comet.com\/site\/blog\/logging-histograms-gradients-and-activations-with-comet\/","title":{"rendered":"Logging Histograms, Gradients and Activations with Comet"},"content":{"rendered":"\n

Introduction<\/h2>\n\n\n\n

3D Histograms or Ridge Plots are a great way to visualize the training progress of your Neural Network. Histogram distributions of the weights, gradients, and activations allow us to get some intuition for the loss surface that we are trying to optimize.<\/p>\n\n\n\n

For example, your model may be approaching a local minima if you observe that your gradients are becoming smaller over time. On the flip side, a trend of large gradients with high variance might imply that you should reduce your learning rate, or use some form of regularization.<\/p>\n\n\n\n

The Comet SDK provides an easy way to visualize your weights, activations and gradients using the `log_histogram_3D` method. For this post we will provide examples of histogram logging with Comet using Tensorflow\u2019s Gradient Tape.<\/p>\n\n\n\n

You can explore the visualizations in this post here<\/a>. We have also included Colab Notebooks at the end of this post so that you can try out the histograms feature for yourself!<\/p>\n\n\n\n

Logging Histograms<\/h2>\n\n\n\n

For this example, we are going to use a simple 2 layer perceptron and train it on the MNIST dataset. Let\u2019s start by loading in our data, and defining our model<\/p>\n\n\n\n

\n
\n\n\n\n
\n
1<\/div>\n
2<\/div>\n
3<\/div>\n
4<\/div>\n
5<\/div>\n
6<\/div>\n
7<\/div>\n
8<\/div>\n
9<\/div>\n
10<\/div>\n
11<\/div>\n
12<\/div>\n
13<\/div>\n
14<\/div>\n
15<\/div>\n
16<\/div>\n
17<\/div>\n
18<\/div>\n
19<\/div>\n
20<\/div>\n
21<\/div>\n
22<\/div>\n
23<\/div>\n
24<\/div>\n
25<\/div>\n
26<\/div>\n
27<\/div>\n
28<\/div>\n<\/td>\n
\n
\n
def get_dataset():<\/code><\/div>\n
    <\/code>num_classes = 10<\/code><\/div>\n
<\/div>\n
    <\/code># the data, shuffled and split between train and test sets<\/code><\/div>\n
    <\/code>(x_train, y_train), (x_test, y_test) = mnist.load_data()<\/code><\/div>\n
<\/div>\n
    <\/code>x_train = x_train.reshape(60000, 784)<\/code><\/div>\n
    <\/code>x_test = x_test.reshape(10000, 784)<\/code><\/div>\n
    <\/code>x_train = x_train.astype(\"float32\")<\/code><\/div>\n
    <\/code>x_test = x_test.astype(\"float32\")<\/code><\/div>\n
    <\/code>x_train \/= 255<\/code><\/div>\n
    <\/code>x_test \/= 255<\/code><\/div>\n
    <\/code>print(x_train.shape[0], \"train samples\")<\/code><\/div>\n
    <\/code>print(x_test.shape[0], \"test samples\")<\/code><\/div>\n
<\/div>\n
    <\/code># convert class vectors to binary class matrices<\/code><\/div>\n
    <\/code>y_train = to_categorical(y_train, num_classes)<\/code><\/div>\n
    <\/code>y_test = to_categorical(y_test, num_classes)<\/code><\/div>\n
<\/div>\n
    <\/code>return x_train, y_train, x_test, y_test<\/code><\/div>\n
<\/div>\n
def build_model_graph():<\/code><\/div>\n
    <\/code>model = Sequential()<\/code><\/div>\n
    <\/code>model.add(Dense(128, activation=\"sigmoid\", input_shape=(784,), name=\"dense1\"))<\/code><\/div>\n
    <\/code>model.add(Dense(64, activation=\"sigmoid\", name=\"dense2\"))<\/code><\/div>\n
    <\/code>model.add(Dense(10, activation=\"softmax\", name=\"output\"))<\/code><\/div>\n
<\/div>\n
    <\/code>return model<\/code><\/div>\n<\/div>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<\/div>\n\n\n\n

Tensorflow\u2019s GradientTape allows eager execution of model code without precomputing a static graph in which inputs are fed in through placeholders.<\/p>\n\n\n\n

This implies that at every training step, we will have to calculate the gradients of our parameters with respect to our loss, and apply them with the optimizer in order to perform backpropagation.<\/p>\n\n\n\n

\n
\n\n\n\n
\n
1<\/div>\n
2<\/div>\n
3<\/div>\n
4<\/div>\n
5<\/div>\n
6<\/div>\n
7<\/div>\n
8<\/div>\n
9<\/div>\n
10<\/div>\n
11<\/div>\n
12<\/div>\n<\/td>\n
\n
\n
def step(model, X, y, gradmap={}, activations={}):<\/code><\/div>\n
    <\/code>with tf.GradientTape() as tape:<\/code><\/div>\n
        <\/code>pred = model(X)<\/code><\/div>\n
        <\/code>loss = categorical_crossentropy(y, pred)<\/code><\/div>\n
<\/div>\n
    <\/code>grads = tape.gradient(loss, model.trainable_variables)<\/code><\/div>\n
    <\/code>opt.apply_gradients(zip(grads, model.trainable_variables))<\/code><\/div>\n
<\/div>\n
    <\/code>gradmap = get_gradients(gradmap, grads, model)<\/code><\/div>\n
    <\/code>activations = get_activations(activations, X, model)<\/code><\/div>\n
<\/div>\n
    <\/code>return loss.numpy().mean(), gradmap, activations<\/code><\/div>\n<\/div>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<\/div>\n\n\n\n

We will use two dictionaries to store layerwise information about the gradients and activations. For each batch of data we will accumulate the values of the gradients and activations. At the end of every epoch we will scale these by the number of batches in an epoch, and log them to Comet.<\/p>\n\n\n\n

\n
\n\n\n\n
\n
1<\/div>\n
2<\/div>\n
3<\/div>\n
4<\/div>\n
5<\/div>\n
6<\/div>\n
7<\/div>\n
8<\/div>\n
9<\/div>\n
10<\/div>\n
11<\/div>\n
12<\/div>\n
13<\/div>\n
14<\/div>\n
15<\/div>\n
16<\/div>\n
17<\/div>\n
18<\/div>\n
19<\/div>\n
20<\/div>\n
21<\/div>\n
22<\/div>\n
23<\/div>\n
24<\/div>\n
25<\/div>\n
26<\/div>\n
27<\/div>\n
28<\/div>\n
29<\/div>\n
30<\/div>\n
31<\/div>\n
32<\/div>\n
33<\/div>\n
34<\/div>\n
35<\/div>\n
36<\/div>\n
37<\/div>\n
38<\/div>\n
39<\/div>\n
40<\/div>\n<\/td>\n
\n
\n
def train(model, X, y, epoch, steps_per_epoch, experiment):<\/code><\/div>\n
    <\/code>gradmap = {}<\/code><\/div>\n
    <\/code>activations = {}<\/code><\/div>\n
    <\/code>total_loss = 0<\/code><\/div>\n
    <\/code>with experiment.train():<\/code><\/div>\n
        <\/code># show the current epoch number<\/code><\/div>\n
        <\/code>print(\"[INFO] starting epoch {}\/{}...\".format(epoch, EPOCHS), end=\"\")<\/code><\/div>\n
        <\/code>for i in range(0, steps_per_epoch):<\/code><\/div>\n
            <\/code>start = i * BS<\/code><\/div>\n
            <\/code>end = start + BS<\/code><\/div>\n
            <\/code>curr_step = ((i + 1) * BS) * epoch<\/code><\/div>\n
            <\/code><\/div>\n
            <\/code>loss, gradmap, activations = step(<\/code><\/div>\n
                <\/code>model,<\/code><\/div>\n
                <\/code>X[start:end],<\/code><\/div>\n
                <\/code>y[start:end],<\/code><\/div>\n
                <\/code>gradmap,<\/code><\/div>\n
                <\/code>activations,<\/code><\/div>\n
            <\/code>)<\/code><\/div>\n
            <\/code>experiment.log_metric(<\/code><\/div>\n
              <\/code>\"batch_loss\", loss, step=curr_step<\/code><\/div>\n
            <\/code>)            <\/code><\/div>\n
            <\/code><\/div>\n
            <\/code>total_loss += loss<\/code><\/div>\n
        <\/code><\/div>\n
        <\/code>experiment.log_metric(<\/code><\/div>\n
            <\/code>\"loss\", total_loss \/ steps_per_epoch, step=epoch * steps_per_epoch<\/code><\/div>\n
        <\/code>)<\/code><\/div>\n
<\/div>\n
    <\/code># scale gradients<\/code><\/div>\n
    <\/code>for k, v in gradmap.items():<\/code><\/div>\n
        <\/code>gradmap[k] = v \/ steps_per_epoch<\/code><\/div>\n
<\/div>\n
    <\/code># scale activations<\/code><\/div>\n
    <\/code>for k, v in activations.items():<\/code><\/div>\n
        <\/code>activations[k] = v \/ steps_per_epoch<\/code><\/div>\n
<\/div>\n
    <\/code>log_weights(experiment, model, epoch * steps_per_epoch)<\/code><\/div>\n
    <\/code>log_histogram(experiment, gradmap, epoch * steps_per_epoch, prefix=\"gradient\")<\/code><\/div>\n
    <\/code>log_histogram(experiment, activations, epoch * steps_per_epoch, prefix=\"activation\")<\/code><\/div>\n<\/div>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<\/div>\n\n\n\n

We can visualize these reported values under the Histograms tab in our Comet Experiment<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

<\/p>\n\n\n\n

Figure 1. Logged Histograms in Comet<\/strong> (link to Comet Experiment<\/a>)<\/figcaption><\/figure>\n\n\n\n

We can sort and search the various histograms using the options in the menu. For example, we might want to group our histograms based on the reported names<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

<\/p>\n\n\n\n

\"\"
Figure 2. Selecting grouping for Histograms (link to Comet Experiment<\/a>)<\/strong><\/strong>
\"\"<\/figure><\/figcaption><\/figure>\n\n\n\n
\"\"
Figure 3. Histograms Grouped by Name (link to Comet Experiment<\/a>)<\/strong><\/strong>
\"\"<\/figure><\/figcaption><\/figure>\n\n\n\n
\n
Figure 4. Histograms Grouped by Name Expanded View<\/strong><\/strong> <\/figcaption>\n<\/figure>\n\n\n\n

For larger models, we may have many such histograms. Comet makes it possible to search for the specific histogram that you\u2019re interested in. In Figure 5 we can see how we are able to search for the histogram distribution of the gradients of a single layer.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

<\/p>\n\n\n\n

\n
Figure 5. Filtered Histograms based on Name<\/strong><\/figcaption>\n<\/figure>\n\n\n\n

Additionally, Comet provides Custom Panels<\/a> that can visualize these histograms as a heatmap over time. This particular custom panel can be found in Panels Gallery \u2192 Public Panels \u2192 Histogram by Step. After adding the Panel to the project, we should see it above the Experiment Table.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

<\/p>\n\n\n\n

\"\"
Figure 6. Adding a custom panel to visualize histograms<\/strong><\/strong>
\"\"<\/figure><\/figcaption><\/figure>\n\n\n\n
\n
Figure 7. Custom Panel to Visualize Histograms in the Project<\/strong> <\/figcaption>\n<\/figure>\n\n\n\n

This custom panel has drop downs for both the experiment ids and the histogram names. We can select a particular experiment from the dropdown and then visualize the individual histograms in that experiment as a heat map over time.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

<\/p>\n\n\n\n

\"\"
Figure 8. Selecting an experiment from the dropdown in a Custom Panel<\/strong><\/strong>
\"\"<\/figure><\/figcaption><\/figure>\n\n\n\n
\n
Figure 9. Selecting a histogram in the experiment from the dropdown in a Custom Panel<\/strong><\/figcaption>\n<\/figure>\n\n\n\n

Conclusion<\/h2>\n\n\n\n

In this post we demonstrated how Comet\u2019s histogram logging and Custom Panels can be used to visualize the weights, gradients, and activations of a Neural Network.<\/p>\n\n\n\n

Colab Notebooks :<\/strong><\/p>\n\n\n\n

TF Gradient Tape Example<\/a><\/p>\n\n\n\n

Pytorch Example<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

Introduction 3D Histograms or Ridge Plots are a great way to visualize the training progress of your Neural Network. Histogram distributions of the weights, gradients, and activations allow us to get some intuition for the loss surface that we are trying to optimize. For example, your model may be approaching a local minima if you […]<\/p>\n","protected":false},"author":4,"featured_media":346,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"customer_name":"","customer_description":"","customer_industry":"","customer_technologies":"","customer_logo":"","footnotes":""},"categories":[8,9],"tags":[],"coauthors":[128],"class_list":["post-345","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-comet-community-hub","category-product"],"yoast_head":"\nLogging Histograms, Gradients and Activations with Comet<\/title>\n<meta name=\"description\" content=\"Guide to logging histograms, gradients and activations with Comet. 3D Histograms or Ridge Plots visualize the training progress of your Neural Network.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.comet.com\/site\/blog\/logging-histograms-gradients-and-activations-with-comet\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Logging Histograms, Gradients and Activations with Comet\" \/>\n<meta property=\"og:description\" content=\"Guide to logging histograms, gradients and activations with Comet. 3D Histograms or Ridge Plots visualize the training progress of your Neural Network.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.comet.com\/site\/blog\/logging-histograms-gradients-and-activations-with-comet\/\" \/>\n<meta property=\"og:site_name\" content=\"Comet\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/cometdotml\" \/>\n<meta property=\"article:published_time\" content=\"2020-09-21T23:12:44+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2025-04-24T17:30:39+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.comet.com\/site\/wp-content\/uploads\/2022\/02\/Screen-Shot-2020-07-29-at-6.55.34-PM-1024x562-1.png\" \/>\n\t<meta property=\"og:image:width\" content=\"1024\" \/>\n\t<meta property=\"og:image:height\" content=\"562\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"Dhruv Nair\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@Cometml\" \/>\n<meta name=\"twitter:site\" content=\"@Cometml\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Dhruv Nair\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"7 minutes\" \/>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"Logging Histograms, Gradients and Activations with Comet","description":"Guide to logging histograms, gradients and activations with Comet. 3D Histograms or Ridge Plots visualize the training progress of your Neural Network.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.comet.com\/site\/blog\/logging-histograms-gradients-and-activations-with-comet\/","og_locale":"en_US","og_type":"article","og_title":"Logging Histograms, Gradients and Activations with Comet","og_description":"Guide to logging histograms, gradients and activations with Comet. 3D Histograms or Ridge Plots visualize the training progress of your Neural Network.","og_url":"https:\/\/www.comet.com\/site\/blog\/logging-histograms-gradients-and-activations-with-comet\/","og_site_name":"Comet","article_publisher":"https:\/\/www.facebook.com\/cometdotml","article_published_time":"2020-09-21T23:12:44+00:00","article_modified_time":"2025-04-24T17:30:39+00:00","og_image":[{"width":1024,"height":562,"url":"https:\/\/www.comet.com\/site\/wp-content\/uploads\/2022\/02\/Screen-Shot-2020-07-29-at-6.55.34-PM-1024x562-1.png","type":"image\/png"}],"author":"Dhruv Nair","twitter_card":"summary_large_image","twitter_creator":"@Cometml","twitter_site":"@Cometml","twitter_misc":{"Written by":"Dhruv Nair","Est. reading time":"7 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/www.comet.com\/site\/blog\/logging-histograms-gradients-and-activations-with-comet\/#article","isPartOf":{"@id":"https:\/\/www.comet.com\/site\/blog\/logging-histograms-gradients-and-activations-with-comet\/"},"author":{"name":"Matt Peternell","@id":"https:\/\/www.comet.com\/site\/#\/schema\/person\/85aa446f8be987e848ea929ef524b67b"},"headline":"Logging Histograms, Gradients and Activations with Comet","datePublished":"2020-09-21T23:12:44+00:00","dateModified":"2025-04-24T17:30:39+00:00","mainEntityOfPage":{"@id":"https:\/\/www.comet.com\/site\/blog\/logging-histograms-gradients-and-activations-with-comet\/"},"wordCount":603,"commentCount":0,"publisher":{"@id":"https:\/\/www.comet.com\/site\/#organization"},"image":{"@id":"https:\/\/www.comet.com\/site\/blog\/logging-histograms-gradients-and-activations-with-comet\/#primaryimage"},"thumbnailUrl":"https:\/\/www.comet.com\/site\/wp-content\/uploads\/2022\/02\/Screen-Shot-2020-07-29-at-6.55.34-PM-1024x562-1.png","articleSection":["Comet Community Hub","Product"],"inLanguage":"en-US","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/www.comet.com\/site\/blog\/logging-histograms-gradients-and-activations-with-comet\/#respond"]}]},{"@type":"WebPage","@id":"https:\/\/www.comet.com\/site\/blog\/logging-histograms-gradients-and-activations-with-comet\/","url":"https:\/\/www.comet.com\/site\/blog\/logging-histograms-gradients-and-activations-with-comet\/","name":"Logging Histograms, Gradients and Activations with Comet","isPartOf":{"@id":"https:\/\/www.comet.com\/site\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.comet.com\/site\/blog\/logging-histograms-gradients-and-activations-with-comet\/#primaryimage"},"image":{"@id":"https:\/\/www.comet.com\/site\/blog\/logging-histograms-gradients-and-activations-with-comet\/#primaryimage"},"thumbnailUrl":"https:\/\/www.comet.com\/site\/wp-content\/uploads\/2022\/02\/Screen-Shot-2020-07-29-at-6.55.34-PM-1024x562-1.png","datePublished":"2020-09-21T23:12:44+00:00","dateModified":"2025-04-24T17:30:39+00:00","description":"Guide to logging histograms, gradients and activations with Comet. 3D Histograms or Ridge Plots visualize the training progress of your Neural Network.","breadcrumb":{"@id":"https:\/\/www.comet.com\/site\/blog\/logging-histograms-gradients-and-activations-with-comet\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.comet.com\/site\/blog\/logging-histograms-gradients-and-activations-with-comet\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.comet.com\/site\/blog\/logging-histograms-gradients-and-activations-with-comet\/#primaryimage","url":"https:\/\/www.comet.com\/site\/wp-content\/uploads\/2022\/02\/Screen-Shot-2020-07-29-at-6.55.34-PM-1024x562-1.png","contentUrl":"https:\/\/www.comet.com\/site\/wp-content\/uploads\/2022\/02\/Screen-Shot-2020-07-29-at-6.55.34-PM-1024x562-1.png","width":1024,"height":562,"caption":"Keras Gradient Image | Comet ML"},{"@type":"BreadcrumbList","@id":"https:\/\/www.comet.com\/site\/blog\/logging-histograms-gradients-and-activations-with-comet\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/www.comet.com\/site\/"},{"@type":"ListItem","position":2,"name":"Logging Histograms, Gradients and Activations with Comet"}]},{"@type":"WebSite","@id":"https:\/\/www.comet.com\/site\/#website","url":"https:\/\/www.comet.com\/site\/","name":"Comet","description":"Build Better Models Faster","publisher":{"@id":"https:\/\/www.comet.com\/site\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.comet.com\/site\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/www.comet.com\/site\/#organization","name":"Comet ML, Inc.","alternateName":"Comet","url":"https:\/\/www.comet.com\/site\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.comet.com\/site\/#\/schema\/logo\/image\/","url":"https:\/\/www.comet.com\/site\/wp-content\/uploads\/2025\/01\/logo_comet_square.png","contentUrl":"https:\/\/www.comet.com\/site\/wp-content\/uploads\/2025\/01\/logo_comet_square.png","width":310,"height":310,"caption":"Comet ML, Inc."},"image":{"@id":"https:\/\/www.comet.com\/site\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/cometdotml","https:\/\/x.com\/Cometml","https:\/\/www.youtube.com\/channel\/UCmN63HKvfXSCS-UwVwmK8Hw"]},{"@type":"Person","@id":"https:\/\/www.comet.com\/site\/#\/schema\/person\/85aa446f8be987e848ea929ef524b67b","name":"Matt Peternell","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.comet.com\/site\/#\/schema\/person\/image\/da003ee51bbeeccfb95147ec69139879","url":"https:\/\/secure.gravatar.com\/avatar\/36058153d701caaf237a96d5d6fb9c2d1678325c3ed0d8e88bf5e487019a2a53?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/36058153d701caaf237a96d5d6fb9c2d1678325c3ed0d8e88bf5e487019a2a53?s=96&d=mm&r=g","caption":"Matt Peternell"},"description":"We re-implemented the architecture of this model to incorporate patient and study information. By comparing our updated model to the original Github repository, we were able to quantify the benefits of classifying by patient as opposed to classifying by individual X-ray. We observed a 0.0254 increase in AUROC when evaluating the DenseNet121 on patients instead of on individual scans.","sameAs":["http:\/\/atre.net"],"url":"https:\/\/www.comet.com\/site\/blog\/author\/mpeternellatre-net\/"}]}},"_links":{"self":[{"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/posts\/345","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/comments?post=345"}],"version-history":[{"count":1,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/posts\/345\/revisions"}],"predecessor-version":[{"id":15693,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/posts\/345\/revisions\/15693"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/media\/346"}],"wp:attachment":[{"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/media?parent=345"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/categories?post=345"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/tags?post=345"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/coauthors?post=345"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}