{"id":9432,"date":"2024-03-12T07:44:42","date_gmt":"2024-03-12T15:44:42","guid":{"rendered":"https:\/\/live-cometml.pantheonsite.io\/?p=9432"},"modified":"2025-04-24T17:02:58","modified_gmt":"2025-04-24T17:02:58","slug":"transfer-learning-with-comet","status":"publish","type":"post","link":"https:\/\/www.comet.com\/site\/blog\/transfer-learning-with-comet\/","title":{"rendered":"Navigating Transfer Learning with Comet"},"content":{"rendered":"\n
\n
\n
\n
\n
\"\"<\/figure>
\n

<\/picture><\/h2>\n<\/div>
Photo by Daniel Korpai<\/a> on Unsplash<\/a><\/figcaption><\/figure>\n

Transfer learning involves using a pre-trained model to solve a deep-learning problem. If your problem is not necessarily unique, it is better to create a model that has already been trained on a given task.<\/p>\n

Comet extensively integrates with multiple libraries, including the most common ones for deep learning, such as TensorFlow, Keras, and Pytorch.<\/p>\n

These integrations are well-supported and allow extensive types of workflows, including those that require transfer learning. This article will dive deep into how you can make great observations and visualizations with Comet. I will also indicate how easy it is to integrate Comet with previous projects, as it takes a few lines of code here and there to get things running.<\/p>\n

Now, for this task, you will need a few things:<\/p>\n

    \n
  1. A Comet account. Sign up here<\/a>.<\/li>\n
  2. A Python 3.9+ install.<\/li>\n
  3. The following libraries: Comet, ScikitLearn, Pandas, and Keras.<\/li>\n
  4. A fierce learning spirit.<\/li>\n<\/ol>\n

    Let’s dive into the tiny project now.<\/p>\n<\/div>\n<\/div>\n<\/div>\n\n\n\n

    \n
    \n
    \n

    Project Overview<\/h2>\n

    In our tiny project, I will use one of my favorite datasets. The beauty of this dataset is that it has just the right amount of difficulty to ensure that you learn numerous concepts in a deep learning problem. We are going to be using data from Kaggle’s famous Cats-and-Dogs<\/a> problem.<\/p>\n

    The objective here is not to gain perfection or even high accuracy, as this is purely a learning exercise; hence, there are a few tradeoffs that I will highlight here.<\/p>\n

    We are going to build a simple, bare-bones model and train it only on a very small sample of the data we downloaded. The data will only be trained for a few epochs. From this, we can ascertain that there will be poor accuracy. It could be better with proper interventions, though.<\/p>\n

    Since that is established, let’s code.<\/p>\n

    Dataset Preparation<\/h2>\n

    We first store the information in the dataset in a Pandas DataFrame. The DataFrame will contain two columns: one for the file names and the other for each image’s class.<\/p>\n

    import<\/span> pandas as<\/span> pd\nimport<\/span> os\n\ntrain_dir = r\".\/src\/train_data\"<\/span>\n\n#path to training data<\/span>\nfilenames = os.listdir(train_dir)\n\n#separating cats from dogs<\/span>\ncategories = []\nfor<\/span> f_name in<\/span> filenames:\n    category = f_name.split('.'<\/span>)[0<\/span>]\n    if<\/span> category == 'cat'<\/span>:\n        categories.append(0<\/span>)\n    else<\/span>:\n        categories.append(1<\/span>)\n\ndf = pd.DataFrame({\n        'filename'<\/span>: filenames,\n        'category'<\/span>: categories,\n})\n\n#replacing booleans with names<\/span>\ndf[\"category\"<\/span>] = df[\"category\"<\/span>].replace({0<\/span>:'cat'<\/span>, 1<\/span>:'dog'<\/span>})\n\n#storing info in csv files<\/span>\ndf.to_csv(\"new_training_info.csv\"<\/span>, index=False<\/span>)<\/span><\/pre>\n
    After doing this, the table will look as follows:<\/p>\n
    \n
    \"\"<\/figure>
    <\/picture><\/div>\n
    Screenshot by author<\/figcaption>\n<\/figure>\n
    Parameter Definition<\/h2>\n
    We need to define a few parameters. One set will go into the model, while the other will be fed into the DataLoaders.<\/p>\n
    The parameters belonging to the model will be logged into Comet, allowing the training process to be monitored and properly visualized.<\/p>\n
    Our parameters will appear as below:<\/p>\n
    img_height = 224<\/span>\nimg_width = 224<\/span>\nimg_channels = 3<\/span>\nimg_size = (img_height, img_width)\ntrain_dir = r\".\/src\/train_data\"<\/span>\nbatch_size = 32<\/span><\/span><\/pre>\n
    The image fed into the model we will use will be 224 x 224 x 3 pixels, so all the images must be transformed into these sizes. The model will take batches of 32 images for training from the “train_data” directory.<\/p>\n
    Initializing Comet and Data Generators<\/h2>\n
    Import the Comet library and initialize it by giving the project a name. This project will appear on your Comet home page. Most importantly, you will be requested to add an API key, which you can easily find on your account’s “Settings” page.<\/p>\n
    import<\/span> comet_ml\n\ncomet_ml.login(project_name=\"transfer_learning_training\"<\/span>)<\/span><\/pre>\n
    After pasting your API key, you will get the image below.<\/p>\n
    \n
    \n
    \"\"<\/figure>
    <\/picture><\/div>\n<\/div>\n
    Screenshot by author<\/figcaption>\n<\/figure>\n
    The next step is to allow Comet to monitor our parameters. These parameters will be logged.<\/p>\n
    from<\/span> comet_ml import<\/span> Experiment\n\n#initializing comet experiment<\/span>\nexperiment = Experiment()\n\nparameters = {\n    \"batch_size\"<\/span>:batch_size,\n    \"epochs\"<\/span>: 5<\/span>,\n    \"optimizer\"<\/span>: \"adam\"<\/span>,\n    \"loss\"<\/span>: \"binary_crossentropy\"<\/span>,\n}\n#Logging Parameters<\/span>\nexperiment.log_parameters(parameters)<\/span><\/pre>\n
    We can now create data generators for our training and validation data fed into the models. Here, we will perform the necessary transformations on the images.<\/p>\n
    from<\/span> sklearn.model_selection import<\/span> train_test_split\nfrom<\/span> keras.preprocessing.image import<\/span> ImageDataGenerator\n\n\n#Importing dataframe<\/span>\ndf = pd.read_csv(\"new_training_info.csv\"<\/span>)\n\n#performing training and validation splits<\/span>\ntrain_df, validation_df = train_test_split(df,test_size=0.2<\/span>, random_state=42<\/span>)\n\n#preparing data generators<\/span>\ntrain_datagen = ImageDataGenerator(rotation_range=15<\/span>,\n                                   rescale=1.<\/span>\/255<\/span>,\n                                   shear_range=0.1<\/span>,\n                                   zoom_range=0.2<\/span>,\n                                   horizontal_flip=True<\/span>,\n                                   width_shift_range=0.1<\/span>,\n                                   height_shift_range=0.1<\/span>)\n\n\ntrain_gen = train_datagen.flow_from_dataframe(train_df,\n                                              train_dir, x_col='filename'<\/span>, y_col='category'<\/span>,\n                                              target_size=img_size,\n                                              class_mode='binary'<\/span>,\n                                              batch_size=batch_size)\n\nvalidation_datagen = ImageDataGenerator(rescale=1.<\/span>\/255<\/span>)\n\n\nvalidation_gen = validation_datagen.flow_from_dataframe(train_df,\n                                              train_dir, x_col='filename'<\/span>, y_col='category'<\/span>,\n                                              target_size=img_size,\n                                              class_mode='binary'<\/span>,\n                                              batch_size=batch_size)<\/span><\/pre>\n
    Now that that part is complete, we can use transfer learning using a pre-trained model.<\/p>\n
    Transfer Learning<\/h2>\n
    We are going to use a ResNet50 as the base model, and we will add a few layers on top of it.<\/p>\n
    #model creation and training<\/span>\nfrom<\/span> keras.applications.resnet import<\/span> ResNet50\nfrom<\/span> keras.layers import<\/span> Dense, Flatten\nfrom<\/span> keras.models import<\/span> Sequential\nfrom<\/span> keras.callbacks import<\/span> EarlyStopping\n\n#initialize sequential model<\/span>\nmodel = Sequential()\n\n#base model<\/span>\nbase = ResNet50(weights= 'imagenet'<\/span>,\n                 include_top=False<\/span>,\n                 input_shape=(img_height, img_width, img_channels),\n                 classes=2<\/span>\n                 )\n\nfor<\/span> layer in<\/span> base.layers:\n    layer.trainable = False<\/span>\n\n#Adding layers to the model<\/span>\nmodel.add(base)\nmodel.add(Flatten())\nmodel.add(Dense(512<\/span>, activation='relu'<\/span>))\nmodel.add(Dense(1<\/span>, activation='sigmoid'<\/span>))\n\n#EarlyStopping callback<\/span>\ncallback = EarlyStopping(monitor='loss'<\/span>,\n                         patience=0<\/span>)<\/span><\/pre>\n
    Above, we define our ResNet50 model and highlight the number of classes we train (only 2). We also specify the shape of the image that we want.<\/p>\n
    We will now compile the model and train it. Training will be recorded as a live event on your “Project page” in Comet. It will be indicated as complete once we run the “experiment.end()” function provided by Comet.<\/p>\n
    Let’s first see the structure of our model:<\/p>\n
    \n
    \"\"<\/figure>
    <\/picture><\/div>\n
    Screenshot by author<\/figcaption>\n<\/figure>\n
    The model can then be compiled using the parameters we initially logged to Comet.<\/p>\n
    model.compile<\/span>(loss=parameters[\"loss\"<\/span>],\n              optimizer=parameters[\"optimizer\"<\/span>],\n              metrics=[\"accuracy\"<\/span>])<\/span><\/pre>\n
    We will do the same when training the model.<\/p>\n
    #removing validation_steps and steps_per_epoch<\/span>\nmodel.fit_generator(train_gen,\n                    validation_data = validation_gen,\n                    epochs=parameters[\"epochs\"<\/span>],\n                    callbacks=[callback])<\/span><\/pre>\n
    \n
    \n
    \"\"<\/figure>
    <\/picture><\/div>\n<\/div>\n
    Screenshot by author<\/figcaption>\n<\/figure>\n
    Training occurs for five epochs, and now we can open our Comet “Project” page and check whether all the information we want has been logged and visualized.<\/p>\n
    \n
    \n
    \"\"<\/figure>
    <\/picture><\/div>\n<\/div>\n
    Screenshot by author<\/figcaption>\n<\/figure>\n
    We can see that our visualizations have been made. The loss and accuracy charts for the training and validation sets are available.<\/p>\n
    To end the run, one should add the line below:<\/p>\n
    experiment.end()<\/span><\/pre>\n<\/div>\n<\/div>\n<\/div>\n\n\n\n
    As shown above, Comet successfully logged the transfer learning run. However, it was fairly inaccurate, as I did not use various techniques to improve model performance. Some of the techniques that could be used to do this include data augmentation and resampling to give but a few examples.<\/span><\/div>\n\n\n\n
    \n
    \n
    \n
    Till we meet again!!<\/p>\n<\/div>\n<\/div>\n<\/div>\n","protected":false},"excerpt":{"rendered":"
    Photo by Daniel Korpai on Unsplash Transfer learning involves using a pre-trained model to solve a deep-learning problem. If your problem is not necessarily unique, it is better to create a model that has already been trained on a given task. Comet extensively integrates with multiple libraries, including the most common ones for deep learning, […]<\/p>\n","protected":false},"author":79,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"customer_name":"","customer_description":"","customer_industry":"","customer_technologies":"","customer_logo":"","footnotes":""},"categories":[6,9,7],"tags":[],"coauthors":[176],"class_list":["post-9432","post","type-post","status-publish","format-standard","hentry","category-machine-learning","category-product","category-tutorials"],"yoast_head":"\nNavigating Transfer Learning with Comet<\/title>\n<meta name=\"description\" content=\"Learn how you can make great observations and visualizations with Comet, using transfer learning and ScikitLearn, Pandas, and Keras.\" \/>\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\/transfer-learning-with-comet\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Navigating Transfer Learning with Comet\" \/>\n<meta property=\"og:description\" content=\"Learn how you can make great observations and visualizations with Comet, using transfer learning and ScikitLearn, Pandas, and Keras.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.comet.com\/site\/blog\/transfer-learning-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=\"2024-03-12T15:44:42+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2025-04-24T17:02:58+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/miro.medium.com\/v2\/resize:fit:700\/0*3SgLgEnaaG9dFM96\" \/>\n<meta name=\"author\" content=\"Mwanikii Njagi\" \/>\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=\"Mwanikii Njagi\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"6 minutes\" \/>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"Navigating Transfer Learning with Comet","description":"Learn how you can make great observations and visualizations with Comet, using transfer learning and ScikitLearn, Pandas, and Keras.","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\/transfer-learning-with-comet","og_locale":"en_US","og_type":"article","og_title":"Navigating Transfer Learning with Comet","og_description":"Learn how you can make great observations and visualizations with Comet, using transfer learning and ScikitLearn, Pandas, and Keras.","og_url":"https:\/\/www.comet.com\/site\/blog\/transfer-learning-with-comet","og_site_name":"Comet","article_publisher":"https:\/\/www.facebook.com\/cometdotml","article_published_time":"2024-03-12T15:44:42+00:00","article_modified_time":"2025-04-24T17:02:58+00:00","og_image":[{"url":"https:\/\/miro.medium.com\/v2\/resize:fit:700\/0*3SgLgEnaaG9dFM96","type":"","width":"","height":""}],"author":"Mwanikii Njagi","twitter_card":"summary_large_image","twitter_creator":"@Cometml","twitter_site":"@Cometml","twitter_misc":{"Written by":"Mwanikii Njagi","Est. reading time":"6 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/www.comet.com\/site\/blog\/transfer-learning-with-comet#article","isPartOf":{"@id":"https:\/\/www.comet.com\/site\/blog\/transfer-learning-with-comet\/"},"author":{"name":"Mwanikii Njagi","@id":"https:\/\/www.comet.com\/site\/#\/schema\/person\/c7043b3e6b992af7b3220aa1f27d2162"},"headline":"Navigating Transfer Learning with Comet","datePublished":"2024-03-12T15:44:42+00:00","dateModified":"2025-04-24T17:02:58+00:00","mainEntityOfPage":{"@id":"https:\/\/www.comet.com\/site\/blog\/transfer-learning-with-comet\/"},"wordCount":805,"publisher":{"@id":"https:\/\/www.comet.com\/site\/#organization"},"image":{"@id":"https:\/\/www.comet.com\/site\/blog\/transfer-learning-with-comet#primaryimage"},"thumbnailUrl":"https:\/\/miro.medium.com\/v2\/resize:fit:700\/0*3SgLgEnaaG9dFM96","articleSection":["Machine Learning","Product","Tutorials"],"inLanguage":"en-US"},{"@type":"WebPage","@id":"https:\/\/www.comet.com\/site\/blog\/transfer-learning-with-comet\/","url":"https:\/\/www.comet.com\/site\/blog\/transfer-learning-with-comet","name":"Navigating Transfer Learning with Comet","isPartOf":{"@id":"https:\/\/www.comet.com\/site\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.comet.com\/site\/blog\/transfer-learning-with-comet#primaryimage"},"image":{"@id":"https:\/\/www.comet.com\/site\/blog\/transfer-learning-with-comet#primaryimage"},"thumbnailUrl":"https:\/\/miro.medium.com\/v2\/resize:fit:700\/0*3SgLgEnaaG9dFM96","datePublished":"2024-03-12T15:44:42+00:00","dateModified":"2025-04-24T17:02:58+00:00","description":"Learn how you can make great observations and visualizations with Comet, using transfer learning and ScikitLearn, Pandas, and Keras.","breadcrumb":{"@id":"https:\/\/www.comet.com\/site\/blog\/transfer-learning-with-comet#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.comet.com\/site\/blog\/transfer-learning-with-comet"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.comet.com\/site\/blog\/transfer-learning-with-comet#primaryimage","url":"https:\/\/miro.medium.com\/v2\/resize:fit:700\/0*3SgLgEnaaG9dFM96","contentUrl":"https:\/\/miro.medium.com\/v2\/resize:fit:700\/0*3SgLgEnaaG9dFM96"},{"@type":"BreadcrumbList","@id":"https:\/\/www.comet.com\/site\/blog\/transfer-learning-with-comet#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/www.comet.com\/site\/"},{"@type":"ListItem","position":2,"name":"Navigating Transfer Learning 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\/c7043b3e6b992af7b3220aa1f27d2162","name":"Mwanikii Njagi","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.comet.com\/site\/#\/schema\/person\/image\/1a3c516cf04aca9418dfb2213081f4df","url":"https:\/\/www.comet.com\/site\/wp-content\/uploads\/2023\/08\/cropped-1_2jy9gyk0G_yaniWm8gJFVA-1-96x96.webp","contentUrl":"https:\/\/www.comet.com\/site\/wp-content\/uploads\/2023\/08\/cropped-1_2jy9gyk0G_yaniWm8gJFVA-1-96x96.webp","caption":"Mwanikii Njagi"},"url":"https:\/\/www.comet.com\/site\/blog\/author\/freddynjagigmail-com\/"}]}},"_links":{"self":[{"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/posts\/9432","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\/79"}],"replies":[{"embeddable":true,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/comments?post=9432"}],"version-history":[{"count":1,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/posts\/9432\/revisions"}],"predecessor-version":[{"id":15374,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/posts\/9432\/revisions\/15374"}],"wp:attachment":[{"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/media?parent=9432"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/categories?post=9432"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/tags?post=9432"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/coauthors?post=9432"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}