{"id":8084,"date":"2023-11-02T10:06:06","date_gmt":"2023-11-02T18:06:06","guid":{"rendered":"https:\/\/live-cometml.pantheonsite.io\/?p=8084"},"modified":"2025-04-24T17:04:48","modified_gmt":"2025-04-24T17:04:48","slug":"defining-marketing-strategy-using-comet","status":"publish","type":"post","link":"https:\/\/www.comet.com\/site\/blog\/defining-marketing-strategy-using-comet\/","title":{"rendered":"Defining Marketing Strategy Using Comet"},"content":{"rendered":"\n
\"\"\/
Photo by Campaign Creators<\/a> on Unsplash<\/a><\/figcaption><\/figure>\n\n\n\n

Without solid marketing efforts companies will have a hard time growing and sustaining their business.<\/p>\n\n\n\n

The marketing department of any organization is crucial for building the company\u2019s brand and engaging customers with relevant content with the intention of increasing sales and revenue. But in order to serve their customer base best, the marketing team needs to understand them \u2014 what is it that their customers want, what do they need?<\/p>\n\n\n\n

If your company is able to determine this, then you can launch a targeted marketing campaign that aims to educate and engage the customer with content that is specific and tailored to their needs. If data about your customers is available, then data science can be applied to perform market segmentation.<\/p>\n\n\n\n

Imagine that you\u2019ve been hired as a consultant for a credit card company. One of the objectives for the marketing team this quarter is to launch a targeted ad campaign. In order for marketers to launch a targeted marketing campaign, they\u2019ll need to learn about their customers: what are their spending habits, what patterns of credit usage present themselves, and so on. Over the years, your company has collected a lot of valuable data about how their customers use their credit card products.<\/p>\n\n\n\n

The marketing team wants to launch a campaign and, in order to target customers appropriately, the team wants to divide the customers into three to five distinctive segments.<\/p>\n\n\n\n

In this blog, we\u2019ll do the following activities:<\/p>\n\n\n\n

\u2022 Profile our data set<\/p>\n\n\n\n

\u2022 Do some light data cleaning<\/p>\n\n\n\n

\u2022 Find initial segments using a clustering technique known as k-means<\/p>\n\n\n\n

\u2022 Use PCA for dimensionality reduction and use k-means to find clusters on the reduced dataset.<\/p>\n\n\n\n

\u2022 Experiment with autoencoders for dimensionality reduction and use k-means to find clusters on the reduced dataset.<\/p>\n\n\n\n

We\u2019ll be tracking our experiments and logging any artifacts using Comet<\/a>.<\/p>\n\n\n\n

The dataset we\u2019re using for this post comes from the Kaggle Credit Card Dataset for Clustering, which you can download here<\/a>.<\/p>\n\n\n\n

\n

Go ahead and open up a fresh Jupyter notebook, adjust your windows sizes according to your preference, and follow along with me.<\/p>\n<\/blockquote>\n\n\n\n

Profiling our dataset<\/h1>\n\n\n\n

The dataset we\u2019re working with summarizes credit card usage behavior for roughly 9000 active credit card holders over the previous six months. We\u2019ve got 18 behavioral features at our disposal, which are described in the table below.<\/p>\n\n\n\n

\"\"\/
Description of features in raw data<\/figcaption><\/figure>\n\n\n\n

Let\u2019s begin by downloading the dataset. Run the code below to download it to your workspace.<\/p>\n\n\n\n

\n
\n
\n
%pip install gdown --upgrade --quiet\nimport gdown\ndef download_from_gdrive(gid, output):\n\"\"\"Download csv file from Google.\nArgs:\ngid (str): Google Drive's file ID.\noutput (str): Output filename.\n\"\"\"\ngdown.download(id = gid, output = output)\ndownload_from_gdrive(gid = \"1LIIW7rcLExMbsC5RFG87laIL_Nn_yy8Q\", output = 'cc_data.csv')<\/pre>\n<\/div>\n<\/div>\n
Download raw data from Google drive<\/figcaption>\n<\/figure>\n\n\n\n
We\u2019ll also install Comet and import it:<\/p>\n\n\n\n
\n
\n
\n
%pip install comet_ml --quiet\nimport comet_ml\nfrom comet_ml import Experiment, Artifact<\/pre>\n<\/div>\n<\/div>\n
Importing Comet<\/figcaption>\n<\/figure>\n\n\n\n
We\u2019ll also import all the other libraries that will help us in this project:<\/p>\n\n\n\n
\n
\n
\n
%pip install sweetviz --quiet\nimport sweetviz\nimport pandas as pd\nimport numpy as np\nimport seaborn as sns\nimport pickle\nimport matplotlib.pyplot as plt\nfrom numpy import save\nfrom sklearn.preprocessing import StandardScaler, normalize\nfrom sklearn.cluster import KMeans\nfrom sklearn.decomposition import PCA\nfrom sklearn.metrics import silhouette_score\nfrom tensorflow.keras.layers import Input, Add, Dense, Activation\nfrom tensorflow.keras.models import Model, load_model\nfrom tensorflow.keras.initializers import glorot_uniform, lecun_normal\nfrom tensorflow.keras.models import Sequential\nfrom tensorflow.keras.optimizers import SGD<\/pre>\n<\/div>\n<\/div>\n
Importing other useful libraries<\/figcaption>\n<\/figure>\n\n\n\n
We can now load in our dataset and inspect it:<\/p>\n\n\n\n
\n
\n
\n
cc_df = pd.read_csv('cc_data.csv')\ncc_df.head()\ncc_df.info()<\/pre>\n<\/div>\n<\/div>\n
Inspecting data<\/figcaption>\n<\/figure>\n\n\n\n
We\u2019ll now initialize an experiment in Comet. Note that you\u2019ll need to pass your API from Comet<\/a> and change the workspace name to whatever yours is.<\/p>\n\n\n\n
Typically, your workspace name is the same as the username you registered with.<\/p>\n\n\n\n
\n
You\u2019ll need a Comet account to follow along. It\u2019s totally free, and easy to sign up<\/a>.<\/p>\n<\/blockquote>\n\n\n\n
Once you\u2019re up and running there, you\u2019re going to need your API key. To obtain your API key, navigate to your Comet.ml<\/a> dashboard. In the top right corner click on your username and select Settings<\/strong> from the dropdown menu. In the Settings<\/strong> page, scroll down to the Developer Information<\/strong> section and click \u201cGenerate API Key.\u201d<\/p>\n\n\n\n
After running the following line of code, you\u2019ll be prompted to enter your API key.<\/p>\n\n\n\n
\n
\n
\n
comet_ml.login()\nexperiment = Experiment(workspace='team-comet-ml', project_name='cc-clustering')<\/pre>\n<\/div>\n<\/div>\n
Initializing Comet Experiment<\/figcaption>\n<\/figure>\n\n\n\n
We\u2019ll now use sweetviz<\/code> to perform some high level EDA and profiling of our dataset.<\/p>\n\n\n\n
\n
\n
\n
experiment.add_tag(\"sweetviz\")\nexperiment.set_name('profiling-data')\nreport = sweetviz.analyze(cc_df, pairwise_analysis = 'on')\nreport.log_comet(experiment)\nexperiment.end()<\/pre>\n<\/div>\n<\/div>\n
Comet and sweetviz for<\/figcaption>\n<\/figure>\n\n\n\n
By following the link in the <\/a>url<\/a><\/code> section <\/a>and navigating to HTML<\/code> on the bottom left in the panel, we can see that our profiling report was automatically saved to Comet. This allows us to easily share with our colleagues without them having to run this code themselves.<\/p>\n\n\n\n
We can also view the report right here in our notebook by running the following line of code or following this link<\/a>.<\/p>\n\n\n\n
\n
\n
\n
report.show_notebook(w=900, h=500, scale=0.8)<\/pre>\n<\/div>\n<\/div>\n
You can show the EDA report right in your notebook by running the above line of code<\/figcaption>\n<\/figure>\n\n\n\n
Some high level observations we can glean from the profiling report:<\/p>\n\n\n\n