Skip to main content

Project description


Documentation is available at


Extracting topics from reflective medical writings.


MedTop is only compatible with 64-bit python. You can check which version of python you're using in your virtual environment with the following code.

import platform; platform.architecture()[0];

pip install medtop

How to use

A template pipeline is provided below using a test dataset. You can read more about the test_data dataset here

Each step of the pipeline has configuration options for experimenting with various methods. These are detailed in the documentation for each method. Notably, the import_docs, get_cluster_topics, visualize_clustering, and evaluate methods all include the option to save results to a file.

Example Pipeline

Import data

Import and pre-process documents from a text file containing a list of all documents.

from medtop.core import *
data, doc_df = import_from_files('test_data/corpus_file_list.txt', stop_words_file='stop_words.txt', save_results = False)

You can also consolidate your documents into a single, pipe-delimited csv file with the columns "doc_name" and "text".

data, doc_df = import_from_csv('test_data/corpus.txt', stop_words_file='stop_words.txt', save_results = False)

Transform data

Create word vectors from the most expressive phrase in each sentence of the imported documents. Seed documents can be passed as a single CSV similar to corpus documents in the import step.

NOTE: If doc_df is NOT passed to create_tfidf, you must set include_input_in_tfidf=False in get_phrases.

tfidf, dictionary = create_tfidf(doc_df, path_to_seed_topics_file_list='test_data/seed_topics_file_list.txt')
data = get_phrases(data, dictionary.token2id, tfidf, include_input_in_tfidf = True, include_sentiment=True)
data = get_vectors("tfidf", data, dictionary = dictionary, tfidf = tfidf)
Removed 67 sentences without phrases.

Cluster data

Cluster the sentences into groups expressing similar ideas or topics. If you aren't sure how many true clusters exist in the data, try running assign_clusters with the optional parameter show_chart = True to visual cluster quality with varying numbers of clusters. When using method='hac', you can also use show_dendrogram = True see the cluster dendrogram.

data = assign_clusters(data, method = "hac")
cluster_df = get_cluster_topics(data, doc_df, save_results = False)
visualize_clustering(data, method = "umap", show_chart = False)

Evaluate results

gold_file = "test_data/gold.txt"
results_df = evaluate(data, gold_file="test_data/gold.txt", save_results = False)

Document Clustering

IMPORTANT: This feature is still in alpha, meaning that we have adapted the pipeline to accomodate the clustering of documents, but have made no rigorous efforts the ensure that it works well.

To cluster documents, simply import data and create the TF-IDF as above, but extract phrase, create the vectors, and cluster using the doc_df dataframe. Passing the parameter window_size=-1 to get_phrases tells the method to use all tokens instead of selecting a subset of length window_size.

doc_df = get_phrases(doc_df, dictionary.token2id, tfidf, include_input_in_tfidf = True, window_size=-1)
doc_df = get_vectors("tfidf", doc_df, dictionary = dictionary, tfidf = tfidf)
doc_df = assign_clusters(doc_df, method = "kmeans", k=4)
cluster_df = get_cluster_topics(data, doc_df, save_results = False)
visualize_clustering(data, method = "svd", show_chart = False)
Removed 0 sentences without phrases.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for medtop, version 0.0.8
Filename, size File type Python version Upload date Hashes
Filename, size medtop-0.0.8.tar.gz (20.0 kB) File type Source Python version None Upload date Hashes View
Filename, size medtop-0.0.8-py3-none-any.whl (18.9 kB) File type Wheel Python version py3 Upload date Hashes View

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page