v0.2.1

• Fixed issue when loading in more than 40.000 images
• Fixed transform only working with pre-trained embeddings

v0.2.0

Added c-TF-IDF as an algorithm to extract textual representations from images.

from concept import ConceptModel

concept_model = ConceptModel(ctfidf=True)
concepts = concept_model.fit_transform(img_names, docs=docs)

From the textual and visual embeddings, we use cosine similarity to find the best matching words
for each image. Then, after clustering the images, we combine all words in a cluster into a single
documents. Finally, c-TF-IDF is used to find the best words for each concept cluster.

The benefit of this method is that it takes the entire cluster structure into account when creating the
representations. This is not the case when we only consider words close to the concept embedding.

v0.1.1

• Fix RAM issues
• Update documentation
• Add ftfy dependency
• Fix .visualize_concepts
• Added .search_concepts

v0.1.0

• Update Readme with a small example
• Create documentation page: https://maartengr.github.io/Concept/
• Fix fit not working properly
• Better visualization of resulting concepts