pixl | Deepnote

Installation

# installing pixl !pip install pixl

Imports

from pixl.utils import Pixl from sklearn.cluster import KMeans import pandas as pd from shutil import unpack_archive import os

Load in sample imagery

Run the code below to download a zip file of 3 sample dog and 3 sample cat images and unzip them into a newly created 'image' directory. We'll use unpack_archive method from shutil to unzip the images and load the filenames into a variable called filenames.

# setting directory of sample images category = 'animals' img_dir = './assets/' + category # grabbing file_names filenames = [] for file in os.listdir(img_dir): filenames.append(os.fsdecode(file))

filenames

Get vectors

# instantiate a pixl object p_object = Pixl()

def get_vec(img_dir, filenames): """ Returns a list of vectors and a list of labels. Args: filenames (str): string specifying the name of each file. Returns: vectors (lst): list of vectors representing image files. labels (lst): list of labels taken from file names. """ vectors = [] labels = [] for file in filenames: vec = p_object.get_vec(img_dir + '/' + file) vectors.append(vec) label = file.split('.')[0] labels.append(label) return labels, vectors

labels, vectors = get_vec(img_dir, filenames)

# examining a label and vector print(labels[0], vectors[0])

Inspecting vectors

ids = range(len(vectors)) img_embedd_df = pd.DataFrame(index = ids)

img_embedd_df['Labels'] = labels img_embedd_df['Vectors'] = vectors

print(img_embedd_df)

KMeans Clustering

We can do a kmeans clustering fit to our newly generated vectors to see if we can easily seperate the two classes using only the information retained in each vector.

# setting n_clusters to 2 because we know we only want to cluster 2 classes km = KMeans(n_clusters=2) # fitting to the list of vectors from get_vec func km.fit(vectors) # grabbing a list of clusters clusters = km.labels_.tolist() # column cluter for each cluster img_embedd_df["Cluster"] = clusters # displaying counts per cluster print(img_embedd_df['Cluster'].value_counts())

# displaying our df to see how we did! print(img_embedd_df)

Installation

Imports

Load in sample imagery

Get vectors

Inspecting vectors

KMeans Clustering

Recommended on Deepnote

Stock market analysis

10 ways to create NumPy arrays

Wide Residual Networks