kitsamho

def text_clustering_loop(df_text_embeddings, tokeniser, model):
    """
    Upon initialisation, pre-computed embeddings for c.1500 images from the conceptual captions dataset are loaded. These are already labelled as one of 7 domains: property, sports, food, musicians, vehicles, illustration and nature
    Users are prompted to type a text input related to one of the domains
    The app uses CLIP to generate image embeddings for this user-inputted text
    The Pre-computed conceptual captions are concatenated with this new user embedding
    The UMAP algorithm is used for dimensionality reduction across the concatenated embedding space
    The resulting markers/clusters are displayed as a scatter plot
    Each marker in the scatter plot represents a text string where similar texts are grouped together

kitsamho

def image_clustering_loop(df_image_embeddings, processor, model):
    """
    Users can input a URL to an image (e.g. their LinkedIn photo or any link that links directly to an image (e.g. jpeg, .png)
    The app uses CLIP to generate image embeddings for this user-inputted image
    Pre-computed rock archive images are concatenated with this new user embedding
    The UMAP algorithm is used for dimensionality reduction across the concatenated embedding space
    The resulting markers/clusters are displayed as a scatter plot
    Each marker in the scatter plot represents an image where similar images are grouped together

    Args:

kitsamho

def text_classification_loop(bbc_headlines, tokeniser, model):
    """
    Displays a random bbc headline from a list of scraped headlines, prompts the user to enter some contrasting labels
    for the headline, and uses the pre-trained model to predict the probabilities of the provided labels.
    Returns the predicted probabilities in a dataframe that get visualised as a bar plot
    
    Args:
        bbc_headlines (List[str]): A list of BBC headlines to use for classification.
        model: A pre-trained CLIP model.
        tokenizer: A tokenizer for the CLIP model.

kitsamho

from transformers import CLIPModel, AutoProcessor, AutoTokenizer
import streamlit as st


@st.cache_resource
def download_clip_model(clip_model='openai/clip-vit-base-patch32'):
    """
    Load the CLIP model and its associated tokenizer and processor from a given pre-trained model.

    Args:

kitsamho

def classify_images(text_inputs: list, images: list, processor, model, tokeniser):
    """
    Calculates the similarity between the text inputs and images.

    Args:
        text_inputs (list): List of text inputs.
        images (list): List of images urls
        processor (transformers.AutoProcessor): The CLIP processor to use.
        model (transformers.CLIPModel): The CLIP model to use.
        tokenizer (transformers.AutoTokenizer): The CLIP tokenizer to use.

kitsamho

a = st.beta_expander('About (click to expand)')

a.write("This is an animated plot where each step is a week. If you want to explore the animation using other features \
    wait until the initial animation has ended or skip through to the end before changing features otherwise you may \
    see some odd behaviour in the animation. Plotly can be a little fickle like that.")

a.write("This is a Plotly chart so you can click on the legend to mask values if needed.")

# set up some columns for the interactive widgets - use a mid point to create some buffer between widgets
c1, c2 = st.beta_columns((2, 3))

kitsamho

a = st.beta_expander('About (click to expand)')

a.write("The crossplot analysis allows you to plot two features against one another with the marker sizes representing \
what we might consider as dependant features e.g. total deaths, total deaths per million, total vaccinations.")
a.write("DataFrame masking allows you to explore the data by continent or include all countries.")
a.write("Each feature's central tendency is represented by the dashed line on each axis so you can see where \
countries are positioned in terms of the distribution for each feature. These update when you change features.")
a.write('The heat map is a summary of the correlations between all features')

# DataFrame set up

kitsamho

def get_indexes(x):
    
    # identifies indexes where no data points exist yet and we will want to fill with a value other than null
    index_fill_1 = [i for i in range(x.index[0], x.dropna().index[0])]
    # identifies indexes where there is data however some missing points exist and we will want to apply interpolation
    index_interpolate = [i for i in range(x.dropna().index[0], x.index[-1])]
    return index_fill_1, index_interpolate


def update_series(x):

kitsamho

def get_data(df,transform_cols):
    """ This is the main function that transforms the raw OWID data into something we can use in the app
    Args:
        Original DataFrame from csv
    Returns:
        Processed / cleaned DataFrame
    """
    # loop through and subset each country to a list
    country_dfs = []

kitsamho

import pandas as pd
import numpy as np
import plotly.graph_objs as go
from sklearn.decomposition import PCA

def musixmatch_scatplot(df,df_embed,n_components=2):
    
    """Comprehensive function that adds components to a scatter plot. Requires musixmatch_PCA function """
    
    df = musixmatch_PCA(df,df_embed,n_components) #gets PCA