Spaces:
Running
Running
| import streamlit as st | |
| import numpy as np | |
| import matplotlib.pyplot as plt | |
| from sklearn.decomposition import PCA | |
| from adjustText import adjust_text | |
| st.subheader('Most Similar Terms (Visual)') | |
| st.write("This app shows how a given input term changes its position in the embeddings space over time relative to its nearest neighbors.") | |
| keyword = st.text_input("Input term", "gay", key="semkey") | |
| keyword = keyword.lower() | |
| def semchange(keyword): | |
| if keyword not in st.session_state['models_all'][1810]: | |
| st.write('Keyword not found in data. Please check for spelling errors.') | |
| return | |
| # get list of all similar words from different periods | |
| sim_words = [] | |
| for year, model in st.session_state['models_all'].items(): | |
| if year in range(1810, 2000, 60): | |
| if model[keyword].all() == st.session_state['models_all'][1810]['biology'].all(): | |
| st.write('Keyword not available for ', year) | |
| if model[keyword].all() != st.session_state['models_all'][1810]['biology'].all(): | |
| tempsim = model.most_similar(keyword, topn=7) | |
| for term, vector in tempsim: | |
| sim_words.append(term) | |
| sim_words = list(set(sim_words)) | |
| # get vectors of similar words in most recent embedding (1990) | |
| sim_vectors1990 = np.array([st.session_state['models_all'][1990][w] for w in sim_words]) | |
| # get vectors of keyword in all periods | |
| keyword_vectors = np.zeros(shape=(0,300)) | |
| for year, model in st.session_state['models_all'].items(): | |
| if year in range(1810, 2000, 60): | |
| if model[keyword].all() != st.session_state['models_all'][1810]['biology'].all(): | |
| temp_keyword_vector = np.array([model[keyword]]) | |
| keyword_vectors = np.append(keyword_vectors, temp_keyword_vector, axis=0) | |
| # add keyword vectors from all periods to vectors of similar words 1990 | |
| allvectors = np.append(sim_vectors1990, keyword_vectors, axis=0) | |
| # "train" PCA model with only similar words | |
| pca = PCA(n_components=2) | |
| pca.fit(sim_vectors1990) | |
| two_dim = pca.transform(allvectors) | |
| # get labels | |
| labels = sim_words | |
| for year, model in st.session_state['models_all'].items(): | |
| if year in range(1810, 2000, 60): | |
| if model[keyword].all() != st.session_state['models_all'][1810]['biology'].all(): | |
| labels.append(keyword + str(year)) | |
| #plot results | |
| fig, ax = plt.subplots() | |
| ax.scatter(two_dim[:, 0], two_dim[:, 1]) | |
| texts = [ax.text(x=two_dim[i, 0], y=two_dim[i, 1], s=labels[i]) for i in range(len(sim_words))] | |
| adjust_text(texts) | |
| #plot arrow between keywords | |
| for i in range(-2, -(len(keyword_vectors)+1), -1): | |
| ax.arrow(two_dim[i,0], two_dim[i,1], | |
| two_dim[i+1, 0] - two_dim[i,0], two_dim[i+1, 1] - two_dim[i,1], | |
| head_width=0.03, length_includes_head=True) | |
| st.pyplot(fig) | |
| fig.clear() | |
| plt.close(fig) | |
| semchange(keyword) | |