Tamara Adokeme
Initial classifier config
09ff543
raw
history blame
11.7 kB
############ 1. IMPORTING LIBRARIES ############
# Import streamlit, requests for API calls, and pandas and numpy for data manipulation
import streamlit as st
import requests
import pandas as pd
import numpy as np
from streamlit_tags import st_tags # to add labels on the fly!
############ 2. SETTING UP THE PAGE LAYOUT AND TITLE ############
# `st.set_page_config` is used to display the default layout width, the title of the app, and the emoticon in the browser tab.
st.set_page_config(
layout="centered", page_title="Zero-Shot Text Classifier", page_icon="❄️"
)
############ CREATE THE LOGO AND HEADING ############
# We create a set of columns to display the logo and the heading next to each other.
c1, c2 = st.columns([0.32, 2])
# The snowflake logo will be displayed in the first column, on the left.
with c1:
st.image(
"https://images.unsplash.com/photo-1508175800969-525c72a047dd?w=500&auto=format&fit=crop&q=60&ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxzZWFyY2h8MTl8fGFmcm8lMjByb2JvdHxlbnwwfHwwfHx8MA%3D%3D",
width=85,
)
# The heading will be on the right.
with c2:
st.caption("")
st.title("Zero-Shot Text Classifier")
# We need to set up session state via st.session_state so that app interactions don't reset the app.
if not "valid_inputs_received" in st.session_state:
st.session_state["valid_inputs_received"] = False
############ SIDEBAR CONTENT ############
st.sidebar.write("")
# For elements to be displayed in the sidebar, we need to add the sidebar element in the widget.
# We create a text input field for users to enter their API key.
API_KEY = st.sidebar.text_input(
"Enter your HuggingFace API key",
help="Once you created you HuggingFace account, you can get your free API token in your settings page: https://huggingface.co/settings/tokens",
type="password",
)
# Adding the HuggingFace API inference URL.
API_URL = "https://api-inference.huggingface.co/models/valhalla/distilbart-mnli-12-3"
# Now, let's create a Python dictionary to store the API headers.
headers = {"Authorization": f"Bearer {API_KEY}"}
st.sidebar.markdown("---")
# Let's add some info about the app to the sidebar.
st.sidebar.write(
"""
App created by [Charly Wargnier](https://twitter.com/DataChaz) using [Streamlit](https://streamlit.io/)🎈 and [HuggingFace](https://huggingface.co/inference-api)'s [Distilbart-mnli-12-3](https://huggingface.co/valhalla/distilbart-mnli-12-3) model.
"""
)
############ TABBED NAVIGATION ############
# First, we're going to create a tabbed navigation for the app via st.tabs()
# tabInfo displays info about the app.
# tabMain displays the main app.
MainTab, InfoTab = st.tabs(["Main", "Info"])
with InfoTab:
st.subheader("What is Streamlit?")
st.markdown(
"[Streamlit](https://streamlit.io) is a Python library that allows the creation of interactive, data-driven web applications in Python."
)
st.subheader("Resources")
st.markdown(
"""
- [Streamlit Documentation](https://docs.streamlit.io/)
- [Cheat sheet](https://docs.streamlit.io/library/cheatsheet)
- [Book](https://www.amazon.com/dp/180056550X) (Getting Started with Streamlit for Data Science)
"""
)
st.subheader("Deploy")
st.markdown(
"You can quickly deploy Streamlit apps using [Streamlit Community Cloud](https://streamlit.io/cloud) in just a few clicks."
)
with MainTab:
# Then, we create a intro text for the app, which we wrap in a st.markdown() widget.
st.write("")
st.markdown(
"""
Classify keyphrases on the fly with this mighty app. No training needed!
"""
)
st.write("")
# Now, we create a form via `st.form` to collect the user inputs.
# All widget values will be sent to Streamlit in batch.
# It makes the app faster!
with st.form(key="my_form"):
############ ST TAGS ############
# We initialize the st_tags component with default "labels"
# Here, we want to classify the text into one of the following user intents:
# Transactional
# Informational
# Navigational
labels_from_st_tags = st_tags(
value=["Transactional", "Informational", "Navigational"],
maxtags=3,
suggestions=["Transactional", "Informational", "Navigational"],
label="",
)
# The block of code below is to display some text samples to classify.
# This can of course be replaced with your own text samples.
# MAX_KEY_PHRASES is a variable that controls the number of phrases that can be pasted:
# The default in this app is 50 phrases. This can be changed to any number you like.
MAX_KEY_PHRASES = 50
new_line = "\n"
pre_defined_keyphrases = [
"I want to buy something",
"We have a question about a product",
"I want a refund through the Google Play store",
"Can I have a discount, please",
"Can I have the link to the product page?",
]
# Python list comprehension to create a string from the list of keyphrases.
keyphrases_string = f"{new_line.join(map(str, pre_defined_keyphrases))}"
# The block of code below displays a text area
# So users can paste their phrases to classify
text = st.text_area(
# Instructions
"Enter keyphrases to classify",
# 'sample' variable that contains our keyphrases.
keyphrases_string,
# The height
height=200,
# The tooltip displayed when the user hovers over the text area.
help="At least two keyphrases for the classifier to work, one per line, "
+ str(MAX_KEY_PHRASES)
+ " keyphrases max in 'unlocked mode'. You can tweak 'MAX_KEY_PHRASES' in the code to change this",
key="1",
)
# The block of code below:
# 1. Converts the data st.text_area into a Python list.
# 2. It also removes duplicates and empty lines.
# 3. Raises an error if the user has entered more lines than in MAX_KEY_PHRASES.
text = text.split("\n") # Converts the pasted text to a Python list
linesList = [] # Creates an empty list
for x in text:
linesList.append(x) # Adds each line to the list
linesList = list(dict.fromkeys(linesList)) # Removes dupes
linesList = list(filter(None, linesList)) # Removes empty lines
if len(linesList) > MAX_KEY_PHRASES:
st.info(
f"❄️ Note that only the first "
+ str(MAX_KEY_PHRASES)
+ " keyphrases will be reviewed to preserve performance. Fork the repo and tweak 'MAX_KEY_PHRASES' in the code to increase that limit."
)
linesList = linesList[:MAX_KEY_PHRASES]
submit_button = st.form_submit_button(label="Submit")
############ CONDITIONAL STATEMENTS ############
# Now, let us add conditional statements to check if users have entered valid inputs.
# E.g. If the user has pressed the 'submit button without text, without labels, and with only one label etc.
# The app will display a warning message.
if not submit_button and not st.session_state.valid_inputs_received:
st.stop()
elif submit_button and not text:
st.warning("❄️ There is no keyphrases to classify")
st.session_state.valid_inputs_received = False
st.stop()
elif submit_button and not labels_from_st_tags:
st.warning("❄️ You have not added any labels, please add some! ")
st.session_state.valid_inputs_received = False
st.stop()
elif submit_button and len(labels_from_st_tags) == 1:
st.warning("❄️ Please make sure to add at least two labels for classification")
st.session_state.valid_inputs_received = False
st.stop()
elif submit_button or st.session_state.valid_inputs_received:
if submit_button:
# The block of code below if for our session state.
# This is used to store the user's inputs so that they can be used later in the app.
st.session_state.valid_inputs_received = True
############ MAKING THE API CALL ############
# First, we create a Python function to construct the API call.
def query(payload):
response = requests.post(API_URL, headers=headers, json=payload)
return response.json()
# The function will send an HTTP POST request to the API endpoint.
# This function has one argument: the payload
# The payload is the data we want to send to HugggingFace when we make an API request
# We create a list to store the outputs of the API call
list_for_api_output = []
# We create a 'for loop' that iterates through each keyphrase
# An API call will be made every time, for each keyphrase
# The payload is composed of:
# 1. the keyphrase
# 2. the labels
# 3. the 'wait_for_model' parameter set to "True", to avoid timeouts!
for row in linesList:
api_json_output = query(
{
"inputs": row,
"parameters": {"candidate_labels": labels_from_st_tags},
"options": {"wait_for_model": True},
}
)
# Let's have a look at the output of the API call
# st.write(api_json_output)
# All the results are appended to the empty list we created earlier
list_for_api_output.append(api_json_output)
# then we'll convert the list to a dataframe
df = pd.DataFrame.from_dict(list_for_api_output)
st.success("βœ… Done!")
st.caption("")
st.markdown("### Check the results!")
st.caption("")
# st.write(df)
############ DATA WRANGLING ON THE RESULTS ############
# Various data wrangling to get the data in the right format!
# List comprehension to convert the score from decimals to percentages
f = [[f"{x:.2%}" for x in row] for row in df["scores"]]
# Join the classification scores to the dataframe
df["classification scores"] = f
# Rename the column 'sequence' to 'keyphrase'
df.rename(columns={"sequence": "keyphrase"}, inplace=True)
# The API returns a list of all labels sorted by score. We only want the top label.
# For that, we need to select the first element in the 'labels' and 'classification scores' lists
df["label"] = df["labels"].str[0]
df["accuracy"] = df["classification scores"].str[0]
# Drop the columns we don't need
df.drop(["scores", "labels", "classification scores"], inplace=True, axis=1)
# st.write(df)
# We need to change the index. Index starts at 0, so we make it start at 1
df.index = np.arange(1, len(df) + 1)
# Display the dataframe
st.write(df)
cs, c1 = st.columns([2, 2])
# The code below is for the download button
# Cache the conversion to prevent computation on every rerun
with cs:
@st.experimental_memo
def convert_df(df):
return df.to_csv().encode("utf-8")
csv = convert_df(df)
st.caption("")
st.download_button(
label="Download results",
data=csv,
file_name="classification_results.csv",
mime="text/csv",
)