pigieligi's picture
Update app.py
c13093b
Raw History Blame Contribute Delete
5.96 kB
import gradio as gr
import os
import numpy as np
import datetime as dt
import hopsworks
import pandas as pd
import utils
import random
from PIL import Image
from io import BytesIO
import requests
import pickle
MAX_MOVIES = 3
def fetch_new_predictions():
project = hopsworks.login()
fs = project.get_feature_store(name='igijul63_featurestore')
fg = fs.get_feature_group('movie_predictions', version=1)
query = fg.select_all()
new_predictions_df = query.read()
hopsworks.logout()
ids = new_predictions_df['id'].tolist()
print(f"Predicted movies are {ids}")
details = []
titles = []
for i, movie_id in enumerate(ids):
movie_details = utils.get_movie_details(movie_id)
if movie_details['budget'] > 0 :
# movie_details['predicted_revenue'] = new_predictions[new_predictions['id'] == movie_id]['revenue'].values[0]
movie_details['predicted_revenue'] = new_predictions_df.iloc[i,:]['revenue']
# movie_details['release_date'] = dt.datetime.strptime(movie_details['release_date'], '%Y-%m-%d').date()
details.append(movie_details)
details = pd.DataFrame(details)
details.sort_values(by = 'release_date', ascending = False, inplace = True)
# details = details[:MAX_MOVIES]
idxs = random.sample(range(len(details)), 3)
details = details.iloc[idxs, :]
new_titles = [f"### {title}" for title in details['title']]
new_revenues = [f"#### Predicted revenue is {int(revenue)}\$" for revenue in details['predicted_revenue']]
get_img = lambda poster_path : Image.open(BytesIO(requests.get(f"https://image.tmdb.org/t/p/w500{poster_path}", stream=True).content))
new_images = [get_img(poster_path) for poster_path in details['poster_path']]
return new_titles[0], new_titles[1], new_titles[2], new_revenues[0], new_revenues[1], new_revenues[2], new_images[0], new_images[1], new_images[2]
def fetch_prediction_for_title(title):
movie_id = utils.get_movie_by_title(title)['results'][0]['id']
project = hopsworks.login()
fs = project.get_feature_store(name='igijul63_featurestore')
fg = fs.get_feature_group('movie_predictions', version=1)
query = fg.select_all()
new_predictions_df = query.read()
print(f"Got predictions df")
movie_details = utils.get_movie_details(movie_id)
if movie_id not in new_predictions_df['id'].tolist():
features = utils.extract_features(movie_id, latest=False)
print(features)
features_df = pd.DataFrame([features])
features_df = features_df[['top_cast_popularity', 'budget', 'crew_popularity', 'similar_revenues']]
features_df[['budget','similar_revenues']] /= 1000000
model = utils.load_model()
output = model.predict(features_df)
new_revenue = output * 1000000
# new_revenue = random.randint(200 * 1e6, 1000 * 1e6)
else:
movie_prediction = new_predictions_df[new_predictions_df['id'] == movie_id]
new_revenue = movie_prediction['revenue']
movie_details['predicted_revenue'] = new_revenue # movie_prediction['revenue']
new_title = f"### {movie_details['title']}"
new_revenue = f"#### Predicted revenue is {int(movie_details['predicted_revenue'])}\$"
get_img = lambda poster_path : Image.open(BytesIO(requests.get(f"https://image.tmdb.org/t/p/w500{poster_path}", stream=True).content))
new_image = get_img(movie_details['poster_path'])
budget = f"#### Movie with budget of {movie_details['budget']} \$"
overview = movie_details['overview']
hopsworks.logout()
return new_revenue, new_image, budget, overview
new_data = fetch_new_predictions()
search_data = fetch_prediction_for_title(new_data[0][4:])
with gr.Blocks("soft") as demo:
live = True
gr.Markdown("## Movie revenue prediction")
with gr.Tabs():
with gr.TabItem("New movies"):
gr.Markdown("# Prediction for recently released movies")
gr.Markdown("Press the button to see other new predictions!")
reload_button = gr.Button(value="Refresh")
with gr.Row():
title_0 = gr.Markdown(new_data[0])
title_1 = gr.Markdown(new_data[1])
title_2 = gr.Markdown(new_data[2])
with gr.Row():
image_0 = gr.Image(new_data[6])
image_1 = gr.Image(new_data[7])
image_2 = gr.Image(new_data[8])
with gr.Row():
prediction_0 = gr.Markdown(new_data[3])
prediction_1 = gr.Markdown(new_data[4])
prediction_2 = gr.Markdown(new_data[5])
with gr.TabItem("Search a movie"):
gr.Markdown("## Predict a revenue for a selected movie! You can type the title you want to search for and we'll see if we can predict how much it will earn.")
with gr.Row():
with gr.Column(scale = 3):
input_title = gr.Textbox(value = new_data[0][4:])
with gr.Column(scale = 1):
search_button = gr.Button(value="Search")
with gr.Row():
with gr.Column(scale=3):
with gr.Row():
prediction_s = gr.Markdown(search_data[0])
with gr.Row():
budget = gr.Markdown(search_data[2])
with gr.Row():
overview = gr.Markdown(search_data[3])
with gr.Column(scale=1):
image_search = gr.Image(search_data[1])
search_button.click(fn = fetch_prediction_for_title, inputs = [input_title], outputs = [prediction_s, image_search, budget, overview])
reload_button.click(fn = fetch_new_predictions, inputs=[], outputs=[title_0, title_1, title_2, prediction_0, prediction_1, prediction_2, image_0, image_1, image_2])
demo.launch()
# iface.launch()