predictor ui

app.py

@@ -0,0 +1,109 @@
+import gradio as gr
+import numpy as np
+from PIL import Image
+import requests
+import re
+
+import hopsworks
+import joblib
+from consts import *
+
+project = hopsworks.login()
+fs = project.get_feature_store()
+
+mr = project.get_model_registry()
+model = mr.get_model("titanic", version=2)
+model_dir = model.download()
+model = joblib.load(model_dir + "/titanic_model.pkl")
+
+
+def get_deck(cabin):
+    if cabin is None:
+        return 0
+    deck = re.compile("([a-zA-Z]+)").search(cabin)
+    if deck is not None:
+        deck = deck.group()
+    if deck in DECK:
+        deck = DECK[deck]
+    else:
+        deck = 0
+    return deck
+
+
+def get_sex(sex):
+    return GENDERS[sex]
+
+
+def get_embarked(embarked):
+    return PORTS[embarked]
+
+
+def get_title(title):
+    if title in TITLES_RARE:
+        title = "Rare"
+    if title not in TITLES:
+        return 0
+    return TITLES[title]
+
+
+def get_age_class(age, p_class):
+    return age * p_class
+
+
+def get_relatives(sib_sp, parch):
+    return sib_sp + parch
+
+
+def get_not_alone(relatives):
+    return 1 if relatives > 0 else 0
+
+
+def get_fare_per_person(fare, relatives):
+    return fare / (relatives + 1)
+
+
+def titanic(p_class, sex, age, sib_sp, parch, fare, cabin, embarked, title):
+    # Model input:
+    # Pclass, Sex, Age, SibSp, Parch, Fare, Embarked, Deck,
+    # Title, Age_Class, Relatives, Not_alone, Fare_Per_Person
+    p_class = p_class + 1
+    deck = get_deck(cabin)
+    sex = get_sex(sex)
+    embarked = get_embarked(embarked)
+    title = get_title(title)
+    age_class = get_age_class(age, p_class)
+    relatives = get_relatives(sib_sp, parch)
+    not_alone = get_not_alone(relatives)
+    fare_per_person = get_fare_per_person(fare, relatives)
+    input_list = [p_class, sex, age, sib_sp, parch, fare, embarked,
+                  deck, title, age_class, relatives, not_alone, fare_per_person]
+    # 'res' is a list of predictions returned as the label.
+    res = model.predict(np.asarray(input_list).reshape(1, -1))
+    # We add '[0]' to the result of the transformed 'res', because 'res' is a list, and we only want
+    # the first element.
+    img_name = "alive.png" if res[0] == 1 else "dead.png"
+    img = Image.open(img_name)
+    return img
+
+
+demo = gr.Interface(
+    fn=titanic,
+    title="Titanic survival predictor",
+    description="Experiment with the parameters to predict if the person would or would not survive.",
+    allow_flagging="never",
+    inputs=[
+        gr.inputs.Dropdown(choices=["Class 1", "Class 2", "Class 3"], type="index", label="Class", default="Class 1"),
+        gr.inputs.Dropdown(choices=["male", "female"], type="value", label="Gender", default="male"),
+        gr.inputs.Number(label="Age"),
+        gr.inputs.Number(default=0, label="Number of sibling and spouses on board"),
+        gr.inputs.Number(default=0, label="Number of children/parents on board"),
+        gr.inputs.Number(label="Fare"),
+        gr.inputs.Textbox(label="Cabin"),
+        gr.inputs.Dropdown(choices=['S', 'C', 'Q'], label="Port of embarkation", default="S"),
+        gr.inputs.Textbox(label="Title")
+    ],
+    outputs=gr.Image(type="pil"))
+
+demo.launch()
+
+titanic(1, "male", 22, 1, 0, 150, "C85", "S", "Mr")

consts.py

@@ -0,0 +1,6 @@
+TITLES = {"Mr": 1, "Miss": 2, "Mrs": 3, "Master": 4, "Rare": 5}
+TITLES_RARE = ['Lady', 'Countess', 'Capt', 'Col', 'Don', 'Dr',
+               'Major', 'Rev', 'Sir', 'Jonkheer', 'Dona']
+DECK = {"A": 1, "B": 2, "C": 3, "D": 4, "E": 5, "F": 6, "G": 7, "U": 8}
+GENDERS = {"male": 0, "female": 1}
+PORTS = {"S": 0, "C": 1, "Q": 2}