Add predictions to new data, using saved ML models.

This template allows you to load a pre-trained ML model and apply the model to new records in a table, to make predictions on a scheduled basis. The predictions are written to a column “Prediction” in the same table in Peliqan.

In this tutorial we’ll use a Lead Conversion prediction model, you can modify the code to work with other ML models as well.

Import required modules

from joblib import load # to load the existing saved model
import pandas as pd

Load new records

We’ll load records that do not have a prediction yet. Make sure to add a “Prediction” column first in the spreadsheet view of the table.

dbconn = pq.dbconnect(pq.DW_NAME)
df = dbconn.fetch(pq.DW_NAME, query='select * from leads where prediction is null limit 5', df=True)

Data Preprocessing

For predicting, we have to prepare the data in the same format as it was trained on.

# Drop unwanted features
drop_features = ['Prospect ID', 'index', 'Prediction'] 

X = df.drop(drop_features, axis=1)

# Apply Label Encoding to convert categorical variables to Numerical
encoder = load('/data_app/encoder_lead_conversion')
cat_cols = X.select_dtypes('object').columns
X[cat_cols] = X[cat_cols].apply(lambda x: encoder[x.name].fit_transform(x))

# Input columns to make predictions and drop null values
X = X.drop('Converted', axis=1)

Note: It's recommended to add try to block to capture if there are no records to update other wise it will throw errors in the future.

Load the model & predict

model = load('/data_app/model_lead_conversion') # loading the model

# Making prediction
pred = model.predict(X)

# Saving the prediction to df
df['prediction'] = pred

Write predictions to table

We can accomplish this using the update_cell method from the Peliqan module pq:

for _, record in df.iterrows():
    pq.update_cell(table_name='leads', field_name='Prediction', row_pk=record['Prospect ID'], value=record['prediction'])
    st.text(f"Updated Prediction for {record['Prospect ID']}")

‣

Expand this to see the full code

What’s Next