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.
from joblib import load # to load the existing saved model
import pandas as pd
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('dw_123')
df = dbconn.fetch('dw_123', query='select * from leads where prediction is null limit 5', df=True)
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.
model = load('/data_app/model_lead_conversion') # loading the model
# Making prediction
pred = model.predict(X)
# Saving the prediction to df
df['prediction'] = pred