Retrain with the training code¶
Training code on the Predict & export page is the complete Python script behind the model you chose, with every setting and random seed the Studio used. Run it on the same file and you get the same score. Run it on a newer file with the same columns and you get a fresh model, built the same way.
1. Save the script¶
Open Training code, copy it with the copy button, and save it as
train.py. The first lines name the scikit-learn and pandas versions it was
written for.
2. Put your data next to it¶
The script reads the file you trained on, by its original name, for example:
Put that file in the same folder, or change the name to your newer file. The columns must match.
Excel workbooks with several sheets
The script reads the first sheet. If the Studio used a later sheet, add
sheet_name="<the sheet's name>" to the pd.read_excel(...) line.
3. Run it¶
pip install scikit-learn==<version> pandas==<version> joblib # add openpyxl for Excel
python train.py
It prints the model's accuracy, or R² when predicting a number, next to what a
simple guess scores, measured on the same rows the Studio used. Both numbers
match the Metrics tab. It then refits on every row and saves the result as
model.joblib.
Not the same file as the Studio's download
The script saves the model on its own, without the extra details the
Studio's download carries. Load it with model = joblib.load("model.joblib")
and call model.predict(...) on the same columns.
Running it on the instance¶
The Python environment already has the right libraries. Copy the script and your data to the instance, for example through JupyterLab's file browser, and run it there.