trained.rst

Use a Trained Model

Load trained model

We can load a trained model from a project directory you've saved:

import tai_chi_engine as tce
PROJECT_DIR = '/path/to/project/dir'
trained = tce.TaiChiTrained(PROJECT_DIR)

tai_chi_engine.trained

TaiChiTrained

__init__

We can also load it for GPU inference:

trained = tce.TaiChiTrained(PROJECT_DIR, device='cuda:0')

Inference in python

The TaiChiTrained class has a predict method that can be used to make prediction on python dictionary:

x = {'img': image}
pred = trained.predict(x)

Notice, the key 'img' is the pandas column name, if you trained the model to use "user_id","movie_id" as the X columns, the input data looks like:

x = {'user_id': 128, 'movie_id': 42}
pred = trained.predict(x)

Other elements in TaiChiTrained

• `TaiChiTrained.final_model`: the trained model
• `TaiChiTrained.x_columns`: the X columns used to train the model
• `TaiChiTrained.y_columns`: the Y columns used to train the model
• `TaiChiTrained.phase`: The configuration object
• `TaiChiTrained.device`: The inference device
• `TaiChiTrained.qdict`: A dictionary of Quantify

Streamlit Deployment

Start a streamlit app to demonstrate your prototype:

from tai_chi_engine.app import StartStreamLit
# You'll have to pick a trained project folder, and assign a port
tc_app = StartStreamLit("./project_directory", port = 8501)
tc_app.start()

Then you can see some thing like following on your browser:

http://localhost:8501/

Stop the streamlit app by doing :

tc_app.stop()

Or pkill -f streamlit