Weather Data Visualization and Prediction with TensorFlow.js
Weather Data Visualization and Prediction with TensorFlow.js
Description
This demo showcases
visualization of temporal sequential data with
the tfjs-vis
library
predicting future values based on sequential input data using
various model types including linear regressors, multilayer
perceptrons (MLPs) and recurrent neural networks (RNNs).
To visualize and explore the data that goes into the model,
select various columns of the dataset in the
"Data series 1" and "Data series 2" dropdown menus. Experiment with normalization
and no-normalization options. Try showing the data at different time spans.
At time spans narrower than "full", you can use the left and right arrow
buttons to navigate along the time axis. Finally, try plotting two data series
against each other as a scatter plot by checking the "Plot against each other"
checkbox.
To train a linear-regression model or a multilayer perceptron (MLP),
specify the number of training epochs
and click "Train model". Wait patiently for the training to finish.
Loss values from the training and validation dataset will be refreshed in the
tfjs-vis visor surface on the right-hand side of the page at the end of
every training epoch. Experiment with regularization and dropout and observe
their effects on overfitting.