Models¶

twgeo.models.geomodel¶

class twgeo.models.geomodel.Model(use_tensorboard=True, batch_size=64)¶

Geolocation prediction model. Consists of 4 layers ( Embedding, LSTM, LSTM and Dense).

Parameters:	use_tensorboard – Track training progress using Tensorboard. Default: true. batch_size – Default: 64

build_model(num_outputs, time_steps=500, vocab_size=20000, hidden_layer_size=128)¶

Build the model.

Parameters:	num_outputs – Number of output classes. For example, in the case of Census regions num of classes is 4. time_steps – Default: 500 vocab_size – Use the top N most frequent words. Default: 20,000 hidden_layer_size – Number of neurons in the hidden layers. Default: 128
Returns:

evaluate(x_test, y_test)¶

Get the loss, accuracy and top 5 accuracy of the model.

Parameters:	x_test – Evaluation samples. y_test – Evaluation labels.
Returns:	A dictionary of metric, value pairs.

load_saved_model(filename)¶

Load a previously trained model from disk.

Parameters:	filename – The H5 model.
Returns:

predict(x)¶

Predict the location of the given samples.

Parameters:	x – A vector of tweets. Each row corresponds to a single user.
Returns:	The prediction results.

save_model(filename)¶

Save the current model and trained weights for later use.

Parameters:	filename – Prefix for the model filenames.

train(x_train, y_train, x_dev, y_dev, epochs=7, reset_model=False)¶

Fit the model to the training data.

Parameters:	x_train – Training samples. y_train – Training labels. Must be a vector of integer values. x_dev – Validation samples. y_dev – Validation labels. Must be a vector of integer values. epochs – Number of times to train on the whole data set. Default: 7 reset_model – If this is set to True, it will discard any previously trained model and start from scratch.
Returns:
Raises:	ValueError: If the number of training samples and the number of labels do not match.