Customizing the model search space

We can customize the model's search to restrict the search space by using AutoModel instead of TextClassifier, for example, by setting TextBlock for some specific configurations.

In the following code snippet, we're telling AutoKeras to only generate models that use 'ngram' to vectorize the sentences. Remember that if we do not specify any of these arguments, AutoKeras will automatically try all the possible combinations until the number reaches the max_trial parameter:

input_node = ak.TextInput()
output_node = ak.TextBlock(block_type="ngram")(input_node)
output_node = ak.ClassificationHead()(output_node)
clf = ak.AutoModel(inputs=input_node, 
                   outputs=output_node, overwrite=True,
                   max_trials...