atmarges / model2module-template

Deploy your trained deep learning models for NLP as simple python module.

Model2Module Template

Deploy your trained deep learning models for NLP as simple python module.

Note: Only applicable for NLP models trained using Keras

Quickstart

Create an instance of ModelTemplate and initialize it with the path of the following files:

• json_file - the .json file that contains the structure of your deep neural network model
• weight_file - the file containing the weights of your trained model
• words_file - a .tsv file containing a list of your word tokens. The first column should be the index of each word followed by the actual word tokens
• classes_file - a .tsv file containing a list of the class labels used on your model. You can include multiple labels by creating multiple columns. The first row will be the headers which will be later used to switch between different class labels.

Below is a sample code using a machine learing model trained for emotion classification:

from model2module.models import ModelTemplate

dataset = [
    "Nakakaasar nman! ahhh... ang dami dami kong gagawin... kulang time...",
    "Asar... Badtrip!",
    "May bagong bagyo ang namataan sa kanlurang bahagi ng bansa.",
    "Thank you po Lord Jesus at gumana din sya.",
    "Congrats po sa inyong lahat!",
]

json_file='C:/location/of/model/json_file.json',
weight_file='C:/location/of/model/weight_file.h5',
words_file='C:/location/of/model/words.tsv',
classes_file='C:/location/of/model/classes.tsv'

model = ModelTemplate(json_file=json_file, weight_file=weight_file,
                       words_file=words_file, classes_file=classes_file)
pred = model.predict_dataset(dataset, output_type='emotion')
print(pred)

You can set raw_output=True in the ModelTemplate.predict_dataset() function to retrieve output in vector form.

Below is what classes_file.tsv looks like. The first row contains the header which can be used in the output_type parameter of the ModelTemplate.predict_dataset() function

index	emoji	emotion	sentiment
0	📝	neutral	neutral
1	😄	happy	positive
2	😌	relief	positive
3	😑	unammused	negative
4	😘	love	positive
5	😜	playful	positive
6	😞	sad	negative
7	😡	angry	negative
8	😱	shocked	negative
9	😷	sick	negative
10	🤔	pondering	neutral

The file words_file.tsv looks like this. The first column contains the indexes of the tokens followed by the actual words.

1	aacts
2	aad
3	aada
4	aadal
5	aadalan
6	aadapt
7	aadc
8	aadd
9	aaddict
10	aadfghjkl