octis 0.1.2

Install

You can install OCTIS with the following command:

pip install octis

You can find the requirements in the requirements.txt file.

Features

• We provide a set of state-of-the-art preprocessed text datasets (or you can preprocess your own dataset)

• We provide a set of well-known topic models (both classical and neurals), or you can integrate your own model

• You can evaluate your model using several state-of-the-art evaluation metrics

• You can optimize the hyperparameters of the models with respect to a given metric using Bayesian Optimization

• We provide a simple web dashboard for starting and controlling the optimization experiments

Get a preprocessed dataset

To acquire a dataset you can use one of the built-in sources.

from octis.dataset.dataset import Dataset
 dataset = Dataset()
 dataset.load("octis/preprocessed_datasets/m10")

Or use your own.

import octis.preprocessing.sources.custom_dataset as source
dataset = source.retrieve("path\to\dataset")

A custom dataset must have a document for each line of the file. Datasets can be partitioned in train and test sets.

Preprocess

To preprocess a dataset Initialize a Pipeline_handler and use the preprocess method.

from octis.preprocessing.pipeline_handler import Pipeline_handler

pipeline_handler = Pipeline_handler(dataset) # Initialize pipeline handler
preprocessed = pipeline_handler.preprocess() # preprocess

preprocessed.save("dataset_folder") # Save the preprocessed dataset

For the customization of the preprocess pipeline see the optimization demo example in the examples folder.

Train a model

To build a model, load a preprocessed dataset, customize the model hyperparameters and use the train_model() method of the model class.

from octis.dataset.dataset import Dataset
from octis.models.LDA import LDA

# Load a dataset
dataset = Dataset()
dataset.load("dataset_folder")

model = LDA(num_topics=25)  # Create model
model_output = model.train_model(dataset) # Train the model

If the dataset is partitioned, you can choose to:

• Train the model on the training set and test it on the test documents

• Train the model on the training set and update it with the test set

• Train the model with the whole dataset, regardless of any partition.

Evaluate a model

To evaluate a model, choose a metric and use the score() method of the metric class.

from octis.evaluation_metrics.diversity_metrics import TopicDiversity

# Set metric parameters
td_parameters ={'topk':10}

metric = TopicDiversity(td_parameters) # Initialize metric
topic_diversity_score = metric.score(model_output) # Compute score of the metric

Optimize a model

To optimize a model you need to select a dataset, a metric and the search space of the hyperparameters to optimize.

from octis.optimization.optimizer import Optimizer

search_space = {
"alpha": Real(low=0.001, high=5.0),
"eta": Real(low=0.001, high=5.0)
}

number_of_call=5
model_runs=3
save_path="results"
# Initialize an optimizer object and start the optimization.
optimizer=Optimizer()
OptObject=optimizer.optimize(model,dataset, npmi,search_space,
                                number_of_call=number_of_call,
                                model_runs=model_runs,
                                save_path=save_path)
#save the results of th optimization in a csv file
OptObject.save_to_csv("results.csv")

The result will provide best-seen value of the metric with the corresponding hyperparameter configuration, and the hyperparameters and metric value for each iteration of the optimization. To visualize this information, you have to set ‘plot’ attribute of Bayesian_optimization to True.

You can find more here: optimizer README

Examples and Tutorials

Our Colab Tutorials:

Available Models

• AVITM

• CTM

• ETM

• HDP

• LDA

• LSI

• NMF

• NeuralLDA

• ProdLDA

Available Datasets

• 20Newsgroup

• BBC News

• DBLP

• M10

Implement your own Model

Models inherit from the class Abstract_Model defined in models/model.py . To build your own model your class must override the train_model(self, dataset, hyperparameters) method which always require at least a Dataset object and a Dictionary of hyperparameters as input and should return a dictionary with the output of the model as output.

To better understand how a model work, let’s have a look at the LDA implementation. The first step in developing a custom model is to define the dictionary of default hyperparameters values:

hyperparameters = {'corpus': None, 'num_topics': 100,
    'id2word': None, 'alpha': 'symmetric',
    'eta': None, # ...
    'callbacks': None}

Defining the default hyperparameters values allows users to work on a subset of them without having to assign a value to each parameter.

The following step is the train_model() override:

def train_model(self, dataset, hyperparameters={}, top_words=10):

The LDA method requires a dataset, the hyperparameters dictionary and an extra (optional) argument used to select how many of the most significative words track for each topic.

With the hyperparameters defaults, the ones in input and the dataset you should be able to write your own code and return as output a dictionary with at least 3 entries:

• topics: the list of the most significative words foreach topic (list of lists of strings).

• topic-word-matrix: an NxV matrix of weights where N is the number of topics and V is the vocabulary length.

• topic-document-matrix: an NxD matrix of weights where N is the number of topics and D is the number of documents in the corpus.

if your model support the training/test partitioning it should also return:

• test-topic-document-matrix: the document topic matrix of the test set.

In case the model isn’t updated with the test set. Or:

• test-topics: the list of the most significative words foreach topic (list of lists of strings) of the model updated with the test set.

• test-topic-word-matrix: an NxV matrix of weights where N is the number of topics and V is the vocabulary length of the model updated with the test set.

• test-topic-document-matrix: an NxD matrix of weights where N is the number of topics and D is the number of documents in the corpus of the model updated with the test set.

If the model is updated with the test set.

Dashboard

OCTIS includes a user friendly graphical interface for creating, monitoring and viewing experiments. Following the implementation standards of datasets, models and metrics the dashboard will automatically update and allow you to use your own custom implementations.

To run rhe dashboard, while in the project directory run the following command:

python OCTIS/dashboard/server.py

The browser will open and you will be redirected to the dashboard. In the dashboard you can:

• Create new experiments organized in batch

• Visualize and compare all the experiments

• Visualize a custom experiment

• Manage the experiment queue

Team

Credits

This package was created with Cookiecutter and the audreyr/cookiecutter-pypackage project template.