Permalink
395 lines (291 sloc) 25.6 KB

Classification models in DeepPavlov

In this repository one can find code for training and using classification models which are implemented as a number of different neural networks (for example, shallow-and-wide Convolutional Neural Network [1]) or sklearn models. Models can be used for binary, multi-class or multi-label classification.

Available classifiers are:

  • deeppavlov.models.classifiers.KerasClassificationModel (registered as keras_classification_model) builds neural network on Keras with tensorflow backend. One of the available network configurations can be chosen in model_name parameter in config. List of implemented networks can be found bellow.
  • deeppavlov.models.sklearn.SklearnComponent (registered as sklearn_component) builds most of sklearn classifiers. Chosen model should be passed to model_class, e.g. "model_class": "sklearn.neighbors:KNeighborsClassifier", as well as infer_method can be assigned to any sklearn model's prediction methods (e.g. predict or predict_proba). As for text classification in DeepPavlov we assign list of labels for each sample, it is required to ensure that output of a classifier-sklearn_component is a list of labels for each sample. Therefore, for sklearn component classifier one should set ensure_list_output to true.

Quick start

First you would need to install additional requirements:

python -m deeppavlov install <path_to_config>

where <path_to_config> is a path to one of the :config:`provided config files <classifiers>` or its name without an extension, for example :config:`intents_snips <classifiers/intents_snips.json>`.

One can run the following command to try provided pipelines out:

python -m deeppavlov interact <path_to_config> [-d]

where <path_to_config> is one of the :config:`provided config files <classifiers>`. With the optional -d parameter all the data required to run selected pipeline will be downloaded.

One can also use these configs in your python code. To download required data one have to set download parameter to True.

from deeppavlov import build_model, configs

CONFIG_PATH = configs.classifiers.intents_snips

model = build_model(CONFIG_PATH, download=True)

print(model(["What is the weather in Boston today?"]))

>>> [['GetWeather']]

Pre-trained models

We also provide with pre-trained models for classification on DSTC 2 dataset, SNIPS dataset, "AG News" dataset, "Detecting Insults in Social Commentary", Twitter sentiment in Russian dataset.

DSTC 2 dataset does not initially contain information about intents, therefore, Dstc2IntentsDatasetIterator (deeppavlov/dataset_iterators/dstc2_intents_interator.py) instance extracts artificial intents for each user reply using information from acts and slots.

Below we give several examples of intent construction:

System: "Hello, welcome to the Cambridge restaurant system. You can ask for restaurants by area, price range or food type. How may I help you?"

User: "cheap restaurant"

In the original dataset this user reply has characteristics

"goals": {"pricerange": "cheap"},
"db_result": null,
"dialog-acts": [{"slots": [["pricerange", "cheap"]], "act": "inform"}]}

This message contains only one intent: inform_pricerange.

User: "thank you good bye",

In the original dataset this user reply has characteristics

"goals": {"food": "dontcare", "pricerange": "cheap", "area": "south"},
"db_result": null,
"dialog-acts": [{"slots": [], "act": "thankyou"}, {"slots": [], "act": "bye"}]}

This message contains two intents (thankyou, bye). Train, valid and test division is the same as on web-site.

SNIPS dataset contains intent classification task for 7 intents (approximately 2.4 samples per intent):

  • GetWeather
  • BookRestaurant
  • PlayMusic
  • AddToPlaylist
  • RateBook
  • SearchScreeningEvent
  • SearchCreativeWork

Initially, classification model on SNIPS dataset was trained only as an example of usage that is why we provide pre-trained model for SNIPS with embeddings trained on DSTC-2 dataset that is not the best choice for this task. Train set is divided to train and validation sets to illustrate basic_classification_iterator work.

Detecting Insults in Social Commentary dataset contains binary classification task for detecting insults for participants of conversation. Train, valid and test division is the same as for the Kaggle challenge.

AG News dataset contains topic classification task for 5 classes (range from 0 to 4 points scale). Test set is initial one from a web-site, valid is a Stratified division 1/5 from the train set from web-site with 42 seed, and the train set is the rest.

Twitter mokoron dataset contains sentiment classification of Russian tweets for positive and negative replies [2]. It was automatically labeled. Train, valid and test division is made by hands (Stratified division: 1/5 from all dataset for test set with 42 seed, then 1/5 from the rest for validation set with 42 seed). Two provided pre-trained models were trained on the same dataset but with and without preprocessing. The main difference between scores is caused by the fact that some symbols (deleted during preprocessing) were used for automatic labelling. Therefore, it can be considered that model trained on preprocessed data is based on semantics while model trained on unprocessed data is based on punctuation and syntax.

RuSentiment dataset contains sentiment classification of social media posts for Russian language within 5 classes 'positive', 'negative', 'neutral', 'speech', 'skip'.

Questions on Yahoo Answers labeled as either informational or conversational dataset contains intent classification of English questions into two category: informational (0) and conversational (1) questions. The dataset includes some additional metadata but for the presented pre-trained model only Title of questions and Label were used. Embeddings were obtained from language model (ELMo) fine-tuned on the dataset L6 - Yahoo! Answers Comprehensive Questions and Answers. We do not provide datasets, both are available upon request to Yahoo Research. Therefore, this model is available only for interaction.

Task Dataset Lang Model Metric Valid Test Downloads
28 intents DSTC 2 En :config:`DSTC 2 emb <classifiers/intents_dstc2.json>` Accuracy 0.7732 0.7868 800 Mb
:config:`Wiki emb <classifiers/intents_dstc2_big.json>` 0.9602 0.9593 8.5 Gb
7 intents SNIPS-2017 :config:`DSTC 2 emb <classifiers/intents_snips.json>` F1 0.8685 -- 800 Mb
:config:`Wiki emb <classifiers/intents_snips_big.json>` 0.9811 -- 8.5 Gb
:config:`Tfidf + SelectKBest + PCA + Wiki emb <classifiers/intents_snips_sklearn.json>` 0.9673 -- 8.6 Gb
:config:`Wiki emb weighted by Tfidf <classifiers/intents_snips_tfidf_weighted.json>` 0.9786 -- 8.5 Gb
Insult detection Insults :config:`Reddit emb <classifiers/insults_kaggle.json>` ROC-AUC 0.9271 0.8618 6.2 Gb
5 topics AG News :config:`Wiki emb <classifiers/topic_ag_news.json>` Accuracy 0.8876 0.9011 8.5 Gb
Sentiment Twitter mokoron Ru :config:`RuWiki+Lenta emb w/o preprocessing <classifiers/sentiment_twitter.json>` 0.9972 0.9971 6.2 Gb
:config:`RuWiki+Lenta emb with preprocessing <classifiers/sentiment_twitter_preproc.json>` 0.7811 0.7749 6.2 Gb
RuSentiment :config:`RuWiki+Lenta emb <classifiers/rusentiment_cnn.json>` F1 0.6393 0.6539 6.2 Gb
:config:`ELMo <classifiers/rusentiment_elmo.json>` 0.7066 0.7301 700 Mb
Intent Yahoo-L31 :config:`Yahoo-L31 on ELMo <classifiers/yahoo_convers_vs_info.json>` pre-trained on Yahoo-L6 ROC-AUC 0.9269 -- 700 Mb

Download pre-trained model

DeepPavlov provides the following pre-trained models:

To download pre-trained models, vocabs, embeddings on the dataset of interest one should run the following command providing corresponding name of the config file (see above):

python -m deeppavlov download deeppavlov/configs/classifiers/intents_dstc2.json

or provide flag -d for commands like interact, interactbot, etc. The flag -d provides downloading all the required components.

Infer from pre-trained model

Pre-trained models can be used for inference in the following way:

from deeppavlov import build_model, configs

snips_model = build_model(configs.classifiers.intents_snips , download=True)
snips_model(["Hello! What is the weather in Boston tomorrow?"])

or from command line:

python -m deeppavlov interact deeppavlov/configs/classifiers/intents_dstc2.json [-d]

or

python -m deeppavlov interactbot deeppavlov/configs/classifiers/intents_dstc2.json -t <TELEGRAM_TOKEN> [-d]

For 'interactbot' mode one should specify a Telegram bot token in -t parameter or in the TELEGRAM_TOKEN environment variable.

Now user can enter a text string and get output of two elements: the first one is an array of classes names which the string belongs to, and the second one is a dictionary with probability distribution among all the considered classes (take into account that for multi-class classification then sum of probabilities is not equal to 1).

An example of interacting the model from :config:`intents_dstc2.json <classifiers/intents_dstc2.json>`

:: hey! I want cheap restaurant
>> (array(['inform_pricerange'], dtype='<U17'), {'ack': 0.0040760376, 'affirm': 0.017633557, 'bye': 0.023906048, 'confirm_area': 0.0040424005, 'confirm_food': 0.012261569, 'confirm_pricerange': 0.007227284, 'deny_food': 0.003502861, 'deny_name': 0.003412795, 'hello': 0.0061915903, 'inform_area': 0.15999688, 'inform_food': 0.18303667, 'inform_name': 0.0042709936, 'inform_pricerange': 0.30197725, 'inform_this': 0.03864918, 'negate': 0.016452404, 'repeat': 0.003964727, 'reqalts': 0.026930325, 'reqmore': 0.0030793257, 'request_addr': 0.08075432, 'request_area': 0.018258458, 'request_food': 0.018060096, 'request_phone': 0.07433994, 'request_postcode': 0.012727374, 'request_pricerange': 0.024933394, 'request_signature': 0.0034591882, 'restart': 0.0038622846, 'thankyou': 0.036836267, 'unknown': 0.045310754})

and an example of interacting the model from :config:`intents_dstc2_big.json <classifiers/intents_dstc2_big.json>`

::I want cheap chinese restaurant
>> (array(['inform_food', 'inform_pricerange'], dtype='<U18'), {'ack': 0.008203662, 'affirm': 0.010941843, 'bye': 0.0058273915, 'confirm_area': 0.011861361, 'confirm_food': 0.017537124, 'confirm_pricerange': 0.012897875, 'deny_food': 0.009804511, 'deny_name': 0.008331243, 'hello': 0.009887574, 'inform_area': 0.009167877, 'inform_food': 0.9627541, 'inform_name': 0.008696462, 'inform_pricerange': 0.98613375, 'inform_this': 0.009358878, 'negate': 0.011380567, 'repeat': 0.00850759, 'reqalts': 0.012249454, 'reqmore': 0.008230184, 'request_addr': 0.006192594, 'request_area': 0.009336099, 'request_food': 0.008417402, 'request_phone': 0.004564096, 'request_postcode': 0.006752021, 'request_pricerange': 0.010917218, 'request_signature': 0.008601435, 'restart': 0.00838949, 'thankyou': 0.0060319724, 'unknown': 0.010502234})

Train model

Available Neural models

DeepPavlov contains a number of different model configurations for classification task. Below the list of available models is presented:

  • cnn_model -- Shallow-and-wide CNN with max pooling after convolution,
  • dcnn_model -- Deep CNN with number of layers determined by the given number of kernel sizes and filters,
  • cnn_model_max_and_aver_pool -- Shallow-and-wide CNN with max and average pooling concatenation after convolution,
  • bilstm_model -- Bidirectional LSTM,
  • bilstm_bilstm_model -- 2-layers bidirectional LSTM,
  • bilstm_cnn_model -- Bidirectional LSTM followed by shallow-and-wide CNN,
  • cnn_bilstm_model -- Shallow-and-wide CNN followed by bidirectional LSTM,
  • bilstm_self_add_attention_model -- Bidirectional LSTM followed by self additive attention layer,
  • bilstm_self_mult_attention_model -- Bidirectional LSTM followed by self multiplicative attention layer,
  • bigru_model -- Bidirectional GRU model.

Please, pay attention that each model has its own parameters that should be specified in config.

Train again on provided datasets

To train from pre-trained model, re-train a model or train it with other parameters on one of the provided datasets, one should set save_path to a directory where the trained model will be saved (pre-trained model will be loaded if load_path is provided and files exist, otherwise it will be created from scratch). All other parameters of the model as well as embedder, tokenizer and preprocessor could be changed. Then training can be run in the following way:

python -m deeppavlov train "path_to_config"

Train on other datasets

Constructing intents from DSTC 2 makes Dstc2IntentsDatasetIterator difficult to use. Therefore, we also provide another dataset reader BasicClassificationDatasetReader and dataset BasicClassificationDatasetIterator to work with .csv and .json files. These classes are described in deeppavlov/dataset_readers/basic_classification_reader.py and deeppavlov/dataset_iterators/basic_classification_dataset_iterator.py.

Data files should be in the following format:

x y
text_0 intent_0
text_1 intent_0
text_2 intent_1,intent_2
text_3 intent_1,intent_0,intent_2
... ...

To train model one should

  • set data_path to the directory to which train.csv should be downloaded,
  • set save_path to the directory where the trained model should be saved,
  • set all other parameters of model as well as embedder, tokenizer and preprocessor to desired ones.

Then training process can be run in the same way:

python -m deeppavlov train "path_to_config"

The current version of :config:`intents_snips.json <classifiers/intents_snips.json>`` contains parameters for intent recognition for SNIPS benchmark dataset that was restored in .csv format and will be downloaded automatically.

Important: we do not provide any special embedding binary file for SNIPS dataset. In order to train the model one should provide own embedding binary file, because embedding file trained on DSTC-2 dataset is not the best choice for this task.

Comparison

As no one had published intent recognition for DSTC-2 data, the comparison of the presented model is given on SNIPS dataset. The evaluation of model scores was conducted in the same way as in [3] to compare with the results from the report of the authors of the dataset. The results were achieved with tuning of parameters and embeddings trained on Reddit dataset.

Model AddToPlaylist BookRestaurant GetWheather PlayMusic RateBook SearchCreativeWork SearchScreeningEvent
api.ai 0.9931 0.9949 0.9935 0.9811 0.9992 0.9659 0.9801
ibm.watson 0.9931 0.9950 0.9950 0.9822 0.9996 0.9643 0.9750
microsoft.luis 0.9943 0.9935 0.9925 0.9815 0.9988 0.9620 0.9749
wit.ai 0.9877 0.9913 0.9921 0.9766 0.9977 0.9458 0.9673
snips.ai 0.9873 0.9921 0.9939 0.9729 0.9985 0.9455 0.9613
recast.ai 0.9894 0.9943 0.9910 0.9660 0.9981 0.9424 0.9539
amazon.lex 0.9930 0.9862 0.9825 0.9709 0.9981 0.9427 0.9581
               
Shallow-and-wide CNN 0.9956 0.9973 0.9968 0.9871 0.9998 0.9752 0.9854

How to improve the performance

  • One can use FastText [4] to train embeddings that are better suited for considered datasets.
  • One can use ELMo [5] embeddings.
  • All the parameters should be tuned on the validation set.

References

[1]Kim Y. Convolutional neural networks for sentence classification //arXiv preprint arXiv:1408.5882. – 2014.
[2]Ю. В. Рубцова. Построение корпуса текстов для настройки тонового классификатора // Программные продукты и системы, 2015, №1(109), –С.72-78
[3]https://www.slideshare.net/KonstantinSavenkov/nlu-intent-detection-benchmark-by-intento-august-2017
[4]
  1. Bojanowski, E. Grave, A. Joulin, T. Mikolov, Enriching Word Vectors with Subword Information.
[5]Peters, Matthew E., et al. "Deep contextualized word representations." arXiv preprint arXiv:1802.05365 (2018).