{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "# NLP model creation and training" ] }, { "cell_type": "code", "execution_count": 1, "metadata": { "hide_input": true }, "outputs": [], "source": [ "from fastai.gen_doc.nbdoc import *\n", "from fastai.text import * \n", "from fastai import *" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "The main thing here is [`RNNLearner`](/text.learner.html#RNNLearner). There are also some utility functions to help create and update text models." ] }, { "cell_type": "code", "execution_count": 2, "metadata": { "hide_input": true }, "outputs": [ { "data": { "text/markdown": [ "

class RNNLearner[source]

\n", "\n", "> RNNLearner(`data`:[`DataBunch`](/basic_data.html#DataBunch), `model`:[`Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module), `bptt`:`int`=`70`, `split_func`:`OptSplitFunc`=`None`, `clip`:`float`=`None`, `adjust`:`bool`=`False`, `alpha`:`float`=`2.0`, `beta`:`float`=`1.0`, `kwargs`) :: [`Learner`](/basic_train.html#Learner)" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "show_doc(RNNLearner, doc_string=False)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Handles the whole creation of a [`Learner`](/basic_train.html#Learner) from data and a `model` with a text data using a certain `bptt`. The `split_func` is used to properly split the model in different groups for gradual unfreezing and differential learning rates. Gradient clipping of `clip` is optionally applied. `adjust`, `alpha` and `beta` are all passed to create an instance of [`RNNTrainer`](/callbacks.rnn.html#RNNTrainer). Can be used for a language model or an RNN classifier. It also handles the conversion of weights from a pretrained model as well as saving or loading the encoder." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Factory methods" ] }, { "cell_type": "code", "execution_count": 3, "metadata": { "hide_input": true }, "outputs": [ { "data": { "text/markdown": [ "

classifier[source]

\n", "\n", "> classifier(`data`:[`DataBunch`](/basic_data.html#DataBunch), `bptt`:`int`=`70`, `max_len`:`int`=`1400`, `emb_sz`:`int`=`400`, `nh`:`int`=`1150`, `nl`:`int`=`3`, `lin_ftrs`:`Collection`\\[`int`\\]=`None`, `ps`:`Collection`\\[`float`\\]=`None`, `pad_token`:`int`=`1`, `drop_mult`:`float`=`1.0`, `qrnn`:`bool`=`False`, `kwargs`) → `RNNLearner`" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "show_doc(RNNLearner.classifier, doc_string=False)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Create an RNNLearner with a classifier model from [`data`](/text.data.html#text.data). The model used is the encoder of an [AWD-LSTM](https://arxiv.org/abs/1708.02182) that is built with embeddings of size `emb_sz`, a hidden size of `nh`, and `nl` layers (the `vocab_size` is inferred from the [`data`](/text.data.html#text.data)). All the dropouts are put to values that we found worked pretty well and you can control their strength by adjusting `drop_mult`. If qrnn is True, the model uses [QRNN cells](https://arxiv.org/abs/1611.01576) instead of LSTMs.\n", "\n", "The input texts are fed into that model by bunch of `bptt` and only the last `max_len` activations are considerated. This gives us the backbone of our model. The head then consists of:\n", "- a layer that concatenates the final outputs of the RNN with the maximum and average of all the intermediate outputs (on the sequence length dimension),\n", "- blocks of ([`nn.BatchNorm1d`](https://pytorch.org/docs/stable/nn.html#torch.nn.BatchNorm1d), [`nn.Dropout`](https://pytorch.org/docs/stable/nn.html#torch.nn.Dropout), [`nn.Linear`](https://pytorch.org/docs/stable/nn.html#torch.nn.Linear), [`nn.ReLU`](https://pytorch.org/docs/stable/nn.html#torch.nn.ReLU)) layers.\n", "\n", "The blocks are defined by the `lin_ftrs` and `drops` arguments. Specifically, the first block will have a number of inputs inferred from the backbone arch and the last one will have a number of outputs equal to data.c (which contains the number of classes of the data) and the intermediate blocks have a number of inputs/outputs determined by `lin_ftrs` (of course a block has a number of inputs equal to the number of outputs of the previous block). The dropouts all have a the same value ps if you pass a float, or the corresponding values if you pass a list. Default is to have an intermediate hidden size of 50 (which makes two blocks model_activation -> 50 -> n_classes) with a dropout of 0.1." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "path = untar_data(URLs.IMDB_SAMPLE)\n", "data = TextClasDataBunch.from_csv(path)\n", "learn = RNNLearner.classifier(data, drop_mult=0.5)" ] }, { "cell_type": "code", "execution_count": 4, "metadata": { "hide_input": true }, "outputs": [ { "data": { "text/markdown": [ "

language_model[source]

\n", "\n", "> language_model(`data`:[`DataBunch`](/basic_data.html#DataBunch), `bptt`:`int`=`70`, `emb_sz`:`int`=`400`, `nh`:`int`=`1150`, `nl`:`int`=`3`, `pad_token`:`int`=`1`, `drop_mult`:`float`=`1.0`, `tie_weights`:`bool`=`True`, `bias`:`bool`=`True`, `qrnn`:`bool`=`False`, `pretrained_model`=`None`, `pretrained_fnames`:`OptStrTuple`=`None`, `kwargs`) → `RNNLearner`" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "show_doc(RNNLearner.language_model, doc_string=False)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Create an RNNLearner with a language model from [`data`](/text.data.html#text.data) of a certain `bptt`. The model used is an [AWD-LSTM](https://arxiv.org/abs/1708.02182) that is built with embeddings of size `emb_sz`, a hidden size of `nh`, and `nl` layers (the `vocab_size` is inferred from the [`data`](/text.data.html#text.data)). All the dropouts are put to values that we found worked pretty well and you can control their strength by adjusting `drop_mult`. If qrnn is True, the model uses [QRNN cells](https://arxiv.org/abs/1611.01576) instead of LSTMs. The flag `tied_weights` control if we should use the same weights for the encoder and the decoder, the flag `bias` controls if the last linear layer (the decoder) has bias or not.\n", "\n", "You can specify `pretrained_fnames` if you want to use the weights of a pretrained model. This should be a list of the name of the weight file and the name of the corresponding dictionary. The dictionary is needed because the function will internally convert the embeddings of the pretrained models to match the dictionary of the [`data`](/text.data.html#text.data) passed (a word may have a different id for the pretrained model). Those two files should be in the models directory of `data.path`. " ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "path = untar_data(URLs.IMDB_SAMPLE)\n", "data = TextLMDataBunch.from_csv(path)\n", "learn = RNNLearner.language_model(data, pretrained_model=URLs.WT103, drop_mult=0.5)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Loading and saving" ] }, { "cell_type": "code", "execution_count": 5, "metadata": { "hide_input": true }, "outputs": [ { "data": { "text/markdown": [ "

load_encoder[source]

\n", "\n", "> load_encoder(`name`:`str`)\n", "\n", "Load the encoder `name` from the model directory. " ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "show_doc(RNNLearner.load_encoder)" ] }, { "cell_type": "code", "execution_count": 6, "metadata": { "hide_input": true }, "outputs": [ { "data": { "text/markdown": [ "

save_encoder[source]

\n", "\n", "> save_encoder(`name`:`str`)\n", "\n", "Save the encoder to `name` inside the model directory. " ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "show_doc(RNNLearner.save_encoder)" ] }, { "cell_type": "code", "execution_count": 7, "metadata": { "hide_input": true }, "outputs": [ { "data": { "text/markdown": [ "

load_pretrained[source]

\n", "\n", "> load_pretrained(`wgts_fname`:`str`, `itos_fname`:`str`)" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "show_doc(RNNLearner.load_pretrained, doc_string=False)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Opens the weights in the `wgts_fname` of `self.model_dir` and the dictionary in `itos_fname` then adapts the pretrained weights to the vocabulary of the data. The two files should be in the models directory of the `learner.path`." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Utility functions" ] }, { "cell_type": "code", "execution_count": 8, "metadata": { "hide_input": true }, "outputs": [ { "data": { "text/markdown": [ "

lm_split[source]

\n", "\n", "> lm_split(`model`:[`Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module)) → `List`\\[[`Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module)\\]\n", "\n", "Split a RNN `model` in groups for differential learning rates. " ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "show_doc(lm_split)" ] }, { "cell_type": "code", "execution_count": 9, "metadata": { "hide_input": true }, "outputs": [ { "data": { "text/markdown": [ "

rnn_classifier_split[source]

\n", "\n", "> rnn_classifier_split(`model`:[`Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module)) → `List`\\[[`Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module)\\]\n", "\n", "Split a RNN `model` in groups for differential learning rates. " ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "show_doc(rnn_classifier_split)" ] }, { "cell_type": "code", "execution_count": 10, "metadata": { "hide_input": true }, "outputs": [ { "data": { "text/markdown": [ "

convert_weights[source]

\n", "\n", "> convert_weights(`wgts`:`Weights`, `stoi_wgts`:`Dict`\\[`str`, `int`\\], `itos_new`:`StrList`) → `Weights`" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "show_doc(convert_weights, doc_string=False)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Convert the `wgts` from an dictionary `stoi_wgts` (mapping of word to id) to a new dictionary `itos_new` (correspondans id to word)." ] } ], "metadata": { "jekyll": { "keywords": "fastai", "summary": "Easy access of language models and ULMFiT", "title": "text.learner" }, "kernelspec": { "display_name": "Python 3", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.6.6" } }, "nbformat": 4, "nbformat_minor": 2 }