Merge pull request #194 from jrzaurin/large-datasets

Large datasets

.coveragerc

@@ -5,7 +5,10 @@ omit =
     pytorch_widedeep/bayesian_models/bayesian_nn/modules/*
 
 [report]
+exclude_also =
+    def __repr__
 omit =
     pytorch_widedeep/optim/*
     pytorch_widedeep/bayesian_models/bayesian_nn/modules/*
-precision = 2
+precision = 2
+

VERSION

@@ -1 +1 @@
-1.3.2
+1.4.0

examples/scripts/adult_census.py

@@ -164,4 +164,4 @@
         )
 
         # 3. Either fit or directly predict
-        preds = trainer_new.predict(X_wide=X_wide, X_tab=X_tab)
+        preds = trainer_new.predict(X_wide=X_wide, X_tab=X_tab, batch_size=256)

examples/scripts/airbnb_load_from_folder_regr.py

@@ -0,0 +1,229 @@
+import numpy as np
+import torch
+import pandas as pd
+from torch.utils.data import DataLoader
+
+from pytorch_widedeep.models import TabResnet  # noqa: F401
+from pytorch_widedeep.models import TabMlp, Vision, BasicRNN, WideDeep
+from pytorch_widedeep.training import TrainerFromFolder
+from pytorch_widedeep.callbacks import EarlyStopping, ModelCheckpoint
+from pytorch_widedeep.preprocessing import (
+    ImagePreprocessor,
+    ChunkTabPreprocessor,
+    ChunkTextPreprocessor,
+)
+from pytorch_widedeep.load_from_folder import (
+    TabFromFolder,
+    TextFromFolder,
+    ImageFromFolder,
+    WideDeepDatasetFromFolder,
+)
+
+use_cuda = torch.cuda.is_available()
+
+if __name__ == "__main__":
+    # The airbnb dataset, which you could get from here:
+    # http://insideairbnb.com/get-the-data.html, is too big to be included in
+    # our datasets module (when including images). Therefore, go there,
+    # download it, and use the download_images.py script to get the images
+    # and the airbnb_data_processing.py to process the data. We'll find
+    # better datasets in the future ;). Note that here we are only using a
+    # small sample to illustrate the use, so PLEASE ignore the results, just
+    # focus on usage
+
+    # For this exercise, we use a small sample of the airbnb dataset,
+    # comprised of tabular data with a text column ('description') and an
+    # image column ('id') that point to the images of the properties listed
+    # in Airbnb. We know the size of the sample before hand (1001) so we set
+    # a series of parameters accordingly
+    train_size = 800
+    eval_size = 100
+    test_size = 101
+    chunksize = 100
+    n_chunks = int(np.ceil(train_size / chunksize))
+
+    data_path = "../tmp_data/airbnb/"
+    train_fname = "airbnb_sample_train.csv"
+    eval_fname = "airbnb_sample_eval.csv"
+    test_fname = "airbnb_sample_test.csv"
+
+    # the images are stored in the 'property_picture' while the text is a
+    # column in the 'airbnb_sample' dataframe. Let's then define the dir and
+    # file variables
+    img_path = "../tmp_data/airbnb/property_picture/"
+    img_col = "id"
+    text_col = "description"
+    target_col = "yield"
+    cat_embed_cols = [
+        "host_listings_count",
+        "neighbourhood_cleansed",
+        "is_location_exact",
+        "property_type",
+        "room_type",
+        "accommodates",
+        "bathrooms",
+        "bedrooms",
+        "beds",
+        "guests_included",
+        "minimum_nights",
+        "instant_bookable",
+        "cancellation_policy",
+        "has_house_rules",
+        "host_gender",
+        "accommodates_catg",
+        "guests_included_catg",
+        "minimum_nights_catg",
+        "host_listings_count_catg",
+        "bathrooms_catg",
+        "bedrooms_catg",
+        "beds_catg",
+        "security_deposit",
+        "extra_people",
+    ]
+    cont_cols = ["latitude", "longitude"]
+
+    # Now, processing the data from here on can be done in two ways:
+    # 1. The tabular data itsel fits in memory as is only the images that do
+    # not: in this case you could use the 'standard' Preprocessors and off
+    # you go, move directly to the '[...]FromFolder' functionalities
+
+    # 2. The tabular data is also very large and does not fit in memory, so we
+    # have to process it in chuncks. For this second case I have created the
+    # Chunk Processors (Wide, Tab and Text). Note that at the moment ONLY csv
+    # format is allowed for the tabular file. More formats will be supported
+    # in the future.
+
+    # For the following I will assume (simply for illustration purposes) that
+    # we are in the second case. Nonetheless, the process, whether 1 or 2,
+    # can be summarised as follows:
+    # 1. Process the data
+    # 2. Define the loaders from folder
+    # 3. Define the datasets and dataloaders
+    # 4. Define the model and the Trainer
+    # 5. Fit the model and Predict
+
+    # Process the data in Chunks
+    tab_preprocessor = ChunkTabPreprocessor(
+        embed_cols=cat_embed_cols,
+        continuous_cols=cont_cols,
+        n_chunks=n_chunks,
+        default_embed_dim=8,
+        verbose=0,
+    )
+
+    text_preprocessor = ChunkTextPreprocessor(
+        n_chunks=n_chunks,
+        text_col=text_col,
+        n_cpus=1,
+    )
+
+    # Note that all the (pre)processing of the images will occur 'on the fly',
+    # as they are loaded from disk. Therefore, the flow for the image dataset
+    # and for the tabular and text data modes is not entirely the same.
+    # Tabular and text data uses Chunk processors while such processing
+    # approach is not needed for the images
+    img_preprocessor = ImagePreprocessor(
+        img_col=img_col,
+        img_path=img_path,
+    )
+
+    for i, chunk in enumerate(
+        pd.read_csv("/".join([data_path, train_fname]), chunksize=chunksize)
+    ):
+        print(f"chunk in loop: {i}")
+        tab_preprocessor.fit(chunk)
+        text_preprocessor.fit(chunk)
+
+    # Instantiate the loaders from folder: again here some explanation is
+    # required. As I mentioned earlier the "[...]FromFolder" functionalities
+    # are thought for the case when we have tabular and text and/or image
+    # datasets and the latter do not fit in memory, so they have to be loaded
+    # from disk. With this in mind, the tabular data is the reference, and
+    # must have columns that point to the image files and to the text files
+    # (in case these exists instead of a column with the texts). Since the
+    # tabular data is used as a reference, is the one that has to be splitted
+    # in train/validation/test. The test and image 'FromFolder' objects only
+    # point to the corresponding column or files, and therefore, we do not
+    # need to create a separate instance per train/validation/test dataset
+    train_tab_folder = TabFromFolder(
+        fname=train_fname,
+        directory=data_path,
+        target_col=target_col,
+        preprocessor=tab_preprocessor,
+        text_col=text_col,
+        img_col=img_col,
+    )
+    eval_tab_folder = TabFromFolder(fname=eval_fname, reference=train_tab_folder)  # type: ignore[arg-type]
+    test_tab_folder = TabFromFolder(
+        fname=test_fname, reference=train_tab_folder, ignore_target=True  # type: ignore[arg-type]
+    )
+
+    text_folder = TextFromFolder(
+        preprocessor=text_preprocessor,
+    )
+
+    img_folder = ImageFromFolder(preprocessor=img_preprocessor)
+
+    # Following 'standard' pytorch approaches, we define the datasets and then
+    # the dataloaders
+    train_dataset_folder = WideDeepDatasetFromFolder(
+        n_samples=train_size,
+        tab_from_folder=train_tab_folder,
+        text_from_folder=text_folder,
+        img_from_folder=img_folder,
+    )
+    eval_dataset_folder = WideDeepDatasetFromFolder(
+        n_samples=eval_size,
+        tab_from_folder=eval_tab_folder,
+        reference=train_dataset_folder,
+    )
+    test_dataset_folder = WideDeepDatasetFromFolder(
+        n_samples=test_size,
+        tab_from_folder=test_tab_folder,
+        reference=train_dataset_folder,
+    )
+    train_loader = DataLoader(train_dataset_folder, batch_size=16, num_workers=1)
+    eval_loader = DataLoader(eval_dataset_folder, batch_size=16, num_workers=1)
+    test_loader = DataLoader(test_dataset_folder, batch_size=16, num_workers=1)
+
+    # And from here on, is all pretty standard within the library
+    basic_rnn = BasicRNN(
+        vocab_size=len(text_preprocessor.vocab.itos),
+        embed_dim=32,
+        hidden_dim=64,
+        n_layers=2,
+        head_hidden_dims=[100, 50],
+    )
+
+    deepimage = Vision(
+        pretrained_model_name="resnet18", n_trainable=0, head_hidden_dims=[200, 100]
+    )
+
+    deepdense = TabMlp(
+        mlp_hidden_dims=[64, 32],
+        column_idx=tab_preprocessor.column_idx,
+        cat_embed_input=tab_preprocessor.cat_embed_input,
+        continuous_cols=cont_cols,
+    )
+
+    model = WideDeep(
+        deeptabular=deepdense,
+        deeptext=basic_rnn,
+        deepimage=deepimage,
+    )
+
+    callbacks = [EarlyStopping, ModelCheckpoint(filepath="model_weights/wd_out.pt")]
+
+    trainer = TrainerFromFolder(
+        model,
+        objective="regression",
+        callbacks=callbacks,
+    )
+
+    trainer.fit(
+        train_loader=train_loader,
+        eval_loader=eval_loader,
+        finetune=True,
+        finetune_epochs=1,
+    )
+    preds = trainer.predict(test_loader=test_loader)

mkdocs/mkdocs.yml

@@ -26,6 +26,7 @@ nav:
             - Image utils: pytorch-widedeep/utils/image_utils.md
             - Text utils: pytorch-widedeep/utils/text_utils.md
         - Preprocessing: pytorch-widedeep/preprocessing.md
+        - Load From Folder: pytorch-widedeep/load_from_folder.md
         - Model Components: pytorch-widedeep/model_components.md
         - Bayesian models: pytorch-widedeep/bayesian_models.md
         - Losses: pytorch-widedeep/losses.md
@@ -59,6 +60,7 @@ nav:
         - 18_feature_importance_via_attention_weights: examples/18_feature_importance_via_attention_weights.ipynb
         - 19_wide_and_deep_for_recsys_pt1: examples/19_wide_and_deep_for_recsys_pt1.ipynb
         - 19_wide_and_deep_for_recsys_pt2: examples/19_wide_and_deep_for_recsys_pt2.ipynb
+        - 20_load_from_folder_functionality.: examples/20_load_from_folder_functionality.ipynb
     - Contributing: contributing.md
 
 theme:

mkdocs/site/404.html

@@ -406,6 +406,8 @@
 
 
 
+
+
 
 
 
@@ -553,6 +555,20 @@
 
 
 
+    <li class="md-nav__item">
+      <a href="/pytorch-widedeep/load_from_folder.html" class="md-nav__link">
+        Load From Folder
+      </a>
+    </li>
+
+
+
+
+
+
+
+
+
     <li class="md-nav__item">
       <a href="/pytorch-widedeep/model_components.html" class="md-nav__link">
         Model Components
@@ -749,6 +765,8 @@
 
 
 
+
+
 
 
 
@@ -1060,6 +1078,20 @@
 
 
 
+
+
+
+
+
+    <li class="md-nav__item">
+      <a href="/examples/20_load_from_folder_functionality.html" class="md-nav__link">
+        20_load_from_folder_functionality.
+      </a>
+    </li>
+
+
+
+
         </ul>
       </nav>
     </li>