Persistence diagrams workflow

.github/workflows/deploy-gh-pages.yml

@@ -17,6 +17,7 @@ jobs:
     - name: Install dependencies
       run: |
         python -m pip install --upgrade pip
+        python -m pip install torch
         if [ -f requirements.txt ]; then pip install -r requirements.txt; fi
     - name: Install sphinx
       run: |

.github/workflows/python-package-windows.yml

@@ -26,7 +26,7 @@ jobs:
     - name: Install dependencies
       run: |
         python -m pip install --upgrade pip
-        pip install flake8 pytest nbconvert jupyter
+        pip install flake8 pytest nbconvert jupyter torch
         if(Test-Path "requirements.txt")
         {
           pip install -r requirements.txt

.github/workflows/python-package.yml

@@ -26,7 +26,7 @@ jobs:
     - name: Install dependencies
       run: |
         python -m pip install --upgrade pip
-        pip install flake8 pytest nbconvert jupyter mypy
+        pip install flake8 pytest nbconvert jupyter mypy torch
         if [ -f requirements.txt ]; then pip install -r requirements.txt; fi
         pip install -e .
     - name: Lint with flake8

.gitignore

@@ -77,6 +77,7 @@ examples/diagrams_5000_1000_0_1.npy
 # virtualenv
 giotto-deep/
 venv/
+myenv/
 
 examples/model_data_specifications
 runs/
@@ -91,3 +92,5 @@ state_dicts/*
 
 # pylance files
 *.pyi
+
+giotto-env/

README.md

@@ -34,7 +34,12 @@ The first step to install the developer version of the package is to `git clone`
 ```
 git clone https://github.com/giotto-ai/giotto-deep.git
 ```
-The change the current working directory to the Repository root folder, e.g. `cd giotto-deep`. 
+The change the current working directory to the Repository root folder, e.g. `cd giotto-deep`.
+Make sure you have the latest version of pytorch installed. 
+You can do this by running the following command (if you have a GPU):
+```
+pip install torch --extra-index-url https://download.pytorch.org/whl/cu113
+```
 Once you are in the root folder, install the package dynamically with:
 ```
 pip install -e .

examples/mutag_pipeline.ipynb

@@ -0,0 +1 @@
+{"cells":[{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["import os\n","from typing import Tuple\n","\n","import torch.nn as nn\n","from gdeep.data import PreprocessingPipeline\n","from gdeep.data.datasets import PersistenceDiagramFromFiles\n","from gdeep.data.datasets.base_dataloaders import (DataLoaderBuilder,\n","                                                  DataLoaderParamsTuples)\n","from gdeep.data.datasets.persistence_diagrams_from_graphs_builder import \\\n","    PersistenceDiagramFromGraphBuilder\n","from gdeep.data.persistence_diagrams.one_hot_persistence_diagram import (\n","    OneHotEncodedPersistenceDiagram, collate_fn_persistence_diagrams)\n","from gdeep.data.preprocessors import (\n","    FilterPersistenceDiagramByHomologyDimension,\n","    FilterPersistenceDiagramByLifetime, NormalizationPersistenceDiagram)\n","from gdeep.search.hpo import GiottoSummaryWriter\n","from gdeep.topology_layers import Persformer, PersformerConfig, PersformerWrapper\n","from gdeep.topology_layers.persformer_config import PoolerType\n","from gdeep.trainer.trainer import Trainer\n","from gdeep.search import HyperParameterOptimization\n","from gdeep.utility import DEFAULT_GRAPH_DIR, PoolerType\n","from gdeep.utility.utils import autoreload_if_notebook\n","from sklearn.model_selection import train_test_split\n","from torch.optim import Adam\n","from torch.utils.data import Subset\n","from torch.utils.tensorboard.writer import SummaryWriter\n","\n","autoreload_if_notebook()\n","\n","# Parameters\n","name_graph_dataset: str = 'MUTAG'\n","diffusion_parameter: float = 0.1\n","num_homology_types: int = 4\n","\n","\n","# Create the persistence diagram dataset\n","pd_creator = PersistenceDiagramFromGraphBuilder(name_graph_dataset, diffusion_parameter)\n","pd_creator.create()"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["# Plot sample extended persistence diagram\n","file_path: str = os.path.join(DEFAULT_GRAPH_DIR,\n","                              f\"MUTAG_{diffusion_parameter}_extended_persistence\", \"diagrams\")\n","graph_idx = 1\n","pd: OneHotEncodedPersistenceDiagram = \\\n","    OneHotEncodedPersistenceDiagram.load(os.path.join(file_path, \n","                                                      f\"{graph_idx}.npy\"))\n","pd.set_homology_dimension_names([\"Ord0\", \"Ext0\", \"Rel1\", \"Ext1\"])\n","pd.plot()"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["\n","pd_mutag_ds = PersistenceDiagramFromFiles(\n","    os.path.join(\n","        DEFAULT_GRAPH_DIR, f\"MUTAG_{diffusion_parameter}_extended_persistence\"\n","        )\n",")\n","\n","pd_sample: OneHotEncodedPersistenceDiagram = pd_mutag_ds[0][0]\n","\n","fig = pd_sample.plot([\"Ord0\", \"Ext0\", \"Rel1\", \"Ext1\"])\n","# add title\n","fig.show()"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["\n","# Create the train/validation/test split\n","\n","train_indices, test_indices = train_test_split(\n","    range(len(pd_mutag_ds)),\n","    test_size=0.2,\n","    random_state=42,\n",")\n","\n","train_indices , validation_indices = train_test_split(\n","    train_indices,\n","    test_size=0.2,\n","    random_state=42,\n",")\n","\n","# Create the data loaders\n","train_dataset = Subset(pd_mutag_ds, train_indices)\n","validation_dataset = Subset(pd_mutag_ds, validation_indices)\n","test_dataset = Subset(pd_mutag_ds, test_indices)\n","\n","# Preprocess the data\n","preprocessing_pipeline = PreprocessingPipeline[Tuple[OneHotEncodedPersistenceDiagram, int]](\n","    (\n","        FilterPersistenceDiagramByHomologyDimension[int]([0, 1]),\n","        FilterPersistenceDiagramByLifetime[int](min_lifetime=-0.1, max_lifetime=1.0),\n","        NormalizationPersistenceDiagram[int](num_homology_dimensions=4),\n","     )\n",")\n","\n","preprocessing_pipeline.fit_to_dataset(train_dataset)\n",""]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["train_dataset = preprocessing_pipeline.attach_transform_to_dataset(train_dataset)  # type: ignore\n","validation_dataset = preprocessing_pipeline.attach_transform_to_dataset(validation_dataset)  # type: ignore\n","test_dataset = preprocessing_pipeline.attach_transform_to_dataset(test_dataset)  # type: ignore\n",""]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["\n","dl_params = DataLoaderParamsTuples.default(\n","    batch_size=32,\n","    num_workers=0,\n","    collate_fn=collate_fn_persistence_diagrams,\n","    with_validation=True,\n",")\n","\n","\n","# Build the data loaders\n","dlb = DataLoaderBuilder((train_dataset, validation_dataset, test_dataset))  # type: ignore\n","dl_train, dl_val, dl_test = dlb.build(dl_params)  # type: ignore"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["\n","# # Define the model\n","# model_config = PersformerConfig(\n","#     num_layers=6,\n","#     num_attention_heads=4,\n","#     input_size= 2 + num_homology_types,\n","#     ouptut_size=2,\n","#     pooler_type=PoolerType.ATTENTION,\n","# )\n","\n","# model = Persformer(model_config)\n","# writer = SummaryWriter()\n","\n","# loss_function =  nn.CrossEntropyLoss()\n","\n","# trainer = Trainer(model, [dl_train, dl_val, dl_test], loss_function, writer)\n","\n","# trainer.train(Adam, 3, False, \n","#               {\"lr\":0.01}, \n","#               {\"batch_size\":16, \"collate_fn\": collate_fn_persistence_diagrams})"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["\n","# Define the model by using a Wrapper for the Persformer model\n","wrapped_model = PersformerWrapper(\n","    num_attention_layers=3,\n","    num_attention_heads=4,\n","    input_size= 2 + num_homology_types,\n","    ouptut_size=2,\n","    pooler_type=PoolerType.ATTENTION,\n",")\n","writer = GiottoSummaryWriter()\n","\n","loss_function =  nn.CrossEntropyLoss()\n","\n","trainer = Trainer(wrapped_model, [dl_train, dl_val, dl_test], loss_function, writer)\n","\n","# initialise hpo object\n","search = HyperParameterOptimization(trainer, \"accuracy\", 2, best_not_last=True)\n","\n","# if you want to store pickle files of the models instead of the state_dicts\n","search.store_pickle = True\n","\n","# dictionaries of hyperparameters\n","optimizers_params = {\"lr\": [0.001, 0.01]}\n","dataloaders_params = {\"batch_size\": [32, 64, 16], \n","                      \"collate_fn\": [collate_fn_persistence_diagrams]}\n","models_hyperparams = {\n","    \"num_attention_layers\": [2, 6, 1],\n","    \"num_attention_heads\": [8, 16, 8],\n","}\n","\n","# starting the HPO\n","search.start(\n","    [Adam],\n","    3,\n","    False,\n","    optimizers_params,\n","    dataloaders_params,\n","    models_hyperparams,\n","    n_accumulated_grads=2,\n",")"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":[]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":[""]}],"nbformat":4,"nbformat_minor":2,"metadata":{"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},"orig_nbformat":4}}