annif.backend.nn_ensemble.NNEnsembleBackend._train() - Code Metrics - NatLibFi/Annif - Measure and Improve Code Quality continuously with Scrutinizer

NNEnsembleBackend._train() A
last analyzed 2026-01-15 15:18 UTC

↳ Parent: annif.backend.nn_ensemble

Complexity

Conditions

Size

Total Lines	13
Code Lines	12

Duplication

Lines	0
Ratio	0 %

Importance

Changes

Metric	Value
cc	1
eloc	12
nop	4
dl	0
loc	13
rs	9.8
c	0
b	0
f	0

"""Neural network based ensemble backend that combines results from multiple
projects."""

from __future__ import annotations

import os.path
import shutil
import sys
from io import BytesIO
from typing import TYPE_CHECKING, Any

import joblib
import lmdb
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
from scipy.sparse import csc_matrix, csr_matrix
from torch.utils.data import DataLoader, Dataset

import annif.corpus
import annif.parallel
import annif.util
from annif.exception import (
    NotInitializedException,
    NotSupportedException,
    OperationFailedException,
)
from annif.suggestion import SuggestionBatch, vector_to_suggestions

from . import backend, ensemble

if TYPE_CHECKING:
    from annif.corpus.document import DocumentCorpus

logger = annif.logger


def idx_to_key(idx: int) -> bytes:
    """convert an integer index to a binary key for use in LMDB"""
    return b"%08d" % idx


def key_to_idx(key: memoryview | bytes) -> int:
    """convert a binary LMDB key to an integer index"""
    return int(key)


class LMDBDataset(Dataset):
    """A sequence of samples stored in a LMDB database."""

    def __init__(self, txn):
        super().__init__()
        self._txn = txn
        cursor = txn.cursor()
        if cursor.last():
            # Counter holds the number of samples in the database
            self._counter = key_to_idx(cursor.key()) + 1
        else:  # empty database
            self._counter = 0

    def add_sample(self, inputs: np.ndarray, targets: np.ndarray) -> None:
        # use zero-padded 8-digit key
        key = idx_to_key(self._counter)
        self._counter += 1
        # convert the sample into a sparse matrix and serialize it as bytes
        sample = (csc_matrix(inputs), csr_matrix(targets))
        buf = BytesIO()
        joblib.dump(sample, buf)
        buf.seek(0)
        self._txn.put(key, buf.read())

    def __getitem__(self, idx: int) -> tuple[np.ndarray, np.ndarray]:
        """get a particular sample"""
        cursor = self._txn.cursor()
        cursor.set_key(idx_to_key(idx))
        value = cursor.value()
        input_csr, target_csr = joblib.load(BytesIO(value))
        input_tensor = torch.from_numpy(input_csr.toarray())
        target_tensor = torch.from_numpy(target_csr.toarray()[0]).float()
        return input_tensor, target_tensor

    def __len__(self) -> int:
        """return the number of available samples"""
        return self._counter


class NNEnsembleModel(nn.Module):
    def __init__(
        self, input_dim: int, hidden_dim: int, output_dim: int, dropout_rate: float
    ):
        super().__init__()
        self.model_config = {
            "input_dim": input_dim,
            "hidden_dim": hidden_dim,
            "output_dim": output_dim,
            "dropout_rate": dropout_rate,
        }
        self.flatten = nn.Flatten()
        self.dropout1 = nn.Dropout(dropout_rate)
        self.hidden = nn.Linear(input_dim, hidden_dim)
        self.dropout2 = nn.Dropout(dropout_rate)
        self.delta_layer = nn.Linear(hidden_dim, output_dim)

    def forward(self, inputs):
        mean = torch.mean(inputs, dim=1)
        x = self.flatten(inputs)
        x = self.dropout1(x)
        x = F.relu(self.hidden(x))
        x = self.dropout2(x)
        delta = self.delta_layer(x)
        return mean + delta

    def save(self, filepath):
        torch.save(
            {
                "model_state_dict": self.state_dict(),
                "model_class": self.__class__.__name__,
                "model_config": self.model_config,
                "pytorch_version": str(torch.__version__),
                "python_version": sys.version,
            },
            filepath,
        )

    @classmethod
    def load(cls, filepath, map_location="cpu"):
        checkpoint = torch.load(filepath, map_location=map_location, weights_only=True)
        config = checkpoint["model_config"]
        model = cls(**config)
        model.load_state_dict(checkpoint["model_state_dict"])
        model.eval()
        return model


class NNEnsembleBackend(backend.AnnifLearningBackend, ensemble.BaseEnsembleBackend):
    """Neural network ensemble backend that combines results from multiple
    projects"""

    name = "nn_ensemble"

    MODEL_FILE = "nn-model.pt"
    LMDB_FILE = "nn-train.mdb"

    DEFAULT_PARAMETERS = {
        "nodes": 100,
        "dropout_rate": 0.2,
        "optimizer": "adam",
        "lr": 0.001,
        "epochs": 10,
        "learn-epochs": 1,
        "lmdb_map_size": 1024 * 1024 * 1024,
    }

    # defaults for uninitialized instances
    _model = None

    def initialize(self, parallel: bool = False) -> None:
        super().initialize(parallel)
        if self._model is not None:
            return  # already initialized
        if parallel:
            # Don't load model just before parallel execution,
            # since it won't work after forking worker processes
            return
        model_filename = os.path.join(self.datadir, self.MODEL_FILE)
        if not os.path.exists(model_filename):
            raise NotInitializedException(
                "model file {} not found".format(model_filename),
                backend_id=self.backend_id,
            )
        self.debug("loading model from {}".format(model_filename))
        try:
            self._model = NNEnsembleModel.load(model_filename)
        except Exception as err:
            message = (
                f"loading model from {model_filename}; "
                f'original error message: "{err}"'
            )
            raise OperationFailedException(message, backend_id=self.backend_id)

    def _merge_source_batches(
        self,
        batch_by_source: dict[str, SuggestionBatch],
        sources: list[tuple[str, float]],
        params: dict[str, Any],
    ) -> SuggestionBatch:
        src_weight = dict(sources)
        score_vectors = np.array(
            [
                [
                    np.sqrt(suggestions.as_vector())
                    * src_weight[project_id]
                    * len(batch_by_source)
                    for suggestions in batch
                ]
                for project_id, batch in batch_by_source.items()
            ],
            dtype=np.float32,
        )
        score_vector_tensor = torch.from_numpy(score_vectors.swapaxes(0, 1))
        with torch.no_grad():
            prediction = self._model(score_vector_tensor)
        return SuggestionBatch.from_sequence(
            [
                vector_to_suggestions(row, limit=int(params["limit"]))
                for row in prediction
            ],
            self.project.subjects,
        )

    def _create_model(self, sources: list[tuple[str, float]]) -> None:
        self.info("creating NN ensemble model")

        # Create PyTorch model
        input_dim = len(self.project.subjects) * len(sources)
        hidden_dim = int(self.params["nodes"])
        output_dim = len(self.project.subjects)
        dropout_rate = float(self.params["dropout_rate"])

        self._model = NNEnsembleModel(
            input_dim=input_dim,
            hidden_dim=hidden_dim,
            output_dim=output_dim,
            dropout_rate=dropout_rate,
        )

    #        summary = []
    #        self._model.summary(print_fn=summary.append)
    #        self.debug("Created model: \n" + "\n".join(summary))

    def _train(
        self,
        corpus: DocumentCorpus,
        params: dict[str, Any],
        jobs: int = 0,
    ) -> None:
        sources = annif.util.parse_sources(self.params["sources"])
        self._create_model(sources)
        self._fit_model(
            corpus,
            epochs=int(params["epochs"]),
            lmdb_map_size=int(params["lmdb_map_size"]),
            n_jobs=jobs,
        )

    def _corpus_to_vectors(
        self,
        corpus: DocumentCorpus,
        seq: LMDBDataset,
        n_jobs: int,
    ) -> None:
        # pass corpus through all source projects
        sources = dict(annif.util.parse_sources(self.params["sources"]))

        # initialize the source projects before forking, to save memory
        self.info(f"Initializing source projects: {', '.join(sources.keys())}")
        for project_id in sources.keys():
            project = self.project.registry.get_project(project_id)
            project.initialize(parallel=True)

        psmap = annif.parallel.ProjectSuggestMap(
            self.project.registry,
            list(sources.keys()),
            backend_params=None,
            limit=None,
            threshold=0.0,
        )

        jobs, pool_class = annif.parallel.get_pool(n_jobs)

        self.info("Processing training documents...")
        with pool_class(jobs) as pool:
            for hits, subject_set in pool.imap_unordered(
                psmap.suggest, corpus.documents
            ):
                doc_scores = []
                for project_id, p_hits in hits.items():
                    vector = p_hits.as_vector()
                    doc_scores.append(
                        np.sqrt(vector) * sources[project_id] * len(sources)
                    )
                score_vector = np.array(doc_scores, dtype=np.float32)
                true_vector = subject_set.as_vector(len(self.project.subjects))
                seq.add_sample(score_vector, true_vector)

    def _open_lmdb(self, cached, lmdb_map_size):
        lmdb_path = os.path.join(self.datadir, self.LMDB_FILE)
        if not cached and os.path.exists(lmdb_path):
            shutil.rmtree(lmdb_path)
        return lmdb.open(lmdb_path, map_size=lmdb_map_size, writemap=True, mode=0o775)

    def _fit_model(
        self,
        corpus: DocumentCorpus,
        epochs: int,
        lmdb_map_size: int,
        n_jobs: int = 1,
    ) -> None:
        env = self._open_lmdb(corpus == "cached", lmdb_map_size)
        if corpus != "cached":
            if corpus.is_empty():
                raise NotSupportedException(
                    "Cannot train nn_ensemble project with no documents"
                )
            with env.begin(write=True, buffers=True) as txn:
                seq = LMDBDataset(txn)
                self._corpus_to_vectors(corpus, seq, n_jobs)
        else:
            self.info("Reusing cached training data from previous run.")

        # fit the model using a read-only view of the LMDB
        self.info("Training neural network model...")
        with env.begin(buffers=True) as txn:
            dataset = LMDBDataset(txn)
            dataloader = DataLoader(dataset, batch_size=32, shuffle=True)

            # Training loop
            optimizer = torch.optim.Adam(
                self._model.parameters(), lr=float(self.params["lr"])
            )
            criterion = nn.BCEWithLogitsLoss()

            self._model.train()
            for epoch in range(epochs):
                for inputs, targets in dataloader:
                    optimizer.zero_grad()
                    outputs = self._model(inputs)
                    loss = criterion(outputs, targets)
                    loss.backward()
                    optimizer.step()

        annif.util.atomic_save(self._model, self.datadir, self.MODEL_FILE)

    def _learn(
        self,
        corpus: DocumentCorpus,
        params: dict[str, Any],
    ) -> None:
        self.initialize()
        self._fit_model(
            corpus, int(params["learn-epochs"]), int(params["lmdb_map_size"])
        )


1			"""Neural network based ensemble backend that combines results from multiple
2			projects."""
3
4			from __future__ import annotations
5
6			import os.path
7			import shutil
8			import sys
9			from io import BytesIO
10			from typing import TYPE_CHECKING, Any
11
12			import joblib
13			import lmdb
14			import numpy as np
15			import torch
16			import torch.nn as nn
17			import torch.nn.functional as F
18			from scipy.sparse import csc_matrix, csr_matrix
19			from torch.utils.data import DataLoader, Dataset
20
21			import annif.corpus
22			import annif.parallel
23			import annif.util
24			from annif.exception import (
25			NotInitializedException,
26			NotSupportedException,
27			OperationFailedException,
28			)
29			from annif.suggestion import SuggestionBatch, vector_to_suggestions
30
31			from . import backend, ensemble
32
33			if TYPE_CHECKING:
34			from annif.corpus.document import DocumentCorpus
35
36			logger = annif.logger
37
38
39			def idx_to_key(idx: int) -> bytes:
40			"""convert an integer index to a binary key for use in LMDB"""
41			return b"%08d" % idx
42
43
44			def key_to_idx(key: memoryview \| bytes) -> int:
45			"""convert a binary LMDB key to an integer index"""
46			return int(key)
47
48
49			class LMDBDataset(Dataset):
50			"""A sequence of samples stored in a LMDB database."""
51
52			def __init__(self, txn):
53			super().__init__()
54			self._txn = txn
55			cursor = txn.cursor()
56			if cursor.last():
57			# Counter holds the number of samples in the database
58			self._counter = key_to_idx(cursor.key()) + 1
59			else: # empty database
60			self._counter = 0
61
62			def add_sample(self, inputs: np.ndarray, targets: np.ndarray) -> None:
63			# use zero-padded 8-digit key
64			key = idx_to_key(self._counter)
65			self._counter += 1
66			# convert the sample into a sparse matrix and serialize it as bytes
67			sample = (csc_matrix(inputs), csr_matrix(targets))
68			buf = BytesIO()
69			joblib.dump(sample, buf)
70			buf.seek(0)
71			self._txn.put(key, buf.read())
72
73			def __getitem__(self, idx: int) -> tuple[np.ndarray, np.ndarray]:
74			"""get a particular sample"""
75			cursor = self._txn.cursor()
76			cursor.set_key(idx_to_key(idx))
77			value = cursor.value()
78			input_csr, target_csr = joblib.load(BytesIO(value))
79			input_tensor = torch.from_numpy(input_csr.toarray())
80			target_tensor = torch.from_numpy(target_csr.toarray()[0]).float()
81			return input_tensor, target_tensor
82
83			def __len__(self) -> int:
84			"""return the number of available samples"""
85			return self._counter
86
87
88			class NNEnsembleModel(nn.Module):
89			def __init__(
90			self, input_dim: int, hidden_dim: int, output_dim: int, dropout_rate: float
91			):
92			super().__init__()
93			self.model_config = {
94			"input_dim": input_dim,
95			"hidden_dim": hidden_dim,
96			"output_dim": output_dim,
97			"dropout_rate": dropout_rate,
98			}
99			self.flatten = nn.Flatten()
100			self.dropout1 = nn.Dropout(dropout_rate)
101			self.hidden = nn.Linear(input_dim, hidden_dim)
102			self.dropout2 = nn.Dropout(dropout_rate)
103			self.delta_layer = nn.Linear(hidden_dim, output_dim)
104
105			def forward(self, inputs):
106			mean = torch.mean(inputs, dim=1)
107			x = self.flatten(inputs)
108			x = self.dropout1(x)
109			x = F.relu(self.hidden(x))
110			x = self.dropout2(x)
111			delta = self.delta_layer(x)
112			return mean + delta
113
114			def save(self, filepath):
115			torch.save(
116			{
117			"model_state_dict": self.state_dict(),
118			"model_class": self.__class__.__name__,
119			"model_config": self.model_config,
120			"pytorch_version": str(torch.__version__),
121			"python_version": sys.version,
122			},
123			filepath,
124			)
125
126			@classmethod
127			def load(cls, filepath, map_location="cpu"):
128			checkpoint = torch.load(filepath, map_location=map_location, weights_only=True)
129			config = checkpoint["model_config"]
130			model = cls(**config)
131			model.load_state_dict(checkpoint["model_state_dict"])
132			model.eval()
133			return model
134
135
136			class NNEnsembleBackend(backend.AnnifLearningBackend, ensemble.BaseEnsembleBackend):
137			"""Neural network ensemble backend that combines results from multiple
138			projects"""
139
140			name = "nn_ensemble"
141
142			MODEL_FILE = "nn-model.pt"
143			LMDB_FILE = "nn-train.mdb"
144
145			DEFAULT_PARAMETERS = {
146			"nodes": 100,
147			"dropout_rate": 0.2,
148			"optimizer": "adam",
149			"lr": 0.001,
150			"epochs": 10,
151			"learn-epochs": 1,
152			"lmdb_map_size": 1024 * 1024 * 1024,
153			}
154
155			# defaults for uninitialized instances
156			_model = None
157
158			def initialize(self, parallel: bool = False) -> None:
159			super().initialize(parallel)
160			if self._model is not None:
161			return # already initialized
162			if parallel:
163			# Don't load model just before parallel execution,
164			# since it won't work after forking worker processes
165			return
166			model_filename = os.path.join(self.datadir, self.MODEL_FILE)
167			if not os.path.exists(model_filename):
168			raise NotInitializedException(
169			"model file {} not found".format(model_filename),
170			backend_id=self.backend_id,
171			)
172			self.debug("loading model from {}".format(model_filename))
173			try:
174			self._model = NNEnsembleModel.load(model_filename)
175			except Exception as err:
176			message = (
177			f"loading model from {model_filename}; "
178			f'original error message: "{err}"'
179			)
180			raise OperationFailedException(message, backend_id=self.backend_id)
181
182			def _merge_source_batches(
183			self,
184			batch_by_source: dict[str, SuggestionBatch],
185			sources: list[tuple[str, float]],
186			params: dict[str, Any],
187			) -> SuggestionBatch:
188			src_weight = dict(sources)
189			score_vectors = np.array(
190			[
191			[
192			np.sqrt(suggestions.as_vector())
193			* src_weight[project_id]
194			* len(batch_by_source)
195			for suggestions in batch
196			]
197			for project_id, batch in batch_by_source.items()
198			],
199			dtype=np.float32,
200			)
201			score_vector_tensor = torch.from_numpy(score_vectors.swapaxes(0, 1))
202			with torch.no_grad():
203			prediction = self._model(score_vector_tensor)
204			return SuggestionBatch.from_sequence(
205			[
206			vector_to_suggestions(row, limit=int(params["limit"]))
207			for row in prediction
208			],
209			self.project.subjects,
210			)
211
212			def _create_model(self, sources: list[tuple[str, float]]) -> None:
213			self.info("creating NN ensemble model")
214
215			# Create PyTorch model
216			input_dim = len(self.project.subjects) * len(sources)
217			hidden_dim = int(self.params["nodes"])
218			output_dim = len(self.project.subjects)
219			dropout_rate = float(self.params["dropout_rate"])
220
221			self._model = NNEnsembleModel(
222			input_dim=input_dim,
223			hidden_dim=hidden_dim,
224			output_dim=output_dim,
225			dropout_rate=dropout_rate,
226			)
227
228			# summary = []
229			# self._model.summary(print_fn=summary.append)
230			# self.debug("Created model: \n" + "\n".join(summary))
231
232			def _train(
233			self,
234			corpus: DocumentCorpus,
235			params: dict[str, Any],
236			jobs: int = 0,
237			) -> None:
238			sources = annif.util.parse_sources(self.params["sources"])
239			self._create_model(sources)
240			self._fit_model(
241			corpus,
242			epochs=int(params["epochs"]),
243			lmdb_map_size=int(params["lmdb_map_size"]),
244			n_jobs=jobs,
245			)
246
247			def _corpus_to_vectors(
248			self,
249			corpus: DocumentCorpus,
250			seq: LMDBDataset,
251			n_jobs: int,
252			) -> None:
253			# pass corpus through all source projects
254			sources = dict(annif.util.parse_sources(self.params["sources"]))
255
256			# initialize the source projects before forking, to save memory
257			self.info(f"Initializing source projects: {', '.join(sources.keys())}")
258			for project_id in sources.keys():
259			project = self.project.registry.get_project(project_id)
260			project.initialize(parallel=True)
261
262			psmap = annif.parallel.ProjectSuggestMap(
263			self.project.registry,
264			list(sources.keys()),
265			backend_params=None,
266			limit=None,
267			threshold=0.0,
268			)
269
270			jobs, pool_class = annif.parallel.get_pool(n_jobs)
271
272			self.info("Processing training documents...")
273			with pool_class(jobs) as pool:
274			for hits, subject_set in pool.imap_unordered(
275			psmap.suggest, corpus.documents
276			):
277			doc_scores = []
278			for project_id, p_hits in hits.items():
279			vector = p_hits.as_vector()
280			doc_scores.append(
281			np.sqrt(vector) * sources[project_id] * len(sources)
282			)
283			score_vector = np.array(doc_scores, dtype=np.float32)
284			true_vector = subject_set.as_vector(len(self.project.subjects))
285			seq.add_sample(score_vector, true_vector)
286
287			def _open_lmdb(self, cached, lmdb_map_size):
288			lmdb_path = os.path.join(self.datadir, self.LMDB_FILE)
289			if not cached and os.path.exists(lmdb_path):
290			shutil.rmtree(lmdb_path)
291			return lmdb.open(lmdb_path, map_size=lmdb_map_size, writemap=True, mode=0o775)
292
293			def _fit_model(
294			self,
295			corpus: DocumentCorpus,
296			epochs: int,
297			lmdb_map_size: int,
298			n_jobs: int = 1,
299			) -> None:
300			env = self._open_lmdb(corpus == "cached", lmdb_map_size)
301			if corpus != "cached":
302			if corpus.is_empty():
303			raise NotSupportedException(
304			"Cannot train nn_ensemble project with no documents"
305			)
306			with env.begin(write=True, buffers=True) as txn:
307			seq = LMDBDataset(txn)
308			self._corpus_to_vectors(corpus, seq, n_jobs)
309			else:
310			self.info("Reusing cached training data from previous run.")
311
312			# fit the model using a read-only view of the LMDB
313			self.info("Training neural network model...")
314			with env.begin(buffers=True) as txn:
315			dataset = LMDBDataset(txn)
316			dataloader = DataLoader(dataset, batch_size=32, shuffle=True)
317
318			# Training loop
319			optimizer = torch.optim.Adam(
320			self._model.parameters(), lr=float(self.params["lr"])
321			)
322			criterion = nn.BCEWithLogitsLoss()
323
324			self._model.train()
325			for epoch in range(epochs):
326			for inputs, targets in dataloader:
327			optimizer.zero_grad()
328			outputs = self._model(inputs)
329			loss = criterion(outputs, targets)
330			loss.backward()
331			optimizer.step()
332
333			annif.util.atomic_save(self._model, self.datadir, self.MODEL_FILE)
334
335			def _learn(
336			self,
337			corpus: DocumentCorpus,
338			params: dict[str, Any],
339			) -> None:
340			self.initialize()
341			self._fit_model(
342			corpus, int(params["learn-epochs"]), int(params["lmdb_map_size"])
343			)
344

NatLibFi / Annif

NNEnsembleBackend._train() A last analyzed 2026-01-15 15:18 UTC

Complexity

Size

Duplication

Importance

Duplication Side-by-Side

Filter issues like

NNEnsembleBackend._train() A
last analyzed 2026-01-15 15:18 UTC