annif.backend.nn_ensemble.NNEnsembleBackend._merge_source_batches() - Code Metrics - Inspection of "refactor ensemble suggest functionality to better..." - NatLibFi/Annif - Measure and Improve Code Quality continuously with Scrutinizer

Passed

Push — issue678-refactor-suggestionre... ( 0d7003...d7e7fa )

by Osma

created 2023-04-06 13:27 UTC

NNEnsembleBackend._merge_source_batches() A

↳ Parent: annif.backend.nn_ensemble

Complexity

Conditions

Size

Total Lines	21
Code Lines	18

Duplication

Lines	0
Ratio	0 %

Importance

Changes

Metric	Value
cc	1
eloc	18
nop	4
dl	0
loc	21
rs	9.5
c	0
b	0
f	0

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


import os.path
import shutil
from io import BytesIO

import joblib
import lmdb
import numpy as np
import tensorflow.keras.backend as K
from scipy.sparse import csc_matrix, csr_matrix
from tensorflow.keras.layers import Add, Dense, Dropout, Flatten, Input, Layer
from tensorflow.keras.models import Model, load_model
from tensorflow.keras.utils import Sequence

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

from . import backend, ensemble


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


def key_to_idx(key):
    """convert a binary LMDB key to an integer index"""
    return int(key)


class LMDBSequence(Sequence):
    """A sequence of samples stored in a LMDB database."""

    def __init__(self, txn, batch_size):
        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
        self._batch_size = batch_size

    def add_sample(self, inputs, targets):
        # 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):
        """get a particular batch of samples"""
        cursor = self._txn.cursor()
        first_key = idx * self._batch_size
        cursor.set_key(idx_to_key(first_key))
        input_arrays = []
        target_arrays = []
        for key, value in cursor.iternext():
            if key_to_idx(key) >= (first_key + self._batch_size):
                break
            input_csr, target_csr = joblib.load(BytesIO(value))
            input_arrays.append(input_csr.toarray())
            target_arrays.append(target_csr.toarray().flatten())
        return np.array(input_arrays), np.array(target_arrays)

    def __len__(self):
        """return the number of available batches"""
        return int(np.ceil(self._counter / self._batch_size))


class MeanLayer(Layer):
    """Custom Keras layer that calculates mean values along the 2nd axis."""

    def call(self, inputs):
        return K.mean(inputs, axis=2)


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

    name = "nn_ensemble"

    MODEL_FILE = "nn-model.h5"
    LMDB_FILE = "nn-train.mdb"

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

    # defaults for uninitialized instances
    _model = None

    def default_params(self):
        params = backend.AnnifBackend.DEFAULT_PARAMETERS.copy()
        params.update(self.DEFAULT_PARAMETERS)
        return params

    def initialize(self, parallel=False):
        super().initialize(parallel)
        if self._model is not None:
            return  # already initialized
        if parallel:
            # Don't load TF 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 Keras model from {}".format(model_filename))
        self._model = load_model(
            model_filename, custom_objects={"MeanLayer": MeanLayer}
        )

    def _merge_source_batches(self, batch_by_source, sources, params):
        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,
        ).transpose(1, 2, 0)
        prediction = self._model(score_vectors).numpy()
        return SuggestionBatch.from_sequence(
            [
                vector_to_suggestions(row, limit=int(params["limit"]))
                for row in prediction
            ],
            self.project.subjects,
        )

    def _create_model(self, sources):
        self.info("creating NN ensemble model")

        inputs = Input(shape=(len(self.project.subjects), len(sources)))

        flat_input = Flatten()(inputs)
        drop_input = Dropout(rate=float(self.params["dropout_rate"]))(flat_input)
        hidden = Dense(int(self.params["nodes"]), activation="relu")(drop_input)
        drop_hidden = Dropout(rate=float(self.params["dropout_rate"]))(hidden)
        delta = Dense(
            len(self.project.subjects),
            kernel_initializer="zeros",
            bias_initializer="zeros",
        )(drop_hidden)

        mean = MeanLayer()(inputs)

        predictions = Add()([mean, delta])

        self._model = Model(inputs=inputs, outputs=predictions)
        self._model.compile(
            optimizer=self.params["optimizer"],
            loss="binary_crossentropy",
            metrics=["top_k_categorical_accuracy"],
        )
        if "lr" in self.params:
            self._model.optimizer.learning_rate.assign(float(self.params["lr"]))

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

    def _train(self, corpus, params, jobs=0):
        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, seq, n_jobs):
        # 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).transpose()
                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)

    def _fit_model(self, corpus, epochs, lmdb_map_size, n_jobs=1):
        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 = LMDBSequence(txn, batch_size=32)
                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:
            seq = LMDBSequence(txn, batch_size=32)
            self._model.fit(seq, verbose=True, epochs=epochs)

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

    def _learn(self, corpus, params):
        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
5			import os.path
6			import shutil
7			from io import BytesIO
8
9			import joblib
10			import lmdb
11			import numpy as np
12			import tensorflow.keras.backend as K
13			from scipy.sparse import csc_matrix, csr_matrix
14			from tensorflow.keras.layers import Add, Dense, Dropout, Flatten, Input, Layer
15			from tensorflow.keras.models import Model, load_model
16			from tensorflow.keras.utils import Sequence
17
18			import annif.corpus
19			import annif.parallel
20			import annif.util
21			from annif.exception import NotInitializedException, NotSupportedException
22			from annif.suggestion import SuggestionBatch, vector_to_suggestions
23
24			from . import backend, ensemble
25
26
27			def idx_to_key(idx):
28			"""convert an integer index to a binary key for use in LMDB"""
29			return b"%08d" % idx
30
31
32			def key_to_idx(key):
33			"""convert a binary LMDB key to an integer index"""
34			return int(key)
35
36
37			class LMDBSequence(Sequence):
38			"""A sequence of samples stored in a LMDB database."""
39
40			def __init__(self, txn, batch_size):
41			self._txn = txn
42			cursor = txn.cursor()
43			if cursor.last():
44			# Counter holds the number of samples in the database
45			self._counter = key_to_idx(cursor.key()) + 1
46			else: # empty database
47			self._counter = 0
48			self._batch_size = batch_size
49
50			def add_sample(self, inputs, targets):
51			# use zero-padded 8-digit key
52			key = idx_to_key(self._counter)
53			self._counter += 1
54			# convert the sample into a sparse matrix and serialize it as bytes
55			sample = (csc_matrix(inputs), csr_matrix(targets))
56			buf = BytesIO()
57			joblib.dump(sample, buf)
58			buf.seek(0)
59			self._txn.put(key, buf.read())
60
61			def __getitem__(self, idx):
62			"""get a particular batch of samples"""
63			cursor = self._txn.cursor()
64			first_key = idx * self._batch_size
65			cursor.set_key(idx_to_key(first_key))
66			input_arrays = []
67			target_arrays = []
68			for key, value in cursor.iternext():
69			if key_to_idx(key) >= (first_key + self._batch_size):
70			break
71			input_csr, target_csr = joblib.load(BytesIO(value))
72			input_arrays.append(input_csr.toarray())
73			target_arrays.append(target_csr.toarray().flatten())
74			return np.array(input_arrays), np.array(target_arrays)
75
76			def __len__(self):
77			"""return the number of available batches"""
78			return int(np.ceil(self._counter / self._batch_size))
79
80
81			class MeanLayer(Layer):
82			"""Custom Keras layer that calculates mean values along the 2nd axis."""
83
84			def call(self, inputs):
85			return K.mean(inputs, axis=2)
86
87
88			class NNEnsembleBackend(backend.AnnifLearningBackend, ensemble.BaseEnsembleBackend):
89			"""Neural network ensemble backend that combines results from multiple
90			projects"""
91
92			name = "nn_ensemble"
93
94			MODEL_FILE = "nn-model.h5"
95			LMDB_FILE = "nn-train.mdb"
96
97			DEFAULT_PARAMETERS = {
98			"nodes": 100,
99			"dropout_rate": 0.2,
100			"optimizer": "adam",
101			"epochs": 10,
102			"learn-epochs": 1,
103			"lmdb_map_size": 1024 * 1024 * 1024,
104			}
105
106			# defaults for uninitialized instances
107			_model = None
108
109			def default_params(self):
110			params = backend.AnnifBackend.DEFAULT_PARAMETERS.copy()
111			params.update(self.DEFAULT_PARAMETERS)
112			return params
113
114			def initialize(self, parallel=False):
115			super().initialize(parallel)
116			if self._model is not None:
117			return # already initialized
118			if parallel:
119			# Don't load TF model just before parallel execution,
120			# since it won't work after forking worker processes
121			return
122			model_filename = os.path.join(self.datadir, self.MODEL_FILE)
123			if not os.path.exists(model_filename):
124			raise NotInitializedException(
125			"model file {} not found".format(model_filename),
126			backend_id=self.backend_id,
127			)
128			self.debug("loading Keras model from {}".format(model_filename))
129			self._model = load_model(
130			model_filename, custom_objects={"MeanLayer": MeanLayer}
131			)
132
133			def _merge_source_batches(self, batch_by_source, sources, params):
134			src_weight = dict(sources)
135			score_vectors = np.array(
136			[
137			[
138			np.sqrt(suggestions.as_vector())
139			* src_weight[project_id]
140			* len(batch_by_source)
141			for suggestions in batch
142			]
143			for project_id, batch in batch_by_source.items()
144			],
145			dtype=np.float32,
146			).transpose(1, 2, 0)
147			prediction = self._model(score_vectors).numpy()
148			return SuggestionBatch.from_sequence(
149			[
150			vector_to_suggestions(row, limit=int(params["limit"]))
151			for row in prediction
152			],
153			self.project.subjects,
154			)
155
156			def _create_model(self, sources):
157			self.info("creating NN ensemble model")
158
159			inputs = Input(shape=(len(self.project.subjects), len(sources)))
160
161			flat_input = Flatten()(inputs)
162			drop_input = Dropout(rate=float(self.params["dropout_rate"]))(flat_input)
163			hidden = Dense(int(self.params["nodes"]), activation="relu")(drop_input)
164			drop_hidden = Dropout(rate=float(self.params["dropout_rate"]))(hidden)
165			delta = Dense(
166			len(self.project.subjects),
167			kernel_initializer="zeros",
168			bias_initializer="zeros",
169			)(drop_hidden)
170
171			mean = MeanLayer()(inputs)
172
173			predictions = Add()([mean, delta])
174
175			self._model = Model(inputs=inputs, outputs=predictions)
176			self._model.compile(
177			optimizer=self.params["optimizer"],
178			loss="binary_crossentropy",
179			metrics=["top_k_categorical_accuracy"],
180			)
181			if "lr" in self.params:
182			self._model.optimizer.learning_rate.assign(float(self.params["lr"]))
183
184			summary = []
185			self._model.summary(print_fn=summary.append)
186			self.debug("Created model: \n" + "\n".join(summary))
187
188			def _train(self, corpus, params, jobs=0):
189			sources = annif.util.parse_sources(self.params["sources"])
190			self._create_model(sources)
191			self._fit_model(
192			corpus,
193			epochs=int(params["epochs"]),
194			lmdb_map_size=int(params["lmdb_map_size"]),
195			n_jobs=jobs,
196			)
197
198			def _corpus_to_vectors(self, corpus, seq, n_jobs):
199			# pass corpus through all source projects
200			sources = dict(annif.util.parse_sources(self.params["sources"]))
201
202			# initialize the source projects before forking, to save memory
203			self.info(f"Initializing source projects: {', '.join(sources.keys())}")
204			for project_id in sources.keys():
205			project = self.project.registry.get_project(project_id)
206			project.initialize(parallel=True)
207
208			psmap = annif.parallel.ProjectSuggestMap(
209			self.project.registry,
210			list(sources.keys()),
211			backend_params=None,
212			limit=None,
213			threshold=0.0,
214			)
215
216			jobs, pool_class = annif.parallel.get_pool(n_jobs)
217
218			self.info("Processing training documents...")
219			with pool_class(jobs) as pool:
220			for hits, subject_set in pool.imap_unordered(
221			psmap.suggest, corpus.documents
222			):
223			doc_scores = []
224			for project_id, p_hits in hits.items():
225			vector = p_hits.as_vector()
226			doc_scores.append(
227			np.sqrt(vector) * sources[project_id] * len(sources)
228			)
229			score_vector = np.array(doc_scores, dtype=np.float32).transpose()
230			true_vector = subject_set.as_vector(len(self.project.subjects))
231			seq.add_sample(score_vector, true_vector)
232
233			def _open_lmdb(self, cached, lmdb_map_size):
234			lmdb_path = os.path.join(self.datadir, self.LMDB_FILE)
235			if not cached and os.path.exists(lmdb_path):
236			shutil.rmtree(lmdb_path)
237			return lmdb.open(lmdb_path, map_size=lmdb_map_size, writemap=True)
238
239			def _fit_model(self, corpus, epochs, lmdb_map_size, n_jobs=1):
240			env = self._open_lmdb(corpus == "cached", lmdb_map_size)
241			if corpus != "cached":
242			if corpus.is_empty():
243			raise NotSupportedException(
244			"Cannot train nn_ensemble project with no documents"
245			)
246			with env.begin(write=True, buffers=True) as txn:
247			seq = LMDBSequence(txn, batch_size=32)
248			self._corpus_to_vectors(corpus, seq, n_jobs)
249			else:
250			self.info("Reusing cached training data from previous run.")
251			# fit the model using a read-only view of the LMDB
252			self.info("Training neural network model...")
253			with env.begin(buffers=True) as txn:
254			seq = LMDBSequence(txn, batch_size=32)
255			self._model.fit(seq, verbose=True, epochs=epochs)
256
257			annif.util.atomic_save(self._model, self.datadir, self.MODEL_FILE)
258
259			def _learn(self, corpus, params):
260			self.initialize()
261			self._fit_model(
262			corpus, int(params["learn-epochs"]), int(params["lmdb_map_size"])
263			)
264

NatLibFi / Annif

Push — issue678-refactor-suggestionre... ( 0d7003...d7e7fa )

NNEnsembleBackend._merge_source_batches() A

Complexity

Size

Duplication

Importance

Duplication Side-by-Side

Filter issues like