File size: 15,736 Bytes
0b9dd2a
 
9eb309a
 
0b9dd2a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9eb309a
0b9dd2a
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
import fasttext
import streamlit as st
import numpy as np
import pandas as pd
from gensim.models import Word2Vec
from sklearn.metrics.pairwise import cosine_similarity
import plotly.express as px
import plotly.graph_objects as go
from collections import Counter
import os
import glob


class UnifiedVectorModel:
    def __init__(self, backend_model, model_type="w2v"):
        self.model = backend_model
        self.model_type = model_type.lower()

        if self.model_type == "w2v":
            self.wv = backend_model.wv
            self.key_to_index = self.wv.key_to_index
            self.vector_size = self.wv.vector_size
            self._words = set(self.wv.key_to_index.keys())

        elif self.model_type == "ft":
            # Для fasttext-wheel
            self.key_to_index = {word: i for i, word in enumerate(backend_model.get_words())}
            self.vector_size = backend_model.get_dimension()
            self._words = set(self.key_to_index.keys())
        else:
            raise ValueError("model_type must be 'w2v' or 'ft'")

    def __contains__(self, word):
        return word in self._words

    def __getitem__(self, word):
        if self.model_type == "w2v":
            return self.wv[word]
        elif self.model_type == "ft":
            return self.model.get_word_vector(word)

    def most_similar(self, positive=None, negative=None, topn=10):
        from sklearn.metrics.pairwise import cosine_similarity

        if not positive:
            positive = []
        if not negative:
            negative = []

        try:
            if self.model_type == "w2v":
                return self.wv.most_similar(positive=positive, negative=negative, topn=topn)

            elif self.model_type == "ft":
                vec = np.zeros(self.vector_size)
                for w in positive:
                    if w in self:
                        vec += self[w]
                    else:
                        continue
                for w in negative:
                    if w in self:
                        vec -= self[w]
                    else:
                        continue

                if np.allclose(vec, 0):
                    return []

                words = list(self._words)
                vectors = np.array([self[w] for w in words])

                sims = cosine_similarity([vec], vectors)[0]
                best = np.argsort(sims)[::-1][:topn + len(positive) + len(negative)]

                result = []
                for i in best:
                    word = words[i]
                    if word not in positive and word not in negative:
                        result.append((word, float(sims[i])))
                        if len(result) >= topn:
                            break
                return result

        except Exception as e:
            print(f"Error in most_similar: {e}")
            return []

    def similar_by_vector(self, vector, topn=10):
        from sklearn.metrics.pairwise import cosine_similarity

        words = list(self._words)
        vectors = np.array([self[w] for w in words])
        sims = cosine_similarity([vector], vectors)[0]
        best = np.argsort(sims)[::-1][:topn]

        return [(words[i], float(sims[i])) for i in best]

    def get_words(self):
        return list(self._words)

    @property
    def vectors(self):
        if not hasattr(self, '_cached_vectors'):
            words = list(self._words)
            self._cached_words = words
            self._cached_vectors = np.array([self[w] for w in words])
        return self._cached_vectors

    @property
    def index_to_key(self):
        if not hasattr(self, '_index_to_key'):
            self._index_to_key = list(self._words)
        return self._index_to_key


@st.cache_resource
def load_model(model_path):
    try:
        if model_path.endswith(".model"):
            raw_model = Word2Vec.load(model_path)
            current_model = UnifiedVectorModel(raw_model, model_type="w2v")

        elif model_path.endswith(".bin"):
            raw_model = fasttext.load_model(model_path)
            current_model = UnifiedVectorModel(raw_model, model_type="ft")
        else:
            raise ValueError(f"wrong path format")
        return current_model
    except Exception as e:
        st.error(f"error loading model {model_path}: {e}")
        return None


MODELS_DIR = "models"

if not os.path.exists(MODELS_DIR):
    st.error(f"Folder `{MODELS_DIR}` not found.")
    st.stop()

model_files = []
for ext in ["*.bin", "*.model", "*.vec"]:
    model_files.extend(glob.glob(os.path.join(MODELS_DIR, ext)))
model_files = [f for f in model_files if os.path.isfile(f)]
model_names = [os.path.basename(f) for f in model_files]

if len(model_names) == 0:
    st.error(f"No models in folder `{MODELS_DIR}` (.bin, .model, .vec).")
    st.info("Supported formats: Word2Vec (binary/text), FastText.")
    st.stop()

selected_model_name = st.sidebar.selectbox(
    "Choose pretrained model",
    model_names
)

selected_model_path = os.path.join(MODELS_DIR, selected_model_name)

st.sidebar.info(f"loading: `{selected_model_name}`")

model = load_model(selected_model_path)

if model is None:
    st.stop()
else:
    st.sidebar.success(f"Model '{selected_model_name}' loaded")
    st.sidebar.write(f"Voc size: {len(model.key_to_index):,}")
    st.sidebar.write(f"Vector size: {model.vector_size}")

def analogy_accuracy(model, file_name):
    right = 0
    count = 0
    results = []
    with open(file_name, encoding='utf-8') as file:
        for line in file:
            words = line.strip().split()
            if len(words) != 4:
                continue
            try:
                most_similar = model.most_similar(positive=[words[0], words[2]], negative=[words[1]], topn=10)
                predicted = [x[0] for x in most_similar]
                correct = words[3]
                if correct in predicted:
                    rank = predicted.index(correct) + 1
                    right += 1
                else:
                    rank = None
                count += 1
                results.append({
                    "query": f"{words[0]} - {words[1]} + {words[2]}",
                    "target": correct,
                    "predicted": predicted[0],
                    "rank": rank,
                    "in_top10": bool(rank)
                })
            except KeyError as e:
                continue
    accuracy = right / count if count > 0 else 0
    return accuracy, results


def avg_similarity(model, file_name):
    res = []
    with open(file_name, encoding='utf-8') as file:
        for line in file:
            words = line.strip().split()
            try:
                vectors = [model[word] for word in words]
            except KeyError:
                continue
            sims = cosine_similarity(vectors)
            for i in range(len(words) - 1):
                for j in range(i + 1, len(words)):
                    res.append(sims[i][j])
    return sum(res) / len(res) if res else 0


def projection(word_vec, axis):
    axis_norm = axis / np.linalg.norm(axis)
    return np.dot(word_vec, axis_norm)


def get_projection_row(model, axis):
    words = list(model.key_to_index.keys())
    projections = [(word, projection(model[word], axis)) for word in words]
    projections = sorted(projections, key=lambda x: x[1])
    return projections


st.title("Vector embeddings")

tab1, tab2, tab3, tab4, tab5 = st.tabs([
    "Vector ariphmetics",
    "Semantic consistency",
    "Semantic axis",
    "Distribution analysis",
    "Report"
])

with tab1:
    st.header("Vector ariphmetics")
    expr = st.text_input("Insert expression", value="рубль - россия + сша")

    if st.button("Compute"):
        words = expr.replace('+', ' + ').replace('-', ' - ').split()
        positive, negative = [], []
        current = 'pos'

        for w in words:
            if w == '+':
                current = 'pos'
            elif w == '-':
                current = 'neg'
            else:
                (positive if current == 'pos' else negative).append(w)

        missing = [w for w in positive + negative if w not in model]
        if missing:
            st.warning(f"Words not found in voc: {', '.join(missing)}")
            st.stop()

        try:
            similar = model.most_similar(
                positive=positive,
                negative=negative,
                topn=10
            )

            st.write("### Result:")
            result_words = [f"{w} ({s:.3f})" for w, s in similar]
            st.write("Nearest words: " + ", ".join(result_words))

            st.write("### In-between steps")

            cum_vec = np.zeros(model.vector_size)

            steps_data = []

            for i in range(len(positive)):
                cum_vec += model[w]
                nearest = model.most_similar(positive=positive[:i + 1], topn=1)
                steps_data.append({
                    "step": f"+ {positive[i]}",
                    "nearest word": nearest[0][0],
                    "similarity": nearest[0][1]
                })

            for i in range(len(negative)):
                cum_vec -= model[w]
                nearest = model.most_similar(positive=positive, negative=negative[:i + 1], topn=1)
                steps_data.append({
                    "step": f"- {negative[i]}",
                    "nearest word": nearest[0][0],
                    "similarity": nearest[0][1]
                })

            df_steps = pd.DataFrame(steps_data)
            st.dataframe(df_steps[["step", "nearest word", "similarity"]])

            result_word = similar[0][0]
            fig = px.scatter(
                x=[cum_vec[0]], y=[cum_vec[1]],
                text=[result_word],
                title="Result (first 2 components)"
            )
            fig.update_traces(textposition='top center', marker=dict(size=12, color='red'))
            st.plotly_chart(fig)

        except Exception as e:
            st.error(f"Error computing: {e}")

with tab2:
    st.header("Similarity calculator")
    col1, col2 = st.columns(2)
    with col1:
        word1 = st.text_input("word 1", value="мужчина")
    with col2:
        word2 = st.text_input("word 2", value="женщина")

    if st.button("Compute similarity"):
        try:
            v1, v2 = model[word1], model[word2]
            sim = cosine_similarity([v1], [v2])[0][0]
            st.metric("Cosine similarity", f"{sim:.4f}")

            st.write("### Nearest neighbors graph")
            neighbors = model.most_similar(word1, topn=5) + model.most_similar(word2, topn=5)
            nodes = list(set([word1, word2] + [n[0] for n in neighbors]))
            edges = [(word1, n[0]) for n in model.most_similar(word1, topn=5)] + \
                    [(word2, n[0]) for n in model.most_similar(word2, topn=5)]

            G = go.Figure()
            pos = np.random.rand(len(nodes), 2) * 2 - 1
            node_x = pos[:, 0]
            node_y = pos[:, 1]

            for edge in edges:
                x0, y0 = pos[nodes.index(edge[0])]
                x1, y1 = pos[nodes.index(edge[1])]
                G.add_trace(go.Scatter(x=[x0, x1], y=[y0, y1], mode='lines', line=dict(width=1, color='gray'), showlegend=False))

            G.add_trace(go.Scatter(x=node_x, y=node_y, mode='text+markers',
                                   marker=dict(size=10, color='lightblue'),
                                   text=nodes, textposition="top center"))
            G.update_layout(title="Semantic links graph", showlegend=False)
            st.plotly_chart(G)

        except KeyError as e:
            st.error(f"Word not found: {e}")

with tab3:
    st.header("Semantic axis projection")
    col1, col2 = st.columns(2)
    with col1:
        pos_axis = st.text_input("positive", value="мужчина")
    with col2:
        neg_axis = st.text_input("negative", value="женщина")

    if st.button("Build axis"):
        try:
            pos_vec = model[pos_axis]
            neg_vec = model[neg_axis]
            axis = pos_vec - neg_vec

            projections = get_projection_row(model, axis)
            top_pos = projections[-10:][::-1]
            top_neg = projections[:10]

            st.write(f"Axis: **{pos_axis}{neg_axis}**")
            st.write("### Top 10 positive:")
            st.write(", ".join([f"{w} ({p:.3f})" for w, p in top_pos]))

            st.write("### Top 10 negative:")
            st.write(", ".join([f"{w} ({p:.3f})" for w, p in top_neg]))

            df_proj = pd.DataFrame(top_pos + top_neg, columns=["word", "projection"])
            fig = px.bar(df_proj, x="projection", y="word", orientation='h', title=f"Projection on axis: {pos_axis}{neg_axis}")
            st.plotly_chart(fig)

        except KeyError as e:
            st.error(f"Error: {e}")

with tab4:
    st.header("Distance distribution analysis")
    all_vectors = model.vectors
    sample = all_vectors[np.random.choice(all_vectors.shape[0], 1000, replace=False)]

    dists = cosine_similarity(sample)
    np.fill_diagonal(dists, 0)
    flat_dists = dists.flatten()
    flat_dists = flat_dists[flat_dists > 0]

    fig = px.histogram(flat_dists, nbins=50, title="Cosine similarity distribution between random words")
    st.plotly_chart(fig)

    st.metric("Mean similarity", f"{np.mean(flat_dists):.3f}")
    st.metric("Std deviation", f"{np.std(flat_dists):.3f}")

with tab5:
    st.header("Report")

    st.subheader("1. Analogy rate")
    analogies_file = "data/analogy.txt"
    if os.path.exists(analogies_file):
        acc, results = analogy_accuracy(model, analogies_file)
        st.metric("Analogy accuracy (in top 10)", f"{acc:.2%}")
        st.dataframe(pd.DataFrame(results))
    else:
        st.warning("File `analogy.txt` not found.")

    st.subheader("2. Average synonyms similarity")
    sim_file = "data/synonyms.txt"
    if os.path.exists(sim_file):
        avg_sim = avg_similarity(model, sim_file)
        st.metric("Average similarity", f"{avg_sim:.4f}")
    else:
        st.warning("File `similarity_words.txt` not found.")

    st.subheader("3. Average antonyms similarity")
    sim_file = "data/antonyms.txt"
    if os.path.exists(sim_file):
        avg_sim = avg_similarity(model, sim_file)
        st.metric("Average similarity", f"{avg_sim:.4f}")
    else:
        st.warning("File `similarity_words.txt` not found.")

    st.subheader("4. Heatmap for nearest words")
    query_words = st.text_input("Enter words", value="мужчина женщина мальчик девочка").split()
    if st.button("Build heatmap"):
        try:
            vectors = [model[w] for w in query_words]
            sims = cosine_similarity(vectors)
            fig = px.imshow(sims, x=query_words, y=query_words, color_continuous_scale="Blues", title="Similarity heatmap")
            st.plotly_chart(fig)
        except KeyError as e:
            st.error(f"Error: {e}")

    st.subheader("5. 2D projection")
    sample_words = st.text_input("Input words", value="мужчина женщина мальчик девочка")
    word_list = sample_words.split()
    if st.button("Show clusters"):
        try:
            from sklearn.manifold import TSNE
            vectors = np.array([model[w] for w in word_list])
            tsne = TSNE(n_components=2, perplexity=len(vectors) - 1, random_state=42)
            embedded = tsne.fit_transform(vectors)

            fig = px.scatter(x=embedded[:, 0], y=embedded[:, 1], text=word_list, title="words projection")
            fig.update_traces(textposition='top center')
            st.plotly_chart(fig)
        except KeyError as e:
            st.error(f"Word not found: {e}")