In [1]:
%run init.ipynb
Using TensorFlow backend.
/home/fanyixing/.local/python3/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: numpy.dtype size changed, may indicate binary incompatibility. Expected 96, got 88
return f(*args, **kwds)
/home/fanyixing/.local/python3/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: numpy.dtype size changed, may indicate binary incompatibility. Expected 96, got 88
return f(*args, **kwds)
matchzoo version 2.1.0
data loading ...
data loaded as `train_pack_raw` `dev_pack_raw` `test_pack_raw`
`ranking_task` initialized with metrics [normalized_discounted_cumulative_gain@3(0.0), normalized_discounted_cumulative_gain@5(0.0), mean_average_precision(0.0)]
loading embedding ...
embedding loaded as `glove_embedding`
In [2]:
preprocessor = mz.preprocessors.BasicPreprocessor(fixed_length_left=10, fixed_length_right=40, remove_stop_words=False)
train_pack_processed = preprocessor.fit_transform(train_pack_raw)
valid_pack_processed = preprocessor.transform(dev_pack_raw)
test_pack_processed = preprocessor.transform(test_pack_raw)
Processing text_left with chain_transform of Tokenize => Lowercase => PuncRemoval: 100%|██████████| 2118/2118 [00:00<00:00, 3665.85it/s]
Processing text_right with chain_transform of Tokenize => Lowercase => PuncRemoval: 100%|██████████| 18841/18841 [00:08<00:00, 2124.56it/s]
Processing text_right with append: 100%|██████████| 18841/18841 [00:00<00:00, 312816.21it/s]
Building FrequencyFilter from a datapack.: 100%|██████████| 18841/18841 [00:00<00:00, 58156.99it/s]
Processing text_right with transform: 100%|██████████| 18841/18841 [00:00<00:00, 36334.58it/s]
Processing text_left with extend: 100%|██████████| 2118/2118 [00:00<00:00, 344069.71it/s]
Processing text_right with extend: 100%|██████████| 18841/18841 [00:00<00:00, 301782.94it/s]
Building Vocabulary from a datapack.: 100%|██████████| 404415/404415 [00:00<00:00, 1562426.96it/s]
Processing text_left with chain_transform of Tokenize => Lowercase => PuncRemoval: 100%|██████████| 2118/2118 [00:00<00:00, 3087.94it/s]
Processing text_right with chain_transform of Tokenize => Lowercase => PuncRemoval: 100%|██████████| 18841/18841 [00:10<00:00, 1847.91it/s]
Processing text_right with transform: 100%|██████████| 18841/18841 [00:00<00:00, 55576.41it/s]
Processing text_left with transform: 100%|██████████| 2118/2118 [00:00<00:00, 37457.67it/s]
Processing text_right with transform: 100%|██████████| 18841/18841 [00:00<00:00, 55490.83it/s]
Processing length_left with len: 100%|██████████| 2118/2118 [00:00<00:00, 329263.75it/s]
Processing length_right with len: 100%|██████████| 18841/18841 [00:00<00:00, 396878.61it/s]
Processing text_left with transform: 100%|██████████| 2118/2118 [00:00<00:00, 38067.78it/s]
Processing text_right with transform: 100%|██████████| 18841/18841 [00:00<00:00, 20157.65it/s]
Processing text_left with chain_transform of Tokenize => Lowercase => PuncRemoval: 100%|██████████| 122/122 [00:00<00:00, 2859.98it/s]
Processing text_right with chain_transform of Tokenize => Lowercase => PuncRemoval: 100%|██████████| 1115/1115 [00:00<00:00, 1982.49it/s]
Processing text_right with transform: 100%|██████████| 1115/1115 [00:00<00:00, 38183.90it/s]
Processing text_left with transform: 100%|██████████| 122/122 [00:00<00:00, 29319.03it/s]
Processing text_right with transform: 100%|██████████| 1115/1115 [00:00<00:00, 43965.86it/s]
Processing length_left with len: 100%|██████████| 122/122 [00:00<00:00, 89240.51it/s]
Processing length_right with len: 100%|██████████| 1115/1115 [00:00<00:00, 260871.81it/s]
Processing text_left with transform: 100%|██████████| 122/122 [00:00<00:00, 26225.15it/s]
Processing text_right with transform: 100%|██████████| 1115/1115 [00:00<00:00, 25368.18it/s]
Processing text_left with chain_transform of Tokenize => Lowercase => PuncRemoval: 100%|██████████| 237/237 [00:00<00:00, 1868.81it/s]
Processing text_right with chain_transform of Tokenize => Lowercase => PuncRemoval: 100%|██████████| 2300/2300 [00:01<00:00, 1421.59it/s]
Processing text_right with transform: 100%|██████████| 2300/2300 [00:00<00:00, 25433.49it/s]
Processing text_left with transform: 100%|██████████| 237/237 [00:00<00:00, 30281.48it/s]
Processing text_right with transform: 100%|██████████| 2300/2300 [00:00<00:00, 47353.48it/s]
Processing length_left with len: 100%|██████████| 237/237 [00:00<00:00, 188088.94it/s]
Processing length_right with len: 100%|██████████| 2300/2300 [00:00<00:00, 356197.59it/s]
Processing text_left with transform: 100%|██████████| 237/237 [00:00<00:00, 19210.92it/s]
Processing text_right with transform: 100%|██████████| 2300/2300 [00:00<00:00, 24293.37it/s]
In [3]:
preprocessor.context
Out[3]:
{'filter_unit': <matchzoo.preprocessors.units.frequency_filter.FrequencyFilter at 0x7fd9f4649be0>,
'vocab_unit': <matchzoo.preprocessors.units.vocabulary.Vocabulary at 0x7fd938f312b0>,
'vocab_size': 16674,
'embedding_input_dim': 16674,
'input_shapes': [(10,), (40,)]}
In [4]:
model = mz.models.KNRM()
model.params.update(preprocessor.context)
model.params['task'] = ranking_task
model.params['embedding_output_dim'] = glove_embedding.output_dim
model.params['embedding_trainable'] = True
model.params['kernel_num'] = 21
model.params['sigma'] = 0.1
model.params['exact_sigma'] = 0.001
model.params['optimizer'] = 'adadelta'
model.build()
model.compile()
#model.backend.summary()
In [5]:
embedding_matrix = glove_embedding.build_matrix(preprocessor.context['vocab_unit'].state['term_index'])
model.load_embedding_matrix(embedding_matrix)
In [6]:
pred_x, pred_y = test_pack_processed.unpack()
evaluate = mz.callbacks.EvaluateAllMetrics(model, x=pred_x, y=pred_y, batch_size=len(pred_x))
In [7]:
train_generator = mz.DataGenerator(
train_pack_processed,
mode='pair',
num_dup=5,
num_neg=1,
batch_size=20
)
print('num batches:', len(train_generator))
num batches: 255
In [8]:
history = model.fit_generator(train_generator, epochs=30, callbacks=[evaluate], workers=30, use_multiprocessing=True)
Epoch 1/30
255/255 [==============================] - 30s 117ms/step - loss: 1.0630
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.525888722537362 - normalized_discounted_cumulative_gain@5(0.0): 0.5969076581909297 - mean_average_precision(0.0): 0.5458421758049457
Epoch 2/30
255/255 [==============================] - 40s 158ms/step - loss: 0.4855
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.540316755776057 - normalized_discounted_cumulative_gain@5(0.0): 0.6183376741777723 - mean_average_precision(0.0): 0.5712611133729898
Epoch 3/30
255/255 [==============================] - 39s 155ms/step - loss: 0.3582
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.541616192577938 - normalized_discounted_cumulative_gain@5(0.0): 0.6153837509826408 - mean_average_precision(0.0): 0.567777172721793
Epoch 4/30
255/255 [==============================] - 40s 156ms/step - loss: 0.2887
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5584765720489868 - normalized_discounted_cumulative_gain@5(0.0): 0.6194525103413756 - mean_average_precision(0.0): 0.5778673022579046
Epoch 5/30
255/255 [==============================] - 41s 162ms/step - loss: 0.2118
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5528840743131206 - normalized_discounted_cumulative_gain@5(0.0): 0.6232796348158448 - mean_average_precision(0.0): 0.5763382056568952
Epoch 6/30
255/255 [==============================] - 41s 160ms/step - loss: 0.1721
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5479498178842863 - normalized_discounted_cumulative_gain@5(0.0): 0.6080774335846967 - mean_average_precision(0.0): 0.5628146118070727
Epoch 7/30
255/255 [==============================] - 42s 163ms/step - loss: 0.1351
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5448249760251662 - normalized_discounted_cumulative_gain@5(0.0): 0.6050334611320453 - mean_average_precision(0.0): 0.5539922350259576
Epoch 8/30
255/255 [==============================] - 41s 161ms/step - loss: 0.1056
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5278279251987681 - normalized_discounted_cumulative_gain@5(0.0): 0.5926550462383007 - mean_average_precision(0.0): 0.546292777345115
Epoch 9/30
255/255 [==============================] - 40s 155ms/step - loss: 0.0843
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5260762848714361 - normalized_discounted_cumulative_gain@5(0.0): 0.5997466236266378 - mean_average_precision(0.0): 0.5463433489984417
Epoch 10/30
255/255 [==============================] - 39s 151ms/step - loss: 0.0641
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.53545626407713 - normalized_discounted_cumulative_gain@5(0.0): 0.5911736172377866 - mean_average_precision(0.0): 0.5511742297719537
Epoch 11/30
255/255 [==============================] - 39s 154ms/step - loss: 0.0506
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5281313957863254 - normalized_discounted_cumulative_gain@5(0.0): 0.5909946137693491 - mean_average_precision(0.0): 0.5455621168460892
Epoch 12/30
255/255 [==============================] - 42s 163ms/step - loss: 0.0411
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5295249224151654 - normalized_discounted_cumulative_gain@5(0.0): 0.5882240550416985 - mean_average_precision(0.0): 0.544818346189001
Epoch 13/30
255/255 [==============================] - 41s 162ms/step - loss: 0.0356
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5332953580728098 - normalized_discounted_cumulative_gain@5(0.0): 0.5923864923271813 - mean_average_precision(0.0): 0.5449250708396572
Epoch 14/30
255/255 [==============================] - 42s 163ms/step - loss: 0.0275
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5344059653683887 - normalized_discounted_cumulative_gain@5(0.0): 0.5852285609827376 - mean_average_precision(0.0): 0.5457076553713109
Epoch 15/30
255/255 [==============================] - 42s 163ms/step - loss: 0.0224
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5326101057106915 - normalized_discounted_cumulative_gain@5(0.0): 0.5896572847849786 - mean_average_precision(0.0): 0.5461713501388032
Epoch 16/30
255/255 [==============================] - 40s 155ms/step - loss: 0.0189
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5209623555446414 - normalized_discounted_cumulative_gain@5(0.0): 0.5907051989832439 - mean_average_precision(0.0): 0.544006088275105
Epoch 17/30
255/255 [==============================] - 39s 153ms/step - loss: 0.0154
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5270528188502702 - normalized_discounted_cumulative_gain@5(0.0): 0.5929241490222487 - mean_average_precision(0.0): 0.5458721064851292
Epoch 18/30
255/255 [==============================] - 39s 152ms/step - loss: 0.0134
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.52963998719156 - normalized_discounted_cumulative_gain@5(0.0): 0.59748806088061 - mean_average_precision(0.0): 0.5504225606330331
Epoch 19/30
255/255 [==============================] - 41s 161ms/step - loss: 0.0129
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5173295480303289 - normalized_discounted_cumulative_gain@5(0.0): 0.5890029693924619 - mean_average_precision(0.0): 0.5393698842857736
Epoch 20/30
255/255 [==============================] - 42s 164ms/step - loss: 0.0136
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5319324815526434 - normalized_discounted_cumulative_gain@5(0.0): 0.599965023036905 - mean_average_precision(0.0): 0.550039654928675
Epoch 21/30
255/255 [==============================] - 42s 163ms/step - loss: 0.0108
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5311517097282423 - normalized_discounted_cumulative_gain@5(0.0): 0.5943106559500474 - mean_average_precision(0.0): 0.5487263041947172
Epoch 22/30
255/255 [==============================] - 42s 165ms/step - loss: 0.0093
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5340421861939937 - normalized_discounted_cumulative_gain@5(0.0): 0.6020890805193887 - mean_average_precision(0.0): 0.5541712982869859
Epoch 23/30
255/255 [==============================] - 40s 157ms/step - loss: 0.0063
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5245100179927665 - normalized_discounted_cumulative_gain@5(0.0): 0.5896176258663225 - mean_average_precision(0.0): 0.5427922534821856
Epoch 24/30
255/255 [==============================] - 40s 157ms/step - loss: 0.00841s - loss
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.528425907303436 - normalized_discounted_cumulative_gain@5(0.0): 0.5952097664878763 - mean_average_precision(0.0): 0.5498493585406231
Epoch 25/30
255/255 [==============================] - 41s 160ms/step - loss: 0.0071
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5303218965746611 - normalized_discounted_cumulative_gain@5(0.0): 0.5948210455548147 - mean_average_precision(0.0): 0.5505741686543946
Epoch 26/30
255/255 [==============================] - 41s 160ms/step - loss: 0.0065
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5223105087495103 - normalized_discounted_cumulative_gain@5(0.0): 0.5901051330822427 - mean_average_precision(0.0): 0.5426582852986084
Epoch 27/30
255/255 [==============================] - 42s 164ms/step - loss: 0.0045
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5259626991148388 - normalized_discounted_cumulative_gain@5(0.0): 0.5922912075108971 - mean_average_precision(0.0): 0.5439769673667663
Epoch 28/30
255/255 [==============================] - 42s 164ms/step - loss: 0.0039
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5326394392626637 - normalized_discounted_cumulative_gain@5(0.0): 0.5991254533953382 - mean_average_precision(0.0): 0.5509986930248422
Epoch 29/30
255/255 [==============================] - 34s 134ms/step - loss: 0.0037
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5250020425449076 - normalized_discounted_cumulative_gain@5(0.0): 0.5952453236394062 - mean_average_precision(0.0): 0.5498564190858788
Epoch 30/30
255/255 [==============================] - 35s 138ms/step - loss: 0.0046
Validation: normalized_discounted_cumulative_gain@3(0.0): 0.5309288871342496 - normalized_discounted_cumulative_gain@5(0.0): 0.5931766459815412 - mean_average_precision(0.0): 0.5474981722703828
In [ ]:
Content source: faneshion/MatchZoo
Similar notebooks: