lkrsnik/stress_asignment
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Created cnn witch look at aditional features as well

Browse Source
This commit is contained in:
lkrsnik 2017-07-01 15:45:46 +02:00
parent ac8f0057c5
commit d314a9ee4f
2 changed files with 92 additions and 33 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

25
.idea/workspace.xml
View File

@ -2,20 +2,7 @@
<project version="4"> <project version="4">
<component name="ChangeListManager"> <component name="ChangeListManager">
<list default="true" id="8a8ba9af-e1a4-433a-9968-475192610776" name="Default" comment=""> <list default="true" id="8a8ba9af-e1a4-433a-9968-475192610776" name="Default" comment="">
<change type="DELETED" beforePath="$PROJECT_DIR$/character_based_ffnn/character_based_ffnn.ipynb" afterPath="" /> <change type="MODIFICATION" beforePath="$PROJECT_DIR$/cnn/character_based_ffnn_keras.ipynb" afterPath="$PROJECT_DIR$/cnn/character_based_ffnn_keras.ipynb" />
<change type="DELETED" beforePath="$PROJECT_DIR$/character_based_ffnn/character_based_ffnn_keras.ipynb" afterPath="" />
<change type="DELETED" beforePath="$PROJECT_DIR$/character_based_ffnn/character_based_ffnn_theano.ipynb" afterPath="" />
<change type="DELETED" beforePath="$PROJECT_DIR$/character_based_ffnn/cnn_per_vowel_3epoch.h5" afterPath="" />
<change type="DELETED" beforePath="$PROJECT_DIR$/character_based_ffnn/create_and_save_inputs.py" afterPath="" />
<change type="DELETED" beforePath="$PROJECT_DIR$/character_based_ffnn/internal_representations/models/cnn_i2_s_c43-3relu_d20_c43-3relu_mp2_f_516relu_d20_121sigmoid_mse_adam_a65.h5" afterPath="" />
<change type="DELETED" beforePath="$PROJECT_DIR$/character_based_ffnn/internal_representations/models/cnn_i2_s_c43-3relu_d20_c43-3relu_mp2_f_516relu_d20_121sigmoid_mse_adam_a65_10epoch.h5" afterPath="" />
<change type="DELETED" beforePath="$PROJECT_DIR$/character_based_ffnn/internal_representations/models/cnn_i2_s_c43-3relu_d20_c43-3relu_mp2_f_516relu_d20_121sigmoid_mse_adam_a65_5epoch.h5" afterPath="" />
<change type="DELETED" beforePath="$PROJECT_DIR$/character_based_ffnn/internal_representations/models/ffnn_i1_s500relu_121sigmoid_mse_adam_a65.h5" afterPath="" />
<change type="DELETED" beforePath="$PROJECT_DIR$/character_based_ffnn/internal_representations/models/ffnn_i1_s500relu_d20_121sigmoid_mse_adam_a65.h5" afterPath="" />
<change type="DELETED" beforePath="$PROJECT_DIR$/character_based_ffnn/test.txt" afterPath="" />
<change type="DELETED" beforePath="$PROJECT_DIR$/character_based_ffnn/word_accetuation/character_based_ffnn_keras.ipynb" afterPath="" />
<change type="DELETED" beforePath="$PROJECT_DIR$/character_based_ffnn/word_accetuation/cnn_i2_s_c43-3relu_d20_c43-3relu_mp2_f_516relu_d20_121sigmoid_mse_adam_a65_10epoch_no_overfitting.h5" afterPath="" />
<change type="MODIFICATION" beforePath="$PROJECT_DIR$/.idea/workspace.xml" afterPath="$PROJECT_DIR$/.idea/workspace.xml" />
<change type="MODIFICATION" beforePath="$PROJECT_DIR$/prepare_data.py" afterPath="$PROJECT_DIR$/prepare_data.py" /> <change type="MODIFICATION" beforePath="$PROJECT_DIR$/prepare_data.py" afterPath="$PROJECT_DIR$/prepare_data.py" />
</list> </list>
<option name="EXCLUDED_CONVERTED_TO_IGNORED" value="true" /> <option name="EXCLUDED_CONVERTED_TO_IGNORED" value="true" />
@ -46,8 +33,8 @@
<file leaf-file-name="prepare_data.py" pinned="false" current-in-tab="true"> <file leaf-file-name="prepare_data.py" pinned="false" current-in-tab="true">
<entry file="file://$PROJECT_DIR$/prepare_data.py"> <entry file="file://$PROJECT_DIR$/prepare_data.py">
<provider selected="true" editor-type-id="text-editor"> <provider selected="true" editor-type-id="text-editor">
<state relative-caret-position="410"> <state relative-caret-position="442">
<caret line="545" column="14" lean-forward="true" selection-start-line="545" selection-start-column="14" selection-end-line="545" selection-end-column="14" /> <caret line="494" column="36" lean-forward="true" selection-start-line="494" selection-start-column="36" selection-end-line="494" selection-end-column="36" />
<folding> <folding>
<element signature="e#24#63#0" expanded="true" /> <element signature="e#24#63#0" expanded="true" />
</folding> </folding>
@ -137,7 +124,6 @@
<find>nearly_zeros</find> <find>nearly_zeros</find>
<find>rand</find> <find>rand</find>
<find>u</find> <find>u</find>
<find>shuffle_inputs</find>
<find>num_all_vowels</find> <find>num_all_vowels</find>
<find>load_shuffle_vector</find> <find>load_shuffle_vector</find>
<find>create_and_save_inputs</find> <find>create_and_save_inputs</find>
@ -147,6 +133,7 @@
<find>generate_inputs</find> <find>generate_inputs</find>
<find>split_number</find> <find>split_number</find>
<find>StringIO</find> <find>StringIO</find>
<find>shuffle_inputs</find>
</findStrings> </findStrings>
</component> </component>
<component name="Git.Settings"> <component name="Git.Settings">
@ -990,8 +977,8 @@
</entry> </entry>
<entry file="file://$PROJECT_DIR$/prepare_data.py"> <entry file="file://$PROJECT_DIR$/prepare_data.py">
<provider selected="true" editor-type-id="text-editor"> <provider selected="true" editor-type-id="text-editor">
<state relative-caret-position="410"> <state relative-caret-position="442">
<caret line="545" column="14" lean-forward="true" selection-start-line="545" selection-start-column="14" selection-end-line="545" selection-end-column="14" /> <caret line="494" column="36" lean-forward="true" selection-start-line="494" selection-start-column="36" selection-end-line="494" selection-end-column="36" />
<folding> <folding>
<element signature="e#24#63#0" expanded="true" /> <element signature="e#24#63#0" expanded="true" />
</folding> </folding>

94
prepare_data.py
View File

@ -8,28 +8,35 @@ import gc
import math import math
# functions for saving, loading and shuffling whole arrays to ram # functions for saving, loading and shuffling whole arrays to ram
def save_inputs(file_name, X, y): def save_inputs(file_name, X, y, other_features=[]):
h5f = h5py.File(file_name, 'w') h5f = h5py.File(file_name, 'w')
if other_features == []:
adict = dict(X=X, y=y) adict = dict(X=X, y=y)
else:
adict = dict(X=X, X_other_features=other_features, y=y)
for k, v in adict.items(): for k, v in adict.items():
h5f.create_dataset(k, data=v) h5f.create_dataset(k, data=v)
h5f.close() h5f.close()
def load_inputs(file_name): def load_inputs(file_name, other_features=False):
h5f = h5py.File(file_name,'r') h5f = h5py.File(file_name,'r')
X = h5f['X'][:] X = h5f['X'][:]
y = h5f['y'][:] y = h5f['y'][:]
if other_features:
X_other_features = h5f['X_other_features'][:]
h5f.close()
return X, X_other_features, y
h5f.close() h5f.close()
return X, y return X, y
def shuffle_inputs(X, y, X_pure=False): def shuffle_inputs(X, y, X_pure=[]):
s = np.arange(X.shape[0]) s = np.arange(X.shape[0])
np.random.shuffle(s) np.random.shuffle(s)
X = X[s] X = X[s]
y = y[s] y = y[s]
if X_pure: if X_pure != []:
X_pure = X_pure[s] X_pure = X_pure[s]
return X, y, X_pure return X, y, X_pure
else: else:
@ -94,7 +101,7 @@ def load_model(file_name):
# functions for creating X and y from content # functions for creating X and y from content
def read_content(): def read_content():
print('READING CONTENT...') print('READING CONTENT...')
with open('../../data/SlovarIJS_BESEDE_utf8.lex') as f: with open('../../../data/SlovarIJS_BESEDE_utf8.lex') as f:
content = f.readlines() content = f.readlines()
print('CONTENT READ SUCCESSFULY') print('CONTENT READ SUCCESSFULY')
return [x.split('\t') for x in content] return [x.split('\t') for x in content]
@ -262,17 +269,22 @@ def generate_presentable_y(accetuations_list, word_list, max_num_vowels):
def generate_full_matrix_inputs(): def generate_full_matrix_inputs():
dictionary, max_word, max_num_vowels, content, vowels, accetuated_vowels = create_dict() dictionary, max_word, max_num_vowels, content, vowels, accetuated_vowels = create_dict()
train_content, validate_content = split_content(content, 0.2) train_content, validate_content = split_content(content, 0.2)
feature_dictionary = create_feature_dictionary(content)
# Generate X and y # Generate X and y
print('GENERATING X AND y...') print('GENERATING X AND y...')
X_train, y_train = generate_X_and_y(dictionary, max_word, max_num_vowels, train_content, vowels, accetuated_vowels) X_train, X_other_features_train, y_train = generate_X_and_y(dictionary, max_word, max_num_vowels, train_content, vowels, accetuated_vowels, feature_dictionary)
X_validate, y_validate = generate_X_and_y(dictionary, max_word, max_num_vowels, validate_content, vowels, accetuated_vowels) X_validate, X_other_features_validate, y_validate = generate_X_and_y(dictionary, max_word, max_num_vowels, validate_content, vowels, accetuated_vowels, feature_dictionary)
print('GENERATION SUCCESSFUL!') print('GENERATION SUCCESSFUL!')
return X_train, y_train, X_validate, y_validate return X_train, X_other_features_train, y_train, X_validate, X_other_features_validate, y_validate
def generate_X_and_y(dictionary, max_word, max_num_vowels, content, vowels, accetuated_vowels):
def generate_X_and_y(dictionary, max_word, max_num_vowels, content, vowels, accetuated_vowels, feature_dictionary):
y = np.zeros((len(content), max_num_vowels * max_num_vowels )) y = np.zeros((len(content), max_num_vowels * max_num_vowels ))
X = np.zeros((len(content), max_word, len(dictionary))) X = np.zeros((len(content), max_word, len(dictionary)))
print('CREATING OTHER FEATURES...')
X_other_features = create_X_features(content, feature_dictionary)
print('OTHER FEATURES CREATED!')
i = 0 i = 0
for el in content: for el in content:
@ -302,9 +314,9 @@ def generate_X_and_y(dictionary, max_word, max_num_vowels, content, vowels, acce
i += 1 i += 1
print('SHUFFELING INPUTS...') print('SHUFFELING INPUTS...')
X, y = shuffle_inputs(X, y) X, y, X_other_features = shuffle_inputs(X, y, X_pure=X_other_features)
print('INPUTS SHUFFELED!') print('INPUTS SHUFFELED!')
return X, y return X, X_other_features, y
def count_vowels(content, vowels): def count_vowels(content, vowels):
@ -473,6 +485,27 @@ def shuffle_full_vowel_inputs(name, orderd_name, parts):
# Decoders for inputs and outputs # Decoders for inputs and outputs
def decode_X_features(feature_dictionary, X_other_features):
for word in X_other_features:
final_word = []
i = 0
for z in range(len(feature_dictionary)):
for j in range(1, len(feature_dictionary[z])):
if j == 1:
if word[i] == 1:
# print feature_dictionary[z][1]
final_word.append(feature_dictionary[z][1])
i += 1
else:
for k in range(len(feature_dictionary[z][j])):
# print (i)
if word[i] == 1:
# print feature_dictionary[z][j][k]
final_word.append(feature_dictionary[z][j][k])
i += 1
print(u''.join(final_word))
def decode_position(y, max_num_vowels): def decode_position(y, max_num_vowels):
max_el = 0 max_el = 0
i = 0 i = 0
@ -566,3 +599,42 @@ def split_content(content, ratio):
train_content = [content[i] for i in range(len(content)) if expanded_content[i] in shuffled_unique_train_content_set] train_content = [content[i] for i in range(len(content)) if expanded_content[i] in shuffled_unique_train_content_set]
validate_content = [content[i] for i in range(len(content)) if expanded_content[i] in shuffled_unique_validate_content_set] validate_content = [content[i] for i in range(len(content)) if expanded_content[i] in shuffled_unique_validate_content_set]
return train_content, validate_content return train_content, validate_content
# create feature dictionary
def create_feature_dictionary(content):
additional_data = [el[2] for el in content]
possible_variants = sorted(set(additional_data))
categories = sorted(set([el[0] for el in possible_variants]))
feature_dictionary = []
for category in categories:
category_features = [1, category]
examples_per_category = [el for el in possible_variants if el[0] == category]
longest_element = max(examples_per_category, key=len)
for i in range(1, len(longest_element)):
possibilities_per_el = sorted(set([el[i] for el in examples_per_category if i < len(el)]))
category_features[0] += len(possibilities_per_el)
category_features.append(possibilities_per_el)
feature_dictionary.append(category_features)
return feature_dictionary
def create_X_features(content, feature_dictionary):
content = content
X_other_features = []
for el in content:
X_el_other_features = []
for feature in feature_dictionary:
if el[2][0] == feature[1]:
X_el_other_features.append(1)
for i in range(2, len(feature)):
for j in range(len(feature[i])):
if i-1 < len(el[2]) and feature[i][j] == el[2][i-1]:
X_el_other_features.append(1)
else:
X_el_other_features.append(0)
else:
X_el_other_features.extend([0] * feature[0])
X_other_features.append(X_el_other_features)
return np.array(X_other_features)