-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathgmm_train.py
218 lines (189 loc) · 11.3 KB
/
gmm_train.py
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
#python kmeans_train.py --train_file data/jobQ123_BOTH/processed/jobQ1_BOTH/split/jobQ1_BOTH_train.json --dev_file data/jobQ123_BOTH/processed/jobQ1_BOTH/split/jobQ1_BOTH_dev.json --lower 2 --upper 12 --iterations 5 --output_file jobQ1_BOTH_split_kmeans --folder_name data/jobQ1_BOTH/kmeans
#https://stackoverflow.com/questions/37604289/tkinter-tclerror-no-display-name-and-no-display-environment-variable/43592515
#for running the pipeline through SSH
import os
import matplotlib as mpl
if os.environ.get('DISPLAY','') == '':
mpl.use('Agg')
from sklearn import mixture
from tqdm import tqdm
import os, math, sys, json, collections
from scipy.stats import entropy
import numpy as np
import joblib
from label_vectorization import get_ans_pct_vectors,get_assignments,tests,get_perplexity
from helper_functions import write_model_logs_to_json,read_labeled_data_KMeans,create_folder,get_index_of_best_iteration,save_trained_model_joblib,save_max_sklearn_model_trained,save_trained_model_joblib_sklearn,KLdivergence,median,create_folder
from helper_functions import sklearn_find_kl,iteration_selection_sklearn,find_item_distribution_clusters_sklearn,get_ids_only
from helper_functions_nlp import clean_text_for_sklean,build_bag_of_words,data_in_cluster_sklearn,save_trained_model_joblib_sklearn_nlp,prep_tokens_for_doc2vec,embed_to_vect,build_glove_embed,glove_embed_vects,text_hybrid_labels,hybrid_flag
import argparse
import sys
from collections import Counter
import pdb
import pandas as pd
from ldl_utils import read_json
import shutil
pretrained_emb = "data/lexicons/glove.twitter.27B/glove.twitter.27B.100d.txt"
#doc2vec parameters
vector_size = 300
window_size = 15
min_count = 1
sampling_threshold = 1e-5
negative_size = 5
train_epoch = 100
dm = 0 #0 = dbow; 1 = dmpv
worker_count = 1 #number of parallel processe
model_selection_measure = "cross"
iterations = 10
# v = {"entropy": entropee, "max": maxy, "distance": scores, "centroid": centroidy, "cross": cross}
def train_dev_gmm_selection(train_answer_counters,dev_answer_counters, ITERATIONS, LOWER, UPPER, output_name, folder_name):
# # Read data splits from file, NOT generate each time
# with open(SPLIT_LOG_DIR + output_name + "_" + split_prep + ".json") as fp:
# results_dict = json.load(fp)
# train_items = results_dict['train_set']
# dev_items = results_dict['dev_set']
#
# train_answer_counters = {}
# for k in train_items:
# train_answer_counters[k] = tweetid_answer_counters[k]
train_vectors = get_ans_pct_vectors(train_answer_counters)
train_message_ids = get_ids_only(train_answer_counters)
dev_vectors = get_ans_pct_vectors(dev_answer_counters)
results_log_dict = {}
results_dict = {}
for n_clusters in tqdm(range(LOWER, UPPER)):
# print(n_clusters)
# maxy = []
# entropee = []
# scores = []
# cross = []
# centroidy = []
kl = []
results = {}
for i in range(iterations):
# Initialize the clusterer with n_clusters value and a random generator seed of 10 for reproducibility
clusterer = mixture.GaussianMixture(n_components=n_clusters)
train_predict = clusterer.fit_predict(train_vectors)
cluster_distributions = data_in_cluster_sklearn(train_predict,n_clusters,train_message_ids,train_answer_counters)
kl.append(sklearn_find_kl(train_answer_counters,train_predict, cluster_distributions))
results[i] = find_item_distribution_clusters_sklearn(train_predict)
create_folder(folder_name + "/logs/models/CL"+str(n_clusters)+"/temp"+str(i))
write_model_logs_to_json(folder_name + "/logs/models/CL"+str(n_clusters)+"/temp"+str(i),cluster_distributions,"cluster_info_"+str(n_clusters))
save_trained_model_joblib_sklearn_nlp(folder_name + "/logs/models/CL"+str(n_clusters)+"/temp"+str(i), clusterer, output_name, n_clusters)
model,cluster_distributions,results_log_dict[n_clusters] = iteration_selection_sklearn(kl,results,folder_name + "/logs/models/CL"+str(n_clusters)+"/temp",n_clusters)
shutil.rmtree(folder_name + "/logs/models/CL"+str(n_clusters))
write_model_logs_to_json(folder_name + "/logs/models",cluster_distributions,"cluster_info_"+str(n_clusters))
# clusterer = KMeans(n_clusters=n_clusters)
# clusterer.fit(train_vectors)
save_trained_model_joblib_sklearn_nlp(folder_name + "/logs/models", model, output_name, n_clusters)
results_log_dict["exp_name"] = output_name
write_model_logs_to_json(folder_name + "/logs/models/",results_log_dict,"cluster_log")
print ("Completed GMM Training")
def train_dev_gmm_nlp(train_answer_counters,dev_answer_counters, ITERATIONS, LOWER, UPPER, output_name, folder_name,label_dict,train_message_dict,dev_message_dict,glove,hybrid,train_vects,dev_vects):
train_messages,train_message_ids,train_cleaned_messages,train_tokens = clean_text_for_sklean(train_message_dict)
dev_messages,dev_message_ids,dev_cleaned_messages,dev_tokens = clean_text_for_sklean(dev_message_dict)
if glove == "bert":
train_vectors = train_vects
dev_vectors = dev_vects
if glove == True:
vec_model = build_glove_embed(train_cleaned_messages)
train_vectors,_ = glove_embed_vects(train_tokens,vec_model)
vec_model.save(folder_name + "/logs/models/gmm_glove.dict")
# else:
# train_vectors,sklearn_bow_model = build_bag_of_words(train_cleaned_messages)
# dev_vectors = sklearn_bow_model.transform(dev_cleaned_messages)
if hybrid:
train_vectors = text_hybrid_labels(train_vectors,train_answer_counters,float(hybrid))
results_log_dict = {}
results_dict = {}
for n_clusters in tqdm(range(LOWER, UPPER)):
# print(n_clusters)
# maxy = []
# entropee = []
# scores = []
# cross = []
# centroidy = []
kl = []
results = {}
for i in range(iterations):
# Initialize the clusterer with n_clusters value and a random generator seed of 10 for reproducibility
clusterer = mixture.GaussianMixture(n_components=n_clusters)
train_predict = clusterer.fit_predict(train_vectors)
cluster_distributions = data_in_cluster_sklearn(train_predict,n_clusters,train_message_ids,train_answer_counters)
kl.append(sklearn_find_kl(train_answer_counters,train_predict, cluster_distributions))
results[i] = find_item_distribution_clusters_sklearn(train_predict)
create_folder(folder_name + "/logs/models/CL"+str(n_clusters)+"/temp"+str(i))
write_model_logs_to_json(folder_name + "/logs/models/CL"+str(n_clusters)+"/temp"+str(i),cluster_distributions,"cluster_info_"+str(n_clusters))
save_trained_model_joblib_sklearn_nlp(folder_name + "/logs/models/CL"+str(n_clusters)+"/temp"+str(i), clusterer, output_name, n_clusters)
model,cluster_distributions,results_log_dict[n_clusters] = iteration_selection_sklearn(kl,results,folder_name + "/logs/models/CL"+str(n_clusters)+"/temp",n_clusters)
shutil.rmtree(folder_name + "/logs/models/CL"+str(n_clusters))
write_model_logs_to_json(folder_name + "/logs/models",cluster_distributions,"cluster_info_"+str(n_clusters))
save_trained_model_joblib_sklearn_nlp(folder_name + "/logs/models/", model, output_name, n_clusters)
results_log_dict["exp_name"] = output_name
write_model_logs_to_json(folder_name + "/logs/models/",results_log_dict,"cluster_log")
print ("Completed GMM NLP Training")
def model_selection(cluster_log,output_dir,output_name,LOWER, UPPER):
max_cluster_id,max_iteration = model_selection_gmm(cluster_log, model_selection_measure,LOWER, UPPER)
model_dir = output_dir + '/logs/models/CL' + str(max_cluster_id) + '/'
model_path = model_dir + "Iter" + str(max_iteration) +'.pkl'
save_max_sklearn_model_trained(model_path,output_dir,output_name)
print ("Model training for GMM completed cluster number: "+str(max_cluster_id)+" and saved to "+model_path)
def model_selection_gmm(cluster_log, measure_name,LOWER, UPPER):
# Select model by the Maximum of **measure_name**
# measure_name = "entropy" or "likelihood"
print(measure_name)
max_meas = cluster_log[LOWER][measure_name]
max_meas_idx = 0
for n_clusters in range(LOWER, UPPER):
# v = {"entropy": entropee, "max": maxy, "likelihood": likelies, "centroid": centroidy}
target_values = cluster_log[n_clusters][measure_name]
if measure_name == "cross":
if target_values <= max_meas:
max_meas_idx = n_clusters
max_meas = target_values
max_iteration = cluster_log[n_clusters]["max_iteration"]
else:
if target_values >= max_meas:
max_meas_idx = n_clusters
max_meas = target_values
max_iteration = cluster_log[n_clusters]["max_iteration"]
print(max_meas_idx, max_meas,max_iteration)
return max_meas_idx,max_iteration
def preprocess_data(input_train_file_name,input_dev_file_name,folder_name):
create_folder(folder_name)
create_folder(folder_name + "/logs")
create_folder(folder_name + "/logs/models")
train_answer_counters,train_message_dict,label_dict = read_labeled_data_KMeans(input_train_file_name)
dev_answer_counters,dev_message_dict,label_dict = read_labeled_data_KMeans(input_dev_file_name)
return train_answer_counters,dev_answer_counters,label_dict,train_message_dict,dev_message_dict
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--train_file", help="Input training file JSON name")
parser.add_argument("--train_file_vects", help="Input training vects .npy",default=False)
parser.add_argument("--dev_file", help="Input dev file JSON name")
parser.add_argument("--dev_file_vects", help="Input dev vects .npy",default=False)
parser.add_argument("--lower", help="Lower Limit")
parser.add_argument("--upper", help="Upper Limit")
parser.add_argument("--iterations", help="Number of iterations")
parser.add_argument("--output_file", help="Output file name", default = False)
parser.add_argument("--folder_name", help="Main folder name",default = False)
parser.add_argument("--nlp_data", help="NLP Data",default = False)
parser.add_argument("--glove", help="Glove Embeddings",default=False)
parser.add_argument("--hybrid", help="Hybrid of Text + Labels", default=False)
args = parser.parse_args()
nlp_flag = args.nlp_data
glove = args.glove
hybrid = hybrid_flag(args.hybrid)
train_vects = args.train_file_vects
dev_vects = args.dev_file_vects
if train_vects:
train_vects = np.load(train_vects,allow_pickle=True)
if dev_vects:
dev_vects = np.load(dev_vects,allow_pickle=True)
#Reading Data
train_answer_counters,dev_answer_counters,label_dict,train_message_dict,dev_message_dict = preprocess_data(args.train_file,args.dev_file,args.folder_name)
if (nlp_flag and hybrid<100):
train_dev_gmm_nlp(train_answer_counters,dev_answer_counters,int(args.iterations), int(args.lower), int(args.upper),args.output_file,args.folder_name,label_dict,train_message_dict,dev_message_dict,glove,hybrid,train_vects,dev_vects)
else:
train_dev_gmm_selection(train_answer_counters,dev_answer_counters,int(args.iterations), int(args.lower), int(args.upper),args.output_file,args.folder_name)
if __name__ == '__main__':
main()