jstr_insert += '],"infomation_gain":['
for idx, line in enumerate(IT_score): # IT
if idx < num_to_show:
jstr_insert += '"' + line.rstrip() + '",'
jstr_insert = jstr_insert[:len(jstr_insert) - 1] # remove last ','
jstr_insert += '],"combined":['
for idx, line in enumerate(combined_score): # Combined
if idx < num_to_show:
jstr_insert += '"' + line.rstrip() + '",'
jstr_insert = jstr_insert[:len(jstr_insert) - 1] # remove last ','
jstr_insert += ']}}'
jstr_insert = jstr_insert.replace("\t", ",")
#print "jstr_insert=",jstr_insert
## write to mongoDB.myml.dataset_info, ignore doc with duplicated key
ret = query_mongo.upsert_doc_t(mongo_tuples, filter, jstr_insert,
upsert_flag)
print "INFO: Upsert count for feature importance=", ret
print 'INFO: Finished!'
return 0
# combint to one table
def combine_with_coef(row_id_str, coef_arr, FIRM_list, IT_list, Prob_list,
out_filename, feat_sample_count_arr):
#print "INFO: combined fname=",out_filename
if os.path.exists(out_filename):
try:
os.remove(out_filename)
except OSError, e:
print("ERROR: %s - %s." % (e.out_filename, e.strerror))
def train(row_id_str, ds_id, hdfs_feat_dir, local_out_dir, ml_opts_jstr, excluded_feat_cslist
, sp_master, spark_rdd_compress, spark_driver_maxResultSize, sp_exe_memory, sp_core_max
, zipout_dir, zipcode_dir, zip_file_name
, mongo_tuples, labelnameflag, fromweb
, training_fraction, jobname, model_data_folder ):
# zip func in other files for Spark workers ================= ================
zip_file_path = ml_build_zip_file(zipout_dir, zipcode_dir, zip_file_name, prefix='zip_feature_util')
print "INFO: zip_file_path=",zip_file_path
# ML model filename ====
model_fname=os.path.join(model_data_folder, row_id_str+'.pkl')
print "INFO: model_data_folder=",model_data_folder
# create out folders and clean up old model files ====
ml_util.ml_prepare_output_dirs(row_id_str,local_out_dir,model_data_folder,model_fname)
# init Spark context ====
sc=ml_util.ml_get_spark_context(sp_master
, spark_rdd_compress
, spark_driver_maxResultSize
, sp_exe_memory
, sp_core_max
, jobname
, [zip_file_path])
t0 = time()
t00 = t0
# check if ml_opts.has_excluded_feat ==1 ===================================
has_excluded_feat=0
if not ml_opts_jstr is None:
ml_opts=json.loads(ml_opts_jstr)
if "has_excluded_feat" in ml_opts:
has_excluded_feat=ml_opts["has_excluded_feat"]
# get excluded feature list from mongo ========== ===
if str(has_excluded_feat) == "1" and excluded_feat_cslist is None:
excluded_feat_cslist=ml_util.ml_get_excluded_feat(row_id_str, mongo_tuples)
print "INFO: excluded_feat_cslist=",excluded_feat_cslist
# source libsvm filename
libsvm_data_file = os.path.join(hdfs_feat_dir , "libsvm_data")
print "INFO: libsvm_data_file=", libsvm_data_file
# load feature count file
feat_count_file=libsvm_data_file+"_feat_count"
feature_count=zip_feature_util.get_feature_count(sc,feat_count_file)
print "INFO: feature_count=",feature_count
# load sample RDD from text file
# also exclude selected features in sample ================ =====
# format (LabeledPoint,hash) from str2LabeledPoint_hash()
#samples_rdd = MLUtils.loadLibSVMFile(sc, libsvm_data_file)
samples_rdd,feature_count = zip_feature_util.get_sample_rdd(sc, libsvm_data_file, feature_count, excluded_feat_cslist)
all_data = samples_rdd.collect()
sample_count=len(all_data)
# 2-D array
features_list = [x.features.toArray() for x,_ in all_data]
# label array
labels_list_all = [x.label for x,_ in all_data]
# hash array
hash_list_all = [x for _,x in all_data]
# convert to np array
labels_list_all = array(labels_list_all)
features_array = np.array(features_list)
hash_list_all=np.array(hash_list_all)
# generate sparse matrix (csr) for all samples
features_sparse_mtx = csr_matrix(features_array)
### randomly split the samples into training and testing data ===============
X_train_sparse, X_test_sparse, labels_train, labels_test, train_hash_list, test_hash_list = \
cross_validation.train_test_split(features_sparse_mtx, labels_list_all, hash_list_all, test_size=(1-training_fraction) )
# X_test_sparse is scipy.sparse.csr.csr_matrix
testing_sample_count = len(labels_test)
training_sample_count=len(labels_train)
training_lbl_cnt_list=Counter(labels_train)
testing_lbl_cnt_list=Counter(labels_test)
print "INFO: training sample count=",training_sample_count,", testing sample count=",testing_sample_count,",sample_count=",sample_count
print "INFO: training label list=",training_lbl_cnt_list,", testing label list=",testing_lbl_cnt_list
print "INFO: train_hash_list count=",len(train_hash_list),", test_hash_list count=",len(test_hash_list)
t1 = time()
print 'INFO: running time: %f' %(t1-t0)
###############################################
###########build learning model################
###############################################
### parse parameters and generate the model ###
(clf, model_name, api, cv, param_dic) = parse_param_and_get_model(ml_opts)
if model_name == "none":
print "ERROR: model name not found!"
return -1
#param_jobj=json.loads(ml_opts_jstr);
#print "param_jobj=",param_jobj
########################################################
##########Grid Search with cross validation#############
########################################################
json2save={}
json2save["rid"]=int(row_id_str)
json2save["key"]="cv_result"
#json2save["param_str"]=ml_opts_jstr
json2save["param_dic"]=param_dic
cv_grid=[]
if api == "centralized":
#########run with Scikit-learn API (for comparison)######
print "INFO: ******************Grid Search with Scikit-learn API************"
t0 = time()
# Set the parameters by cross-validation
#tuned_parameters = [{'C': [0.0001, 0.001, 0.01, 0.1, 1, 10, 100, 1000, 10000, 100000]}]
#tuned_parameters = [{'kernel': ['rbf'], 'gamma': [1e-3, 1e-4], \
# 'C': [1, 10, 100, 1000]}, \
# {'kernel': ['linear'], 'C': [1, 10, 100, 1000]}]
scores = ['accuracy']
json2save["scores"]=scores
#print json2save
for score in scores: # for one item only? score=accuracy
print("INFO: # Tuning hyper-parameters for %s" % score)
#print()
grid = grid_search.GridSearchCV(estimator = clf, param_grid = param_dic, cv=cv, scoring= score)
grid.fit(X_train_sparse, labels_train)
print "INFO: Best parameters set found on development set:"
print "INFO: grid.best_params_=",grid.best_params_
print "INFO: Grid scores on development set:"
for key in grid.best_params_:
print "INFO: best_params["+key+"]=", grid.best_params_[key]
if key.lower()=="regtype":
ml_opts['regularization']=str(grid.best_params_[key]) # add best param to
else:
ml_opts[key.lower()]=str(grid.best_params_[key]) # add best param to
# save best param to db as json string
j_str=json.dumps(ml_opts);
json2save["param_str"]=j_str;
print "INFO: grid_scores_ with params:"
for params, mean_score, scores in grid.grid_scores_:
print "INFO: %0.3f (+/-%0.03f) for %r" % (mean_score, scores.std() * 2, params)
#outstr='%s,%0.3f,%0.03f,%s' % (params,mean_score, scores.std() * 2,"Selected" if params==grid.best_params_ else "")
outj={}
outj["param"]=params
outj["average_accuracy"]="%0.3f" % (mean_score)
outj["std_deviation"]="%0.3f" % (scores.std() * 2)
outj["selected"]="%s" % ("Selected" if params==grid.best_params_ else "")
cv_grid.append(outj)
clf_best = grid.best_estimator_
t1 = time()
############# END run with SKlearn ######
print 'INFO: Grid Search with SKlearn running time: %f' %(t1-t0)
t0 = time()
else:
#############run with SPARK######
print "INFO: ******************Grid Search with SPARK************"
all_comb_list_of_dic = get_all_combination_list_of_dic(param_dic)
print "INFO: Total number of searching combinations=", len(all_comb_list_of_dic)
#print "all_comb_list_of_dic: ", all_comb_list_of_dic
params_rdd = sc.parallelize(all_comb_list_of_dic)
###broad cast clf, traning data, testing data to all workers###
X_broadcast = sc.broadcast(X_train_sparse)
y_broadcast = sc.broadcast(labels_train)
clf_broadcast = sc.broadcast(clf)
### Grid Search with CV in multiple workers ###
models = params_rdd.map(lambda x: learn_with_params(clf_broadcast.value, X_broadcast.value, y_broadcast.value, cv, x)).sortByKey(ascending = False).cache()
(ave_accuracy, (clf_best, p_dic_best, std2)) = models.first()
# output results #
print "INFO: Best parameters set found for ", model_name, " is: "
print "INFO: ",
for key in p_dic_best:
print key, " = ", p_dic_best[key],
if key.lower()=="regtype":
ml_opts['regularization']=str(p_dic_best[key])
else:
ml_opts[key.lower()]=str(p_dic_best[key]) # add best param to
# save best param to db as json string
print ""
j_str=json.dumps(ml_opts);
json2save["param_str"]=j_str;
print "INFO: Average accuracy with CV = ", cv, ": ", ave_accuracy
######## print complete report #######
print "INFO: Grid scores on development set:"
all_results = models.collect()
for i in range(0, len(all_results)):
(ave_accu_i, (clf_i, p_dic_i, std2_i)) = all_results[i]
print "INFO: ",ave_accu_i, " for ", p_dic_i
print "INFO: %0.3f (+/-%0.03f) for " % (ave_accu_i, std2_i), p_dic_i
#outstr='%s,%0.3f,%0.03f,%s' % ( p_dic_i, ave_accu_i, std2_i, "Selected" if p_dic_i==p_dic_best else "")
outj={}
outj["param"]=p_dic_i
outj["average_accuracy"]="%0.3f" % (ave_accu_i)
outj["std_deviation"]="%0.3f" % (std2_i)
outj["selected"]="%s" % ("Selected" if p_dic_i==p_dic_best else "")
cv_grid.append(outj)
print " "
t1 = time()
############# END run with SPARK######
print 'INFO: Grid search with SPARK running time: %f' %(t1-t0)
##################################################################################
#print "cv_grid=",cv_grid
#json2save["cv_grid_title"]='param,average_accuracy,std_deviation,selected'
json2save["cv_grid_data"]=cv_grid
json2save['clf_best']=str(clf_best).replace("\n","").replace(" ","")
cv_result=json.dumps(json2save)
#print "INFO: cv_result=",cv_result
filter='{"rid":'+row_id_str+',"key":"cv_result"}'
upsert_flag=True
## write to mongoDB.myml.dataset_info, ignore doc with duplicated key
# db.dataset_info.createIndex({"rid":1,"key":1},{unique:true})
ret=query_mongo.upsert_doc_t(mongo_tuples,filter,cv_result,upsert_flag)
print "INFO: Upsert count for cv_result: ret=",ret
##################################################################################
##########Retrain with best model for training set and output results#############
##################################################################################
print "INFO: **********Retrain with best model for training set and output results************"
clf_best.fit(X_train_sparse, labels_train)
#### save clf_best for future use ####
#joblib.dump(clf_best, model_data_folder + row_id_str+'.pkl')
joblib.dump(clf_best, model_fname)
### Evaluating the model on testing data
labels_pred = clf_best.predict(X_test_sparse)
accuracy = clf_best.score(X_test_sparse, labels_test)
print "INFO: Accuracy = ", accuracy
######################################the rest of the code is the same as train_sklean.py (replace clf with clf_best)#####################################################################
clf=clf_best
print "INFO: model type=",type(clf)," clf=",clf
# get data from model ================================
coef=None
intercept=None
try:
if type(clf) in ( classes.SVC , classes.NuSVC) :# svm didn't have coef_
col_num=clf.support_vectors_.shape[1]
else: #linear only
# coef_ is only available when using a linear kernel
col_num = len(clf.coef_[0])
coef=clf.coef_[0]
intercept=clf.intercept_[0] # only get 1st item?
#print "**model:clf.coef_[0] =",clf.coef_[0]
except Exception as e:
print "WARNING: Can't get clf.coef_[0]. e=",e,", get total features from meta-data"
col_num = 0 #how to get feature number for sparse array?
print "INFO: total feature # in the model: ", col_num
jfeat_coef_dict={}
# create feature coefficient file ================================
if coef is None:
print "WARNING: model weights not found!"
else:
feat_filename=os.path.join(local_out_dir,row_id_str+"_feat_coef.json")
print "INFO: feat_filename=",feat_filename
# save coef_arr to mongo & create jfeat_coef_dict===
jfeat_coef_dict=ml_util.ml_save_coef_build_feat_coef(row_id_str, mongo_tuples, coef, intercept, feat_filename, ds_id)
#print "INFO: jfeat_coef_dict=", jfeat_coef_dict
print "INFO: jfeat_coef_dict len=", len(jfeat_coef_dict )
# filename for false pred
false_pred_fname=os.path.join(local_out_dir,row_id_str+"_false_pred.json")
print "INFO: false_pred_fname=", false_pred_fname
# build files for false pred & score graph
(score_arr_0, score_arr_1, max_score,min_score)=ml_build_false_pred(X_test_sparse,coef,intercept
, labels_test, labels_pred, test_hash_list, model_name, jfeat_coef_dict, false_pred_fname)
# save pred output
pred_out_arr=[]
for i in range(0,len(labels_test)):
pred_out_arr.append((labels_test[i], labels_pred[i], test_hash_list[i]))
pred_ofname=os.path.join(local_out_dir,row_id_str+"_pred_output.pkl")
print "INFO: pred_ofname=", pred_ofname
ml_util.ml_pickle_save(pred_out_arr,pred_ofname)
###################################################
### generate label names (family names) ###########
### connect to database to get the column list which contains all column number of the corresponding feature####
###################################################
if labelnameflag == 1:
key = "dic_name_label"
jstr_filter='{"rid":'+row_id_str+',"key":"'+key+'"}'
jstr_proj='{"value":1}'
# get parent dataset's data
if ds_id != row_id_str:
jstr_filter='{"rid":'+ds_id+',"key":"'+key+'"}'
doc=query_mongo.find_one_t(mongo_tuples, jstr_filter, jstr_proj)
dic_list = doc['value']
label_dic = {}
for i in range(0, len(dic_list)):
for key in dic_list[i]:
label_dic[dic_list[i][key]] = key.encode('UTF8')
print "INFO: label_dic:", label_dic
else:
label_dic = {}
label_set = set(labels_list_all)
for label_value in label_set:
label_dic[int(label_value)] = str(int(label_value))
print "INFO: ******generated label_dic:", label_dic
labels_list = []
for key in sorted(label_dic):
labels_list.append(label_dic[key])
### generate sample numbers of each family in testing data###
testing_sample_number = len(labels_test)
print "INFO: testing_sample_number=", testing_sample_number
test_cnt_dic = {}
for key in label_dic:
test_cnt_dic[key] = 0
for i in range (0, testing_sample_number):
for key in label_dic:
if labels_test[i] == key:
test_cnt_dic[key] = test_cnt_dic[key] + 1
print "INFO: Number of samples in each label is=", test_cnt_dic
###############################################
###########plot prediction result figure#######
###############################################
pred_fname=os.path.join(local_out_dir,row_id_str+"_1"+".png")
true_fname=os.path.join(local_out_dir,row_id_str+"_2"+".png")
pred_xlabel='Prediction (Single Run)'
true_xlabel='True Labels (Single Run)'
test_cnt_dic=ml_util.ml_plot_predict_figures(labels_pred.tolist(), labels_test.tolist(), labels_list, label_dic, testing_sample_count
, pred_xlabel, pred_fname, true_xlabel, true_fname)
print "INFO: figure files: ", pred_fname, true_fname
print "INFO: Number of samples in each label is=", test_cnt_dic
roc_auc=None
#fscore=None
perf_measures=None
class_count=len(labels_list)
dataset_info={"training_fraction":training_fraction, "class_count":class_count,"dataset_count":sample_count}
#############################################################
###################for 2 class only (plot ROC curve)#########
#############################################################
if len(labels_list) == 2:
# build data file for score graph
score_graph_fname=os.path.join(local_out_dir,row_id_str+"_score_graph.json")
print "INFO: score_graph_fname=", score_graph_fname
ml_build_pred_score_graph(score_arr_0,score_arr_1,model_name, score_graph_fname,max_score,min_score)
do_ROC=True
reverse_label_dic = dict((v,k) for k, v in label_dic.items())
if 'clean' in reverse_label_dic:
flag_clean = reverse_label_dic['clean']
elif 'benign' in reverse_label_dic:
flag_clean = reverse_label_dic['benign']
elif '0' in reverse_label_dic:
flag_clean = 0
else:
print "No ROC curve generated: 'clean' or '0' must be a label for indicating negative class!"
do_ROC=False
if do_ROC:
# calculate fscore ==========
perf_measures=ml_util.calculate_fscore(labels_test, labels_pred)
print "INFO: perf_measures=",perf_measures
confidence_score = clf_best.decision_function(X_test_sparse)
if flag_clean == 0:
scores = [x for x in confidence_score]
s_labels = [x for x in labels_test]
testing_N = test_cnt_dic[0]
testing_P = test_cnt_dic[1]
else:
scores = [-x for x in confidence_score]
s_labels = [1-x for x in labels_test]
testing_N = test_cnt_dic[1]
testing_P = test_cnt_dic[0]
# create ROC data file ======== ====
roc_auc=ml_create_roc_files(row_id_str, scores, s_labels, testing_N, testing_P
, local_out_dir, row_id_str)
perf_measures["roc_auc"]=roc_auc
# only update db for web request
if fromweb=="1":
#print "database update"
str_sql="UPDATE atdml_document set "+"accuracy = '"+str(accuracy*100)+"%" \
+"', status = 'learned', processed_date ='"+str(datetime.datetime.now()) \
+"',ml_opts='"+j_str \
+"', perf_measures='"+json.dumps(perf_measures) \
+"', dataset_info='"+json.dumps(dataset_info) \
+"' where id="+row_id_str
ret=exec_sqlite.exec_sql(str_sql)
print "INFO: Data update done! ret=", str(ret)
else:
print "INFO: accuracy = '"+str(accuracy*100)+"%"
print 'INFO: total running time: %f' %(t1-t00)
print 'INFO: Finished!'
return 0
def train(row_id_str, ds_id, hdfs_feat_dir, local_out_dir, ml_opts_jstr,
excluded_feat_cslist, sp_master, spark_rdd_compress,
spark_driver_maxResultSize, sp_exe_memory, sp_core_max, zipout_dir,
zipcode_dir, zip_file_name, mongo_tuples, labelnameflag, fromweb,
training_fraction, jobname):
if not os.path.exists(local_out_dir):
os.makedirs(local_out_dir)
# zip func in other files for Spark workers ================= ================
zip_file_path = ml_build_zip_file(zipout_dir,
zipcode_dir,
zip_file_name,
prefix='zip_feature_util')
print "INFO: zip_file_path=", zip_file_path
# get_spark_context
sc = ml_util.ml_get_spark_context(sp_master, spark_rdd_compress,
spark_driver_maxResultSize,
sp_exe_memory, sp_core_max, jobname,
[zip_file_path])
t0 = time()
t00 = t0
# check if ml_opts.has_excluded_feat ==1 ===================================
has_excluded_feat = 0
ml_opts = {}
if not ml_opts_jstr is None:
ml_opts = json.loads(ml_opts_jstr)
if "has_excluded_feat" in ml_opts:
has_excluded_feat = ml_opts["has_excluded_feat"]
#print "has_excluded_feat=",has_excluded_feat,",excluded_feat_cslist=",excluded_feat_cslist
# get excluded feature list from mongo ========== ===
if str(has_excluded_feat) == "1" and excluded_feat_cslist is None:
excluded_feat_cslist = ml_util.ml_get_excluded_feat(
row_id_str, mongo_tuples)
print "INFO: excluded_feat_cslist=", excluded_feat_cslist
### generate Labeled point
libsvm_data_file = os.path.join(hdfs_feat_dir, "libsvm_data")
print "INFO: libsvm_data_file:", libsvm_data_file
# load feature count file
feat_count_file = libsvm_data_file + "_feat_count"
feature_count = zip_feature_util.get_feature_count(sc, feat_count_file)
print "INFO: feature_count=", feature_count
# load sample RDD from text file
# also exclude selected features in sample ================ =====
# format (LabeledPoint,hash) from str2LabeledPoint_hash()
#samples_rdd = MLUtils.loadLibSVMFile(sc, libsvm_data_file)
samples_rdd, feature_count = zip_feature_util.get_sample_rdd(
sc, libsvm_data_file, feature_count, excluded_feat_cslist)
#samples_rdd = MLUtils.loadLibSVMFile(sc, libsvm_data_file)
# get distinct label list
labels_list_all = samples_rdd.map(
lambda p: p[0].label).distinct().collect()
### generate training and testing data
training_rdd, testing_rdd = samples_rdd.randomSplit(
[training_fraction, 1 - training_fraction])
training_rdd = training_rdd.map(lambda p: p[0]) # keep LabeledPoint only
training_rdd.cache()
training_sample_count = training_rdd.count()
training_lbl_cnt_list = training_rdd.map(
lambda p: (p.label, 1)).reduceByKey(add).collect()
testing_rdd.cache()
testing_sample_count = testing_rdd.count()
testing_lbl_cnt_list = testing_rdd.map(
lambda p: (p[0].label, 1)).reduceByKey(add).collect()
sample_count = training_sample_count + testing_sample_count
t1 = time()
print "INFO: training sample count=", training_sample_count, ", testing sample count=", testing_sample_count
print "INFO: training label list=", training_lbl_cnt_list, ", testing label list=", testing_lbl_cnt_list
print "INFO: labels_list_all=", labels_list_all
print "INFO: training and testing samples generated!"
print 'INFO: running time: %f' % (t1 - t0)
t0 = t1
##############################################
########### Grid Search with CV ##############
##############################################
### get the parameters for cross validation and grid search ###
(cv, model_name, param_dict) = generate_param(ml_opts)
### generate label names (family names) #####
### connect to database to get the column list which contains all column number of the corresponding feature####
if labelnameflag == 1:
label_dic = ml_util.ml_get_label_dict(row_id_str, mongo_tuples, ds_id)
print "INFO: label_dic:", label_dic
else:
label_dic = {}
label_set = set(labels_list_all)
for label_value in label_set:
label_dic[int(label_value)] = str(int(label_value))
print "INFO: generated label_dic:", label_dic
labels_list = []
for key in sorted(label_dic):
labels_list.append(label_dic[key])
#print "labels:", labels_list
class_num = len(labels_list)
if class_num > 2:
print "INFO: Multi-class classification! Number of classes = ", class_num
#### generate training and testing rdd(s) for CV#####
split_prob = 1.0 / float(cv)
split_prob_list = []
for i in range(0, cv):
split_prob_list.append(split_prob)
list_rdd = training_rdd.randomSplit(split_prob_list)
list_train_rdd = []
list_test_rdd = []
for i in range(0, cv):
list_rdd[i].cache()
for i in range(0, cv):
tr_rdd = sc.emptyRDD()
for j in range(0, cv):
if j == i:
pass
else:
tr_rdd = tr_rdd + list_rdd[j]
tr_rdd.cache()
list_train_rdd.append(tr_rdd)
list_test_rdd.append(list_rdd[i])
all_comb_list_of_dic = get_all_combination_list_of_dic(param_dict)
print "INFO: Total number of searching combinations:", len(
all_comb_list_of_dic)
### loop for all parameter combinations and search the best parameters with CV###
results = []
for p in range(0, len(all_comb_list_of_dic)):
params = all_comb_list_of_dic[p]
C = params['C']
iteration_num = params['iterations']
regularization = params['regType']
scores = []
for i in range(0, cv):
train_rdd = list_train_rdd[i]
test_rdd = list_test_rdd[i]
train_number = train_rdd.count()
regP = C / float(train_number)
### build model ###
if model_name == "linear_svm_with_sgd":
#print "====================1: Linear SVM============="
model_classification = SVMWithSGD.train(
train_rdd,
regParam=regP,
iterations=iteration_num,
regType=regularization) # regParam = 1/(sample_number*C)
elif model_name == "logistic_regression_with_lbfgs":
#print "====================2: LogisticRegressionWithLBFGS============="
model_classification = LogisticRegressionWithLBFGS.train(
train_rdd,
regParam=regP,
iterations=iteration_num,
regType=regularization,
numClasses=class_num) # regParam = 1/(sample_number*C)
elif model_name == "logistic_regression_with_sgd":
#print "====================3: LogisticRegressionWithSGD============="
model_classification = LogisticRegressionWithSGD.train(
train_rdd,
regParam=regP,
iterations=iteration_num,
regType=regularization) # regParam = 1/(sample_number*C)
else:
print "ERROR: Training model selection error: no valid ML model selected!"
return
### Evaluating the model on testing data
labelsAndPreds = test_rdd.map(
lambda p: (p.label, model_classification.predict(p.features)))
labelsAndPreds.cache()
test_sample_number = test_rdd.count()
testErr = labelsAndPreds.filter(
lambda (v, p): v != p).count() / float(test_sample_number)
accuracy = 1 - testErr
#print "Accuracy = ", accuracy
scores.append(accuracy)
ss = np.asarray(scores)
#print "%0.3f (+/-%0.03f) for " % (ss.mean(), ss.std() * 2), params
results.append((ss.mean(), ss.std() * 2, params))
sorted_results = sorted(results, key=lambda x: x[0], reverse=1)
(best_accuracy, best_std2, best_param) = sorted_results[0]
print "INFO: ml_opts_jstr=", ml_opts_jstr
print "INFO: best_param=", best_param
#ml_opts=json.loads(ml_opts_jstr);
print "INFO: ml_opts=", ml_opts
##############################################
######output Grid Search results##############
##############################################
json2save = {}
json2save["rid"] = int(row_id_str)
json2save["key"] = "cv_result"
#json2save["param_str"]=ml_opts_jstr
json2save["param_dic"] = param_dict
cv_grid = []
print ""
print "INFO: =====Grid Search Results for SPARK ======"
print "INFO: Best parameters set found for ", model_name, " is: "
for key in best_param:
print "INFO:", key, "=", best_param[key]
if key.lower() == "regtype":
ml_opts['regularization'] = str(best_param[key])
else:
ml_opts[key.lower()] = str(best_param[key]) # add best param to
ml_opts_jstr = json.dumps(ml_opts)
json2save["param_str"] = ml_opts_jstr
print "INFO: Average accuracy with CV = ", cv, ": ", best_accuracy
print ""
print "INFO: Grid scores on development set:"
for i in range(0, len(sorted_results)):
(ave_accu_i, std2_i, param_i) = sorted_results[i]
print "%0.3f (+/-%0.03f) for " % (ave_accu_i, std2_i), param_i
#outstr='%s,%0.3f,%0.03f,%s' % (param_i,ave_accu_i, std2_i,"Selected" if param_i==best_param else "")
outj = {}
outj["param"] = param_i
outj["average_accuracy"] = "%0.3f" % (ave_accu_i)
outj["std_deviation"] = "%0.3f" % (std2_i)
outj["selected"] = "%s" % ("Selected" if param_i == best_param else "")
cv_grid.append(outj)
print " "
t1 = time()
print 'INFO: Grid Search with CV run time: %f' % (t1 - t0)
t0 = time()
##################################################################################
json2save["cv_grid_data"] = cv_grid
cv_result = json.dumps(json2save)
print "INFO: cv_result=", cv_result
filter = '{"rid":' + row_id_str + ',"key":"cv_result"}'
upsert_flag = True
## write to mongoDB.myml.dataset_info, ignore doc with duplicated key
# db.dataset_info.createIndex({"rid":1,"key":1},{unique:true})
ret = query_mongo.upsert_doc_t(mongo_tuples, filter, cv_result,
upsert_flag)
print "INFO: Upsert count for mllib cv_result: ret=", ret
############################################################################################
########### retrain with all training data and generate the final model with results #######
############################################################################################
C = best_param['C']
iteration_num = best_param['iterations']
regularization = best_param['regType']
regP = C / float(training_sample_count)
######################################the rest of the code is the same as train_MLlib.py #####################################################################
if model_name == "linear_svm_with_sgd":
### 1: linearSVM
print "INFO: ====================1: Linear SVM============="
model_classification = SVMWithSGD.train(
training_rdd,
regParam=regP,
iterations=iteration_num,
regType=regularization) # regParam = 1/(sample_number*C)
#print model_classification
elif model_name == "logistic_regression_with_lbfgs":
### 2: LogisticRegressionWithLBFGS
print "INFO: ====================2: LogisticRegressionWithLBFGS============="
model_classification = LogisticRegressionWithLBFGS.train(
training_rdd,
regParam=regP,
iterations=iteration_num,
regType=regularization,
numClasses=class_num) # regParam = 1/(sample_number*C)
elif model_name == "logistic_regression_with_sgd":
### 3: LogisticRegressionWithSGD
print "INFO: ====================3: LogisticRegressionWithSGD============="
model_classification = LogisticRegressionWithSGD.train(
training_rdd,
regParam=regP,
iterations=iteration_num,
regType=regularization) # regParam = 1/(sample_number*C)
else:
print "INFO: Training model selection error: no valid ML model selected!"
return
print "INFO: model type=", type(model_classification)
# create feature coefficient file ================================
coef_arr = None
intercept = None
if model_classification.weights is None:
print "WARNING: model weights not found!"
else:
coef_arr = model_classification.weights.toArray().tolist()
# save to mongo
key = "coef_arr"
ret = ml_util.save_json_t(row_id_str, key, coef_arr, mongo_tuples)
# save intercept to mongo
key = "coef_intercept"
intercept = model_classification.intercept
ret = ml_util.save_json_t(row_id_str, key, intercept, mongo_tuples)
# feature list + coef file =============
feat_filename = os.path.join(local_out_dir,
row_id_str + "_feat_coef.json")
print "INFO: feat_filename=", feat_filename
# create feature list + coef file =============================================== ============
# expect a dict of {"fid":(coef, feature_raw_string)}
jret = ml_util.build_feat_list_t(row_id_str, feat_filename, None, None,
coef_arr, ds_id, mongo_tuples)
# special featuring for IN or libsvm
if jret is None:
jret = ml_util.build_feat_coef_raw_list_t(row_id_str,
feat_filename, coef_arr,
ds_id, mongo_tuples)
if jret is None:
print "WARNING: Cannot create sample list for testing dataset. "
jfeat_coef_dict = jret
print "INFO: coef_arr len=", len(
coef_arr), ", feature_count=", feature_count
# for multi-class
if len(coef_arr) != feature_count:
jfeat_coef_dict = {}
print "WARNING: feature list can't be shown for multi-class classification"
# Calculate prediction and Save testing dataset
bt_coef_arr = sc.broadcast(coef_arr)
bt_intercept = sc.broadcast(intercept)
bt_jfeat_coef_dict = sc.broadcast(jfeat_coef_dict)
### Evaluating the model on testing dataset: label, predict label, score, feature list
print "INFO: intercept=", intercept
print "INFO: coef_arr len=", len(coef_arr)
print "INFO: jfeat_coef_dict len=", len(jfeat_coef_dict)
# get prediction of testing dataset : (tlabel, plabel, score, libsvm, raw feat str, hash) ==============================
if len(coef_arr) == feature_count:
testing_pred_rdd = testing_rdd.map(lambda p: (
p[0].label \
,model_classification.predict(p[0].features) \
,zip_feature_util.calculate_hypothesis(p[0].features, bt_coef_arr.value, bt_intercept.value, model_name) \
,p[0].features \
,p[1] \
) ).cache()
else: # for multi-class, no prediction score;, TBD for better solution: how to display multiple weights for each class
testing_pred_rdd = testing_rdd.map(lambda p: (
p[0].label \
,model_classification.predict(p[0].features) \
,0 \
,p[0].features \
,p[1] \
) ).cache()
# save false prediction to local file
false_pred_fname = os.path.join(local_out_dir,
row_id_str + "_false_pred.json")
print "INFO: false_pred_fname=", false_pred_fname
false_pred_data=testing_pred_rdd.filter(lambda p: p[0] != p[1])\
.map(lambda p: (p[0],p[1],p[2] \
,zip_feature_util.get_dict_coef_raw4feat(zip_feature_util.sparseVector2dict(p[3]), bt_jfeat_coef_dict.value)
,p[4] ) ) \
.collect()
print "INFO: false predicted count=", len(false_pred_data)
false_pred_arr = []
with open(false_pred_fname, "w") as fp:
for sp in false_pred_data:
jsp = {
"tlabel": sp[0],
"plabel": sp[1],
"score": sp[2],
"feat": sp[3],
"hash": sp[4]
}
#print "jsp=",jsp
false_pred_arr.append(jsp)
fp.write(json.dumps(false_pred_arr))
# save prediction results, format: label, prediction, hash
pred_ofname = os.path.join(local_out_dir,
row_id_str + "_pred_output.pkl")
print "INFO: pred_ofname=", pred_ofname
pred_out_arr = testing_pred_rdd.map(lambda p:
(p[0], p[1], p[4])).collect()
ml_util.ml_pickle_save(pred_out_arr, pred_ofname)
### Evaluating the model on testing data
#labelsAndPreds = testing_rdd.map(lambda p: (p.label, model_classification.predict(p.features)))
labelsAndPreds = testing_pred_rdd.map(lambda p: (p[0], p[1]))
labelsAndPreds.cache()
#testing_sample_count = testing_rdd.count()
testErr = labelsAndPreds.filter(lambda (v, p): v != p).count() / float(
testing_sample_count)
accuracy = 1 - testErr
print "INFO: Accuracy = ", accuracy
### Save model
#save_dir = config.get('app', 'HADOOP_MASTER')+'/user/hadoop/yigai/row_6/'
#save_dir = config.get('app', 'HADOOP_MASTER')+config.get('app', 'HDFS_MODEL_DIR')+'/'+row_id_str
save_dir = os.path.join(config.get('app', 'HADOOP_MASTER'),
config.get('app', 'HDFS_MODEL_DIR'), row_id_str)
try:
hdfs.ls(save_dir)
#print "find hdfs folder"
hdfs.rmr(save_dir)
#print "all files removed"
except IOError as e:
print "WARNING: I/O error({0}): {1}".format(
e.errno, e.strerror), ". At HDFS=", save_dir
except:
print "WARNING: Unexpected error:", sys.exc_info(
)[0], ". At HDFS=", save_dir
model_classification.save(sc, save_dir)
###load model if needed
#sameModel = SVMModel.load(sc, save_dir)
t1 = time()
print 'INFO: training run time: %f' % (t1 - t0)
t0 = t1
###############################################
###########plot prediction result figure#######
###############################################
labels = labelsAndPreds.collect()
true_label_list = [x for x, _ in labels]
pred_label_list = [x for _, x in labels]
pred_fname = os.path.join(local_out_dir, row_id_str + "_1" + ".png")
true_fname = os.path.join(local_out_dir, row_id_str + "_2" + ".png")
pred_xlabel = 'Prediction (Single Run)'
true_xlabel = 'True Labels (Single Run)'
test_cnt_dic = ml_util.ml_plot_predict_figures(
pred_label_list, true_label_list, labels_list, label_dic,
testing_sample_count, pred_xlabel, pred_fname, true_xlabel, true_fname)
plt.show()
perf_measures = None
dataset_info = {
"training_fraction": training_fraction,
"class_count": class_num,
"dataset_count": sample_count
}
#############################################################
###################for 2 class only (plot ROC curve)#########
#############################################################
if len(labels_list) == 2:
do_ROC = True
reverse_label_dic = dict((v, k) for k, v in label_dic.items())
if 'clean' in reverse_label_dic:
flag_clean = reverse_label_dic['clean']
elif 'benign' in reverse_label_dic:
flag_clean = reverse_label_dic['benign']
elif '0' in reverse_label_dic:
flag_clean = 0
else:
print "WARNING: No ROC curve generated: 'clean' or '0' must be a label for indicating negative class!"
do_ROC = False
# build data file for score graph
score_graph_fname = os.path.join(local_out_dir,
row_id_str + "_score_graph.json")
print "INFO: score_graph_fname=", score_graph_fname
# build score_arr_0, score_arr_1
# format: tlabel, plabel, score, libsvm, raw feat str, hash
graph_arr = testing_pred_rdd.map(lambda p:
(int(p[0]), float(p[2]))).collect()
score_arr_0 = []
score_arr_1 = []
max_score = 0
min_score = 0
for p in graph_arr:
if p[0] == 0:
score_arr_0.append(p[1])
else:
score_arr_1.append(p[1])
# save max,min score
if p[1] > max_score:
max_score = p[1]
elif p[1] < min_score:
min_score = p[1]
ml_build_pred_score_graph(score_arr_0, score_arr_1, model_name,
score_graph_fname, max_score, min_score)
#print "score_arr_0=",score_arr_0
#print "score_arr_1=",score_arr_1
#print "max_score=",max_score
#print "min_score=",min_score
if do_ROC:
perf_measures = ml_util.calculate_fscore(true_label_list,
pred_label_list)
print "RESULT: perf_measures=", perf_measures
model_classification.clearThreshold()
scoreAndLabels = testing_rdd.map(lambda p: (
model_classification.predict(p[0].features), int(p[0].label)))
#metrics = BinaryClassificationMetrics(scoreAndLabels)
#areROC = metrics.areaUnderROC
#print areROC
scoreAndLabels_list = scoreAndLabels.collect()
if flag_clean == 0:
scores = [x for x, _ in scoreAndLabels_list]
s_labels = [x for _, x in scoreAndLabels_list]
testing_N = test_cnt_dic[0]
testing_P = test_cnt_dic[1]
else:
scores = [-x for x, _ in scoreAndLabels_list]
s_labels = [1 - x for _, x in scoreAndLabels_list]
testing_N = test_cnt_dic[1]
testing_P = test_cnt_dic[0]
#print scores
#print s_labels
# create ROC data file ======== ====
roc_auc = ml_create_roc_files(row_id_str, scores, s_labels,
testing_N, testing_P, local_out_dir,
row_id_str)
perf_measures["roc_auc"] = roc_auc
# only update db for web request
if fromweb == "1":
#print "database update"
str_sql="UPDATE atdml_document set "+"accuracy = '"+str(accuracy*100)+"%" \
+"', status = 'learned', processed_date ='"+str(datetime.datetime.now()) \
+"',ml_opts='"+ml_opts_jstr \
+"', perf_measures='"+json.dumps(perf_measures) \
+"', dataset_info='"+json.dumps(dataset_info) \
+"' where id="+row_id_str
ret = exec_sqlite.exec_sql(str_sql)
print "INFO: Data update done! ret=", str(ret)
else:
print "INFO: accuracy = '" + str(accuracy * 100) + "%"
t1 = time()
print 'INFO: total run time: %f' % (t1 - t00)
print 'INFO: Finished!'
return 0
def pca(row_id_str, ds_id, hdfs_feat_dir, local_out_dir
, sp_master, spark_rdd_compress, spark_driver_maxResultSize, sp_exe_memory, sp_core_max
, zipout_dir, zipcode_dir, zip_file_name
, mongo_tuples, fromweb, pca_jstr
, jobname, model_data_folder ):
# create zip files for Spark workers ================= ================
zip_file_path = ml_build_zip_file(zipout_dir, zipcode_dir, zip_file_name, prefix='zip_feature_util')
print "INFO: zip_file_path=",zip_file_path
# init Spark context ====
sc=ml_util.ml_get_spark_context(sp_master
, spark_rdd_compress
, spark_driver_maxResultSize
, sp_exe_memory
, sp_core_max
, jobname
, [zip_file_path])
pca_param=json.loads(pca_jstr)
if "k" in pca_param:
k=pca_param["k"]
else:
k=None
if "threshold" in pca_param:
threshold=pca_param["threshold"]
else:
threshold=None
if "lib" in pca_param:
lib=pca_param["lib"]
else:
lib='mllib'
ret=-1
# start here =================================================================== ===============
t0 = time()
# source libsvm filename
libsvm_data_file = os.path.join(hdfs_feat_dir , "libsvm_data")
print "INFO: libsvm_data_file=", libsvm_data_file
# load sample RDD from text file
# format Row(label, features, hash) from get_sample_dataframe()
samples_df, feature_count = zip_feature_util.get_sample_dataframe(sc, libsvm_data_file, 0, None)
print "INFO: feature_count=",feature_count
#df_pcaed format: hash,label, features
(df_pcaed, k, pca_model)=PCA_transform(sc, samples_df,feature_count, threshold, k)
print "INFO: Doing PCA... threshold=",threshold,",k=",k
#print "df_pcaed=",df_pcaed.first()
#print "k=",k
#print "pca_model=",pca_model
#print "pc=",pca_model.pc
# pca model filename ============================= ===============
if model_data_folder is None:
if row_id_str != ds_id:
# get from parent dataset
model_data_folder = os.path.join(config.get('app', 'HADOOP_MASTER'),config.get('app', 'HDFS_MODEL_DIR'), ds_id+"_pca")
else:
model_data_folder = os.path.join(config.get('app', 'HADOOP_MASTER'),config.get('app', 'HDFS_MODEL_DIR'), row_id_str+"_pca")
# create HDFS folder
try:
hdfs.mkdir(model_data_folder)
except IOError as e:
print "WARNING: I/O error({0}): {1}".format(e.errno, e.strerror),". At HDFS=", save_dir
except:
print "WARNING: Unexpected error:", sys.exc_info()[0] ,". At HDFS=", save_dir
if not threshold is None:
#pca_fname=os.path.join(hdfs_feat_dir , row_id_str+'_pca_'+str(threshold)+'.ml')
pca_fname=os.path.join(model_data_folder , 'pca_model_'+str(threshold))
libsvm_data_pca = os.path.join(hdfs_feat_dir , "libsvm_data_pca_"+str(threshold)+'.ml')
else:
pca_fname=os.path.join(model_data_folder , 'pca_model_'+str(k))
libsvm_data_pca = os.path.join(hdfs_feat_dir , "libsvm_data_pca_"+str(k)+'.ml')
# save pca model to HDFS ===============
print "INFO: pca_fname=",pca_fname
pca_model.write().overwrite().save(pca_fname)
# save pca data to HDFS ============================= ===============
print "INFO: libsvm_data_pca=",libsvm_data_pca
# construct libsvm string
libsvm_rdd=df_pcaed.rdd.map(lambda p: p[0]+" "+str(int(p[1]))+zip_feature_util.dv2libsvm(p[2].toArray()))
# clean up old libsvm file ============================= ===============
try:
hdfs.rmr(libsvm_data_pca)
except IOError as e:
print "WARNING: I/O error({0}): {1}".format(e.errno, e.strerror)
except:
print "WARNING: Unexpected error at rmr():", sys.exc_info()[0]
# overwrite pca file at hdfs
libsvm_rdd.saveAsTextFile(libsvm_data_pca)
t1 = time()
print 'INFO: PCA processing time: %f' %(t1-t0)
### insert pca_param into mongoDB ###
filter='{"rid":'+row_id_str+',"key":"pca_param"}'
if not threshold is None:
pca_param["threshold"]=threshold
if not k is None:
pca_param["k"]=k
print "INFO: pca_param=",pca_param
upsert_flag=True
jstr_insert = '{ "rid":'+row_id_str+',"key":"pca_param", "value":'+json.dumps(pca_param)+'}'
ret=query_mongo.upsert_doc_t(mongo_tuples,filter,jstr_insert,upsert_flag)
print "INFO: Upsert count for pca_param=",ret
# only update db for web request ===========
if fromweb=="1":
#print "database update"
str_sql="UPDATE atdml_document set " \
+" status = 'pca-ed', processed_date ='"+str(datetime.datetime.now()) \
+"' , ml_pca_opts = '"+json.dumps(pca_param) \
+"' where id="+row_id_str
ret=exec_sqlite.exec_sql(str_sql)
print "INFO: Update Sqlite DB done! ret=", str(ret)
t1 = time()
print 'INFO: running time: %f' %(t1-t0)
#print 'Finished!'
return 0
def feat_extraction(row_id_str,
hdfs_dir_list,
hdfs_feat_dir,
model_data_folder,
sp_master,
spark_rdd_compress,
spark_driver_maxResultSize,
sp_exe_memory,
sp_core_max,
zipout_dir,
zipcode_dir,
zip_file_name,
mongo_tuples,
fromweb,
label_arr,
metadata_count,
label_idx,
data_idx,
pattern_str,
ln_delimitor,
data_field_list,
jkey_dict,
jobname,
num_gram,
feature_count_threshold,
token_dict=None,
HDFS_RETR_DIR=None,
remove_duplicated="N",
cust_featuring=None,
cust_featuring_params=None,
local_out_dir=None,
filter_ratio=None,
binary_flag=False):
# zip func in other files for Spark workers ================= ================
zip_file_path = ml_util.ml_build_zip_file(zipout_dir,
zipcode_dir,
zip_file_name,
user_custom=cust_featuring)
# get_spark_context
spark = ml_util.ml_get_spark_session(sp_master, spark_rdd_compress,
spark_driver_maxResultSize,
sp_exe_memory, sp_core_max, jobname,
zip_file_path)
if spark:
sc = spark.sparkContext
# log time ================================================================ ================
t0 = time()
# input filename
input_filename = "*"
ext_type = '.gz'
gz_list = None
# single hdfs file
if not ',' in hdfs_dir_list: # single dir having *.gz ==== =========
# read raw data from HDFS as .gz format ==========
hdfs_files = os.path.join(hdfs_dir_list, input_filename + ext_type)
# check if gz files in hdfs ============
try:
gz_list = hdfs.ls(hdfs_dir_list)
print "INFO: check hdfs folder=", hdfs_dir_list
except IOError as e:
print "WARNING: I/O error({0}): {1}".format(e.errno, e.strerror)
except:
print "WARNING: Error at checking HDFS file:", sys.exc_info()[0]
# use whole folder
#print "gz_list",gz_list
if gz_list is None or len(gz_list) == 0:
print "ERROR: No file found by ", input_filename + ext_type #,", use",hdfs_dir_list,"instead"
return -2
elif len(gz_list) == 1:
# use dir as filename
hdfs_files = hdfs_dir_list[0:-1]
else: # multiple dirs ==== =========
hdfs_files = ""
cnt = 0
temp_lbl_list = []
comma = ""
print "INFO: before label_arr=", label_arr
# check each folder
for dr in hdfs_dir_list.split(','):
#print "****=",dr
if not len(dr) > 0:
continue
try:
# remove space etc.
dr = dr.strip()
fdr = os.path.join(HDFS_RETR_DIR, dr)
# ls didn't like "*"
if '*' in fdr:
#gz_list=hdfs.ls(fdr.replace("*",""))
dn = os.path.dirname(fdr).strip()
bn = os.path.basename(
fdr).strip() #print "dn=",dn,",bn=",bn
# get all names under folder and do filtering
gz_list = fnmatch.filter(hdfs.ls(dn), '*' + bn)
else:
gz_list = hdfs.ls(fdr)
cnt = cnt + len(gz_list)
if len(gz_list) > 0:
hdfs_files = hdfs_files + comma + fdr
comma = ","
except IOError as e:
print "WARNING: I/O error({0}): {1}".format(
e.errno, e.strerror)
except:
print "WARNING: Error at checking HDFS file:", sys.exc_info(
)[0]
# use whole folder
if cnt is None or cnt == 0:
print "ERROR: No file found at", hdfs_files
return -2
else:
print "INFO: total file count=", cnt
# set convert flag only when multiple dir and label_arr has dirty label
if not label_arr is None and len(
label_arr) == 2 and label_arr[1] == "dirty":
convert2dirty = "Y"
print "INFO: hdfs_dir_list=", hdfs_dir_list
print "INFO: hdfs_files=", hdfs_files
cust_featuring_jparams = None
# custom featuring
if not cust_featuring is None and len(cust_featuring) > 0:
# load user module =======
user_func, cust_featuring_jparams = get_user_custom_func(
cust_featuring, cust_featuring_params)
# TBD apply user_func
all_hashes_cnt_dic = None
all_hash_str_dic = None
all_hashes_seq_dic = None
else:
print "ERROR: custom featuring type is needed"
print "INFO: cust_featuring=", cust_featuring, "cust_featuring_jparams=", cust_featuring_jparams
dnn_flag = False
has_header = None
label_col = None
label_index = None
# get featuring params
if cust_featuring_jparams:
if 'label_index' in cust_featuring_jparams: # idx number for label, 0 based
label_index = cust_featuring_jparams['label_index']
if 'has_header' in cust_featuring_jparams: # True/False
has_header = eval(cust_featuring_jparams['has_header'])
if has_header == 1:
has_header = True
if 'dnn_flag' in cust_featuring_jparams: # True/False
dnn_flag = cust_featuring_jparams['dnn_flag']
if dnn_flag == 1:
dnn_flag = True
elif dnn_flag == 0:
dnn_flag = False
if label_index is None:
label_index = 0
elif not isinstance(label_index, int):
label_index = eval(label_index)
print "INFO: label_index=", label_index, ",has_header=", has_header, ",dnn_flag=", dnn_flag
# read as DataFrame ===============================================
df = spark.read.csv(hdfs_files, header=has_header)
df.show()
print "INFO: col names=", df.columns
# get column name for label
label_col = None
for i, v in enumerate(df.columns):
if i == label_index:
label_col = v
# get all distinct labels into an array =============== provided by parameter?
if label_arr is None and not label_col is None:
label_arr = sorted([
rw[label_col] for rw in df.select(label_col).distinct().collect()
])
print "INFO: label_arr=", label_arr
label_dic = {}
# convert label_arr to dict; {label:number|
for idx, label in enumerate(sorted(label_arr)):
if not label in label_dic:
label_dic[
label] = idx #starting from 0, value = idx, e.g., clean:0, dirty:1
# add params for dataframe conversion
cust_featuring_jparams["label_dict"] = label_dic
# convert to int
cust_featuring_jparams["label_index"] = label_index
featuring_params = json.dumps(cust_featuring_jparams)
# convert DataFrame row to libsvm string
libsvm_rdd = df.rdd.map(lambda x: user_func(list(x), featuring_params))
print "INFO: sample df row=", (libsvm_rdd.collect()[0])
print "INFO: featuring_params=", featuring_params
total_input_count = df.count()
print "INFO: Total input sample count=", total_input_count
#print "INFO: feature_count_threshold=",feature_count_threshold
#get all hashes and total occurring count ===============
# all_hashes_cnt_dic: {'col index': total count,... }
# build all_hashes_cnt_dic
cnt_df = df.select(
[count(when(~isnull(c), c)).alias(c) for c in df.columns])
#cnt_df.show()
cnt_arr = cnt_df.rdd.map(lambda x: list(x)).collect()
feat_sample_count_arr = cnt_arr[0]
#print "feat_sample_count_arr=",feat_sample_count_arr
if all_hashes_cnt_dic is None:
all_hashes_cnt_dic = {}
idx = 1
for i, v in enumerate(feat_sample_count_arr):
if i != label_index:
all_hashes_cnt_dic[idx] = v
idx += 1
#print "all_hashes_cnt_dic=",all_hashes_cnt_dic
#get all hashes and their extracted string ===============
# all_hash_str_dic: {hash:'str1', ...
if all_hash_str_dic is None:
# convert header to dict=index:string; excude label column
all_hash_str_dic = {}
idx = 1
for i, v in enumerate(df.schema.names):
if i != label_index:
all_hash_str_dic[idx] = v
idx += 1
#print "all_hash_str_dic=",all_hash_str_dic
# save labels to hdfs as text file==================================== ============
hdfs_folder = hdfs_feat_dir #+ "/" # "/" is needed to create the folder correctly
print "INFO: hdfs_folder=", hdfs_folder
try:
hdfs.mkdir(hdfs_folder)
except IOError as e:
print "WARNING: I/O error({0}): {1}".format(e.errno, e.strerror)
except:
print "WARNING: Unexpected error at mkdir:", sys.exc_info()[0]
# clean up metadata_file
metadata_file = os.path.join(hdfs_folder, metadata) #"metadata"
print "INFO: metadata_file=", metadata_file
try:
hdfs.rmr(metadata_file)
except IOError as e:
print "WARNING: I/O error({0}): {1}".format(e.errno, e.strerror)
except:
print "WARNING: Unexpected error at rmr():", sys.exc_info()[0]
sc.parallelize(label_arr, 1).saveAsTextFile(metadata_file)
#remap all hash values to continuous key/feature number ==============
# all_hashes_seq_dic: { hash : sequential_numb }
if all_hashes_seq_dic is None:
all_hashes_seq_dic = {}
# csv column index as sequentail number
remap2seq(all_hash_str_dic, all_hashes_seq_dic)
#print "all_hashes_seq_dic=",all_hashes_seq_dic
total_feature_numb = len(all_hashes_seq_dic)
print "INFO: Total feature count=", len(all_hashes_seq_dic)
# save feat_sample_count_arr data ==================================== ============
filter = '{"rid":' + row_id_str + ',"key":"feat_sample_count_arr"}'
upsert_flag = True
jo_insert = {}
jo_insert["rid"] = eval(row_id_str)
jo_insert["key"] = "feat_sample_count_arr"
jo_insert["value"] = feat_sample_count_arr
jstr_insert = json.dumps(jo_insert)
ret = query_mongo.upsert_doc_t(mongo_tuples, filter, jstr_insert,
upsert_flag)
print "INFO: Upsert count for feat_sample_count_arr=", ret
# insert failed, save to local
if ret == 0:
# drop old record in mongo
ret = query_mongo.delete_many(mongo_tuples, None, filter)
if not os.path.exists(local_out_dir):
os.makedirs(local_out_dir)
fsca_hs = os.path.join(local_out_dir, row_id_str,
row_id_str + "_feat_sample_count_arr.pkl")
print "WARNING: save feat_sample_count_arr to local"
ml_util.ml_pickle_save(feat_sample_count_arr, fsca_hs)
# get rdd statistics info
# remove duplicated libsvm string; only keep the first duplicated item, assume space following key_idx
if remove_duplicated == "Y":
libsvm_rdd=libsvm_rdd \
.map(lambda x: ( ','.join(x.split(' ')[metadata_count:]), x)) \
.groupByKey().map(lambda x: list(x[1])[0] ) \
.cache()
cnt_list = libsvm_rdd.map(lambda x: (x.split(' ')[1], 1)).reduceByKey(
add).collect()
stats = libsvm_rdd.map(
lambda x: len(x.split(' ')[metadata_count:])).stats()
feat_count_max = stats.max()
feat_count_stdev = stats.stdev()
feat_count_mean = stats.mean()
sample_count = stats.count()
print "INFO: Non-Duplicated libsvm data: sample count=", sample_count, ",Feat count mean=", feat_count_mean, ",Stdev=", feat_count_stdev
print "INFO: ,max feature count=", feat_count_max
print "INFO: Non-Duplicated Label count list=", cnt_list
# clean up libsvm data ==================================== ============
libsvm_data_file = os.path.join(hdfs_folder,
libsvm_alldata_filename) #"libsvm_data"
print "INFO: libsvm_data_file=", libsvm_data_file
try:
hdfs.rmr(libsvm_data_file)
except IOError as e:
print "WARNING: I/O error({0}): {1} at libsvm_data_file clean up".format(
e.errno, e.strerror)
except:
print "WARNING: Unexpected error at libsvm file clean up:", sys.exc_info(
)[0]
#codec = "org.apache.hadoop.io.compress.GzipCodec"
#libsvm_rdd.saveAsTextFile(libsvm_data_file, codec)
libsvm_rdd.saveAsTextFile(libsvm_data_file) # TBD encrypted
feat_count_file = libsvm_data_file + "_feat_count"
print "INFO: feat_count_file=", feat_count_file
try:
hdfs.rmr(feat_count_file)
except IOError as e:
print "WARNING: I/O error({0}): {1} at feat_count clean up".format(
e.errno, e.strerror)
except:
print "WARNING: Unexpected error at libsvm feature count clean up:", sys.exc_info(
)[0]
sc.parallelize([total_feature_numb], 1).saveAsTextFile(feat_count_file)
# TBD ??? output text for DNN:[meta-data1,meta-data2,..., [feature tokens]] ================= DNN ===========
if dnn_flag: # special flag to tokenize and keep input orders
print "INFO: processing data for DNN..."
# create token dict
# str_hash_dict: string to hash
# all_hashes_seq_dic: hash to seq id
if token_dict is None or len(token_dict) == 0:
token_dict = {}
str_hash_dict = {v: k for k, v in all_hash_str_dic.iteritems()}
for k, v in str_hash_dict.iteritems():
token_dict[k] = int(all_hashes_seq_dic[str(v)])
#print "token_dict=",len(token_dict),token_dict
# TBD here: need to implement non-binary feature
dnn_rdd=df.rdd \
.map(lambda x: tokenize_by_dict(x, data_idx, token_dict,label_idx, label_dic)) \
.filter(lambda x: len(x) > metadata_count) \
.filter(lambda x: type(x[metadata_count]) is list)
#.cache()
# filter duplication here
#print dnn_rdd.take(3)
dnn_data_file = os.path.join(hdfs_folder,
dnn_alldata_filename) #"dnn_data"
print "INFO: dnn_data_file=", dnn_data_file
try:
hdfs.rmr(dnn_data_file)
except IOError as e:
print "WARNING: I/O error({0}): {1} at dnn_data_file clean up".format(
e.errno, e.strerror)
except:
print "WARNING: Unexpected error at libsvm file clean up:", sys.exc_info(
)[0]
# clean up data
dnn_npy_gz_file = os.path.join(hdfs_folder, row_id_str + "_dnn_")
print "INFO: dnn_npy_gz_file=", dnn_npy_gz_file
try:
hdfs.rmr(dnn_npy_gz_file + "data.npy.gz")
hdfs.rmr(dnn_npy_gz_file + "label.npy.gz")
hdfs.rmr(dnn_npy_gz_file + "info.npy.gz")
except IOError as e:
print "WARNING: I/O error({0}): {1} at dnn_npy clean up".format(
e.errno, e.strerror)
except:
print "WARNING: Unexpected error at dnn_npy file clean up:", sys.exc_info(
)[0]
# save new data
try:
dnn_rdd.saveAsTextFile(dnn_data_file)
except:
print "WARNING: Unexpected error at saving dnn data:", sys.exc_info(
)[0]
# show data statistics
try:
stats = dnn_rdd.map(lambda p: len(p[metadata_count])).stats()
feat_count_max = stats.max()
feat_count_stdev = stats.stdev()
feat_count_mean = stats.mean()
sample_count = stats.count()
print "INFO: DNN data: sample count=", sample_count, ",Feat count mean=", feat_count_mean, ",Stdev=", feat_count_stdev
print "INFO: ,max feature count=", feat_count_max
except:
print "WARNING: Unexpected error at getting stats of dnn_rdd:", sys.exc_info(
)[0]
# clean up pca data in hdfs ============ ========================
pca_files = '*' + libsvm_alldata_filename + "_pca_*"
#print "INFO: pca_files=", pca_files
try:
f_list = hdfs.ls(hdfs_folder)
if len(f_list) > 0:
df_list = fnmatch.filter(f_list, pca_files)
for f in df_list:
print "INFO: rm ", f
hdfs.rmr(f)
except IOError as e:
print "WARNING: I/O error({0}): {1}".format(e.errno, e.strerror)
except:
print "WARNING: Unexpected error at libsvm pca file clean up:", sys.exc_info(
)[0]
# clean up pca data in web local ============ ========================
pca_fname = os.path.join(model_data_folder, row_id_str + '_pca_*.pkl*')
print "INFO: pca_fname=", pca_fname
try:
for fl in glob.glob(pca_fname):
print "INFO: remove ", fl
os.remove(fl)
except OSError, e:
print("Error: %s - %s." % (e.pca_fname, e.strerror))
def feat_extr_ngram(row_id_str,
hdfs_dir_list,
hdfs_feat_dir,
model_data_folder,
sp_master,
spark_rdd_compress,
spark_driver_maxResultSize,
sp_exe_memory,
sp_core_max,
zipout_dir,
zipcode_dir,
zip_file_name,
mongo_tuples,
fromweb,
label_arr,
metadata_count,
label_idx,
data_idx,
pattern_str,
ln_delimitor,
data_field_list,
jkey_dict,
jobname,
num_gram,
feature_count_threshold,
token_dict=None,
HDFS_RETR_DIR=None,
remove_duplicated="N",
cust_featuring=None,
cust_featuring_params=None,
local_out_dir=None,
filter_ratio=None,
binary_flag=True):
# zip func in other files for Spark workers ================= ================
zip_file_path = ml_util.ml_build_zip_file(zipout_dir,
zipcode_dir,
zip_file_name,
user_custom=cust_featuring)
# get_spark_context
sc = ml_util.ml_get_spark_context(sp_master, spark_rdd_compress,
spark_driver_maxResultSize,
sp_exe_memory, sp_core_max, jobname,
[zip_file_path])
# log time ================================================================ ================
t0 = time()
# input filename
input_filename = "*"
ext_type = '.gz'
gz_list = None
convert2dirty = "N"
if not ',' in hdfs_dir_list: # single dir having *.gz ==== =========
# read raw data from HDFS as .gz format ==========
rdd_files = os.path.join(hdfs_dir_list, input_filename + ext_type)
# check if gz files in hdfs ============
try:
gz_list = hdfs.ls(hdfs_dir_list)
print "INFO: check hdfs folder=", hdfs_dir_list
except IOError as e:
print "WARNING: I/O error({0}): {1}".format(e.errno, e.strerror)
except:
print "WARNING: Error at checking HDFS file:", sys.exc_info()[0]
# use whole folder
if gz_list is None or len(gz_list) == 0:
print "ERROR: No file found by ", input_filename + ext_type #,", use",hdfs_dir_list,"instead"
return -2
elif len(gz_list) == 1:
# use dir as filename
rdd_files = hdfs_dir_list[0:-1]
else: # multiple dirs ==== =========
rdd_files = ""
cnt = 0
temp_lbl_list = []
comma = ""
print "INFO: before label_arr=", label_arr
# check each folder
for dr in hdfs_dir_list.split(','):
#print "****=",dr
if not len(dr) > 0:
continue
try:
# remove space etc.
dr = dr.strip()
fdr = os.path.join(HDFS_RETR_DIR, dr)
#print "fdr=",fdr
# ls didn't like "*"
if '*' in fdr:
#gz_list=hdfs.ls(fdr.replace("*",""))
dn = os.path.dirname(fdr).strip()
bn = os.path.basename(fdr).strip()
#print "dn=",dn,",bn=",bn
# get all names under folder and do filtering
gz_list = fnmatch.filter(hdfs.ls(dn), '*' + bn)
#print "gz_list=",gz_list
else:
gz_list = hdfs.ls(fdr)
cnt = cnt + len(gz_list)
if len(gz_list) > 0:
rdd_files = rdd_files + comma + fdr
comma = ","
except IOError as e:
print "WARNING: I/O error({0}): {1}".format(
e.errno, e.strerror)
except:
print "WARNING: Error at checking HDFS file:", sys.exc_info(
)[0]
# use whole folder
if cnt is None or cnt == 0:
print "ERROR: No file found at", rdd_files
return -2
else:
print "INFO: total file count=", cnt
# set convert flag only when multiple dir and label_arr has dirty label
#if label_arr is None: # create label arr if None
# label_arr=temp_lbl_list
if not label_arr is None and len(
label_arr) == 2 and label_arr[1] == "dirty":
convert2dirty = "Y"
print "INFO: rdd_files=", rdd_files
txt_rdd = sc.textFile(rdd_files) #, use_unicode=False
total_input_count = txt_rdd.count()
print "INFO: Total input sample count=", total_input_count
# debug only
#for x in txt_rdd.collect():
# print "t=",x
print "INFO: hdfs_dir_list=", hdfs_dir_list
print "INFO: label_arr=", label_arr
print "INFO: feature_count_threshold=", feature_count_threshold
#jkey_dict={"meta_list":["label","md5","mdate"], "data_key":"logs"}
# this dict depends on the format of input data
if not data_field_list is None:
jkey_dict = json.loads(jkey_dict)
data_key = jkey_dict["data_key"]
meta_list = jkey_dict["meta_list"]
metadata_count = len(meta_list)
data_idx = metadata_count
print "INFO: jkey_dict=", jkey_dict
print "INFO: meta_list=", meta_list
print "INFO: data_key=", data_key
print "INFO: data_field_list=", data_field_list
print "INFO: metadata_count=", metadata_count
featured_rdd = txt_rdd \
.map(lambda x: preprocess_json(x,meta_list,data_key,data_field_list)) \
.filter(lambda x: len(x) > metadata_count) \
.filter(lambda x: type(x[metadata_count]) is list) \
.map(lambda x: feature_extraction_ngram(x, data_idx, MAX_FEATURES, num_gram)) \
.filter(lambda x: len(x) > metadata_count) \
.filter(lambda x: type(x[metadata_count]) is dict) \
.filter(lambda x: type(x[metadata_count+1]) is dict) \
.filter(lambda x: len(x[metadata_count])> int(feature_count_threshold) ) \
.cache()
#print "INFO: featured_rdd="
#for x in featured_rdd.collect():
# print "INFO: **** f=",x
# user custom code for featuring ============================================= ==========
# input txt_rdd format (string): each text row for each sample
# output featured_rdd format (list):[meta-data1,meta-data2,..., hash_cnt_dic, hash_str_dic]
elif not cust_featuring is None and len(cust_featuring) > 0:
user_module = None
user_func = None
user_func_dnn = None
# load user module =======
try:
modules = map(__import__, [CUSTOM_PREFIX + cust_featuring])
user_module = modules[0]
user_func = getattr(user_module, CUSTOM_FUNC)
except Exception as e:
print "ERROR: module=", CUSTOM_PREFIX + cust_featuring
print "ERROR: user module error.", e.__doc__, e.message
return -101
try:
jparams = json.loads(cust_featuring_params)
if jparams and 'n-gram' in jparams:
num_gram = jparams['n-gram']
elif jparams and 'ngram' in jparams:
num_gram = jparams['ngram']
if jparams and 'binary_flag' in jparams:
binary_flag = eval(jparams['binary_flag'])
except Exception as e:
print "ERROR: user params error.", e.__doc__, e.message
return -200
# convert feast into array. output format: [ meta1,meta2,..., [feat1,feat2,...]]
tmp_rdd = txt_rdd.map(lambda x: user_func(x, cust_featuring_params)) \
.filter(lambda x: len(x) > metadata_count) \
.filter(lambda x: type(x[metadata_count]) is list).cache()
print " tmp_rdd cnt=", tmp_rdd.count(
), ",ix=", data_idx, ",max f=", MAX_FEATURES, "ngram=", num_gram
print "take(1) rdd=", tmp_rdd.take(1)
# TBD for multivariant output format: [ meta1,meta2,..., [[feat1,feat2,...],[feat1,feat2,...],...]]
# TBD only for num_gram available
# for traditional ML, feat in a dict
# output format: [ meta1,meta2,..., [[feat1,feat2,...],[feat1,feat2,...],...]]
featured_rdd = tmp_rdd \
.map(lambda x: feature_extraction_ngram(x, data_idx, MAX_FEATURES, num_gram)) \
.filter(lambda x: len(x) > metadata_count) \
.filter(lambda x: type(x[metadata_count]) is dict) \
.filter(lambda x: type(x[metadata_count+1]) is dict) \
.filter(lambda x: len(x[metadata_count])> int(feature_count_threshold) ) \
.cache()
all_hashes_cnt_dic = None
all_hash_str_dic = None
all_hashes_seq_dic = None
else:
print "INFO: pattern_str=", pattern_str + "<--"
print "INFO: ln_delimitor=", ln_delimitor + "<--"
print "INFO: label_idx=", label_idx
print "INFO: data_idx=", data_idx
print "INFO: metadata_count=", metadata_count
print "INFO: filter_ratio=", filter_ratio
# filter top and least percentage of feature
if not filter_ratio is None and filter_ratio > 0 and filter_ratio < 1:
# check total count here before continue
upper_cnt = total_input_count * (1 - filter_ratio)
lower_cnt = total_input_count * filter_ratio
# set limit for lower bound. if total count is large, lower_cnt may exclude all features...
# max lower count = min( MAX_FILTER_LOWER_CNT, total_input_count/100 )
if not MAX_FILTER_LOWER_CNT is None and lower_cnt > MAX_FILTER_LOWER_CNT:
if MAX_FILTER_LOWER_CNT > total_input_count / 100:
lower_cnt = total_input_count / 100
else:
lower_cnt = MAX_FILTER_LOWER_CNT
print "INFO: filtering by count, upper bound=", upper_cnt, ",lower bound=", lower_cnt
# find unique feature, count them, remove them if in highest and lowest % and then create a dict
f_feat_set = Set (txt_rdd.map(lambda x:x.split(ln_delimitor)).flatMap(lambda x:Set(x[metadata_count:])) \
.map(lambda x:(x,1)).reduceByKey(lambda a, b: a + b) \
.filter(lambda x:x[1]<= upper_cnt and x[1]>= lower_cnt) \
.map(lambda x:x[0]).collect() )
print "INFO: f_feat_set len=", len(f_feat_set)
broadcast_f_set = sc.broadcast(f_feat_set)
#txt_rdd=txt_rdd.map(lambda x: filter_by_list(x, metadata_count,ln_delimitor, broadcast_f_list.value ))
txt_rdd=txt_rdd.map(lambda x: x.split(ln_delimitor)) \
.map(lambda x: x[:metadata_count]+ [w for w in x[metadata_count:] if w and w in broadcast_f_set.value]) \
.map(lambda x: ln_delimitor.join(x))
# preprocess by pattern matching and then extract n-gram features #.encode('UTF8')
# input txt_rdd format (string): meta-data1\tmeta-data2\t...\tdataline1\tdataline2\t...datalineN\n
# output featured_rdd format (list):[meta-data1,meta-data2,..., hash_cnt_dic, hash_str_dic]
# hash_cnt_dic: {hash,hash:count,...} hash_str_dic: {hash: 'str1',... }
tmp_rdd = txt_rdd \
.map(lambda x: preprocess_pattern(x, metadata_count, pattern_str, ln_delimitor \
, label_idx, label_arr, convert2dirty )) \
.filter(lambda x: len(x) > metadata_count) \
.filter(lambda x: type(x[metadata_count]) is list) #.cache() memory issue...
#tmp_rdd_count=tmp_rdd.count()
#print "INFO: After preprocessing count=",tmp_rdd_count
featured_rdd = tmp_rdd \
.map(lambda x: feature_extraction_ngram(x, data_idx, MAX_FEATURES, num_gram)) \
.filter(lambda x: len(x) > metadata_count) \
.filter(lambda x: type(x[metadata_count]) is dict) \
.filter(lambda x: type(x[metadata_count+1]) is dict) \
.filter(lambda x: len(x[metadata_count])> int(feature_count_threshold) ) \
.cache()
#feat_rdd_count=featured_rdd.count()
#print "INFO: After featuring count=",feat_rdd_count
all_hashes_cnt_dic = None
all_hash_str_dic = None
all_hashes_seq_dic = None
#get all hashes and total occurring count ===============
# all_hashes_cnt_dic: {'hash,hash': total count,... }
if all_hashes_cnt_dic is None:
#all_hashes_cnt_dic = featured_rdd.map(lambda x: x[metadata_count]).reduce(lambda a, b: combine_dic_cnt(a, b))
all_hashes_cnt_dic = dict(
featured_rdd.flatMap(lambda x: x[metadata_count].items()).
reduceByKey(lambda a, b: a + b).collect())
#get all hashes and their extracted string ===============
# all_hash_str_dic: {hash:'str1', ...
if all_hash_str_dic is None:
#all_hash_str_dic = featured_rdd.map(lambda x: x[metadata_count+1]).reduce(lambda a, b: combine_dic(a, b))
all_hash_str_dic = dict(
featured_rdd.flatMap(
lambda x: x[metadata_count + 1].items()).distinct().collect())
# get all labels into an array =============== provided by parameter?
if label_arr is None:
# will force "clean" be 0 here
label_arr = sorted(
featured_rdd.map(
lambda x: x[label_idx].lower()).distinct().collect())
# debug only
print "INFO: label_arr.=", json.dumps(sorted(label_arr))
# save labels to hdfs as text file==================================== ============
hdfs_folder = hdfs_feat_dir #+ "/" # "/" is needed to create the folder correctly
print "INFO: hdfs_folder=", hdfs_folder
try:
hdfs.mkdir(hdfs_folder)
except IOError as e:
print "WARNING: I/O error({0}): {1}".format(e.errno, e.strerror)
except:
print "WARNING: Unexpected error at mkdir:", sys.exc_info()[0]
# clean up metadata_file
metadata_file = os.path.join(hdfs_folder, metadata) #"metadata"
print "INFO: metadata_file=", metadata_file
try:
hdfs.rmr(metadata_file)
except IOError as e:
print "WARNING: I/O error({0}): {1}".format(e.errno, e.strerror)
except:
print "WARNING: Unexpected error at rmr():", sys.exc_info()[0]
sc.parallelize(label_arr, 1).saveAsTextFile(metadata_file)
#remap all hash values to continuous key/feature number ==============
# all_hashes_seq_dic: { hash : sequential_numb }
if all_hashes_seq_dic is None:
all_hashes_seq_dic = {}
remap2seq(
all_hashes_cnt_dic,
all_hashes_seq_dic) #all_hashes_seq_dic has continuous key number
#print "all_hashes_seq_dic=",all_hashes_seq_dic
total_feature_numb = len(all_hashes_seq_dic)
print "INFO: Total feature count=", len(all_hashes_seq_dic)
# featured_rdd (list): [meta-data1,meta-data2,..., hash_cnt_dic, hash_str_dic]
# seq_featured_rdd(list): [meta-data1,meta-data2,..., hash_cnthsh_dict, hash_str_dic] (feat id in sorted sequence)
# hash_cnt_dic: {hash: count} hash_str_dic: {hash: 'str1,str2...' }
# set binary_flag to True, all feature:value will be 1
broadcast_dic = sc.broadcast(all_hashes_seq_dic)
seq_featured_rdd = featured_rdd.map(lambda x: convert2seq(
x, label_idx, data_idx, broadcast_dic.value, binary_flag=binary_flag)
).cache()
# get hash_cnthsh_dict then flatMap and reduce to (feat id, count)
ct_rdd = seq_featured_rdd.flatMap(lambda x: [(i[0], i[1]) for i in x[
data_idx].iteritems()]).reduceByKey(lambda a, b: a + b)
# sorted by feature id as int
feat_sample_count_arr = ct_rdd.sortBy(lambda x: int(x[0])).map(
lambda x: x[1]).collect()
# sort after collect may fail when rdd is huge
#feat_sample_count_arr=[]
#for i in sorted(ct_rdd.collect(), key=lambda t: int(t[0])):
# feat_sample_count_arr.append(i[1])
print "INFO: feat_sample_count_arr len=", len(feat_sample_count_arr)
# save feat_sample_count_arr data ==================================== ============
filter = '{"rid":' + row_id_str + ',"key":"feat_sample_count_arr"}'
upsert_flag = True
jo_insert = {}
jo_insert["rid"] = eval(row_id_str)
jo_insert["key"] = "feat_sample_count_arr"
jo_insert["value"] = feat_sample_count_arr
jstr_insert = json.dumps(jo_insert)
ret = query_mongo.upsert_doc_t(mongo_tuples, filter, jstr_insert,
upsert_flag)
print "INFO: Upsert count for feat_sample_count_arr=", ret
# insert failed, save to local
if ret == 0:
# drop old record in mongo
ret = query_mongo.delete_many(mongo_tuples, None, filter)
if not os.path.exists(local_out_dir):
os.makedirs(local_out_dir)
fsca_hs = os.path.join(local_out_dir, row_id_str,
row_id_str + "_feat_sample_count_arr.pkl")
print "WARNING: save feat_sample_count_arr to local"
ml_util.ml_pickle_save(feat_sample_count_arr, fsca_hs)
# save feature data; TBD. not used. ==================================== ============
#libsvm_rdd=seq_featured_rdd.map(lambda x: convert2libsvm(x,label_idx,data_idx,label_arr))
# put hash to the front of each row, assume hash is after label
libsvm_rdd = seq_featured_rdd.map(
lambda x: x[label_idx + 1] + " " + convert2libsvm(
x, label_idx, data_idx, label_arr))
# debug only
#print "libsvm_rdd="
#for i in libsvm_rdd.collect():
# print i
# get rdd statistics info
stats = featured_rdd.map(lambda p: len(p[metadata_count])).stats()
feat_count_max = stats.max()
feat_count_stdev = stats.stdev()
feat_count_mean = stats.mean()
sample_count = stats.count()
print "INFO: libsvm data: sample count=", sample_count, ",Feat count mean=", feat_count_mean, ",Stdev=", feat_count_stdev
print "INFO: ,max feature count=", feat_count_max
# find sample count
lbl_arr = featured_rdd.map(lambda x: (x[label_idx], 1)).reduceByKey(
add).collect()
print "INFO: Sample count by label=", lbl_arr
# remove duplicated libsvm string; only keep the first duplicated item, assume space following key_idx
if remove_duplicated == "Y":
libsvm_rdd=libsvm_rdd \
.map(lambda x: ( ','.join(x.split(' ')[metadata_count:]), x)) \
.groupByKey().map(lambda x: list(x[1])[0] ) \
.cache()
cnt_list = libsvm_rdd.map(lambda x: (x.split(' ')[1], 1)).reduceByKey(
add).collect()
stats = libsvm_rdd.map(
lambda x: len(x.split(' ')[metadata_count:])).stats()
feat_count_max = stats.max()
feat_count_stdev = stats.stdev()
feat_count_mean = stats.mean()
sample_count = stats.count()
print "INFO: Non-Duplicated libsvm data: sample count=", sample_count, ",Feat count mean=", feat_count_mean, ",Stdev=", feat_count_stdev
print "INFO: ,max feature count=", feat_count_max
print "INFO: Non-Duplicated Label count list=", cnt_list
# clean up libsvm data ==================================== ============
libsvm_data_file = os.path.join(hdfs_folder,
libsvm_alldata_filename) #"libsvm_data"
print "INFO: libsvm_data_file=", libsvm_data_file
try:
#hdfs.ls(save_dir)
#print "find hdfs folder"
hdfs.rmr(libsvm_data_file)
#if num_gram == 1:
# hdfs.rmr(dnn_data_file)
#print "all files removed"
except IOError as e:
print "WARNING: I/O error({0}): {1} at libsvm_data_file clean up".format(
e.errno, e.strerror)
except:
print "WARNING: Unexpected error at libsvm file clean up:", sys.exc_info(
)[0]
#codec = "org.apache.hadoop.io.compress.GzipCodec"
#libsvm_rdd.saveAsTextFile(libsvm_data_file, codec)
libsvm_rdd.saveAsTextFile(libsvm_data_file) # TBD encrypted
feat_count_file = libsvm_data_file + "_feat_count"
print "INFO: feat_count_file=", feat_count_file
try:
hdfs.rmr(feat_count_file)
except IOError as e:
print "WARNING: I/O error({0}): {1} at feat_count clean up".format(
e.errno, e.strerror)
except:
print "WARNING: Unexpected error at libsvm feature count clean up:", sys.exc_info(
)[0]
sc.parallelize([total_feature_numb], 1).saveAsTextFile(feat_count_file)
label_dic = {}
# assign label a number
for idx, label in enumerate(sorted(label_arr)):
if not label in label_dic:
label_dic[
label] = idx #starting from 0, value = idx, e.g., clean:0, dirty:1
# output text for DNN:[meta-data1,meta-data2,..., [feature tokens]] ================= DNN ===========
if num_gram == 1: # special flag to tokenize and keep input orders
print "INFO: processing data for DNN..."
# create token dict
# str_hash_dict: string to hash
# all_hashes_seq_dic: hash to seq id
if token_dict is None or len(token_dict) == 0:
token_dict = {}
str_hash_dict = {v: k for k, v in all_hash_str_dic.iteritems()}
for k, v in str_hash_dict.iteritems():
token_dict[k] = int(all_hashes_seq_dic[str(v)])
#print "token_dict=",len(token_dict),token_dict
dnn_rdd = tmp_rdd \
.map(lambda x: tokenize_by_dict(x, data_idx, token_dict,label_idx, label_dic)) \
.filter(lambda x: len(x) > metadata_count) \
.filter(lambda x: type(x[metadata_count]) is list)
#.cache()
# filter duplication here
#print dnn_rdd.take(3)
dnn_data_file = os.path.join(hdfs_folder,
dnn_alldata_filename) #"dnn_data"
print "INFO: dnn_data_file=", dnn_data_file
try:
hdfs.rmr(dnn_data_file)
except IOError as e:
print "WARNING: I/O error({0}): {1} at dnn_data_file clean up".format(
e.errno, e.strerror)
except:
print "WARNING: Unexpected error at libsvm file clean up:", sys.exc_info(
)[0]
# clean up data
dnn_npy_gz_file = os.path.join(hdfs_folder, row_id_str + "_dnn_")
print "INFO: dnn_npy_gz_file=", dnn_npy_gz_file
try:
hdfs.rmr(dnn_npy_gz_file + "data.npy.gz")
hdfs.rmr(dnn_npy_gz_file + "label.npy.gz")
hdfs.rmr(dnn_npy_gz_file + "info.npy.gz")
except IOError as e:
print "WARNING: I/O error({0}): {1} at dnn_npy clean up".format(
e.errno, e.strerror)
except:
print "WARNING: Unexpected error at dnn_npy file clean up:", sys.exc_info(
)[0]
# save new data
try:
dnn_rdd.saveAsTextFile(dnn_data_file)
except:
print "WARNING: Unexpected error at saving dnn data:", sys.exc_info(
)[0]
# show data statistics
try:
stats = dnn_rdd.map(lambda p: len(p[metadata_count])).stats()
feat_count_max = stats.max()
feat_count_stdev = stats.stdev()
feat_count_mean = stats.mean()
sample_count = stats.count()
print "INFO: DNN data: sample count=", sample_count, ",Feat count mean=", feat_count_mean, ",Stdev=", feat_count_stdev
print "INFO: ,max feature count=", feat_count_max
except:
print "WARNING: Unexpected error at getting stats of dnn_rdd:", sys.exc_info(
)[0]
# clean up pca data in hdfs ============ ========================
pca_files = '*' + libsvm_alldata_filename + "_pca_*"
#print "INFO: pca_files=", pca_files
try:
f_list = hdfs.ls(hdfs_folder)
if len(f_list) > 0:
df_list = fnmatch.filter(f_list, pca_files)
for f in df_list:
print "INFO: rm ", f
hdfs.rmr(f)
except IOError as e:
print "WARNING: I/O error({0}): {1}".format(e.errno, e.strerror)
except:
print "WARNING: Unexpected error at libsvm pca file clean up:", sys.exc_info(
)[0]
# clean up pca data in web local ============ ========================
pca_fname = os.path.join(model_data_folder, row_id_str + '_pca_*.pkl*')
print "INFO: pca_fname=", pca_fname
try:
for fl in glob.glob(pca_fname):
print "INFO: remove ", fl
os.remove(fl)
except OSError, e:
print("Error: %s - %s." % (e.pca_fname, e.strerror))