def AggregateResults (GoldTEESXML,PredFileList,EpochFocus,AggeragationFolder,Description, ShouldWriteBackAggregation):
import copy as cp ;
print "-"*80
print Description
print GoldTEESXML
print "-"*20
for idx , fileaddr in enumerate (PredFileList):
#print idx+1 , fileaddr
assert "EpochNo"+ str(EpochFocus)+".xml" in fileaddr
assert GE.FILE_CheckFileExists (fileaddr) == True
#print "-"*20
assert GE.FILE_CheckFileExists (GoldTEESXML) == True
if ShouldWriteBackAggregation:
if not GE.OS_IsDirectory (AggeragationFolder):
GE.OS_MakeDir (AggeragationFolder)
if AggeragationFolder[-1]<>"/":
AggeragationFolder+="/"
#Process GOLD
GoldTree = LoadTEESFile_GetTreeRoot (GoldTEESXML, True)
GoldData = Get_Interaction_Results (GoldTree)
GoldKeys = sorted(GoldData.keys())
y_true = np.array ([GoldData[interaction_id] for interaction_id in GoldKeys] , dtype=np.int16)
PredMetric = []
Agg_results = {}
for FileAddress in PredFileList:
print "/".join (FileAddress.split("/")[-4:])
PredTree = LoadTEESFile_GetTreeRoot (FileAddress, True)
PredData = Get_Interaction_Results (PredTree)
assert sorted(PredData.keys ()) == GoldKeys
y_pred = []
for interaction_key in GoldKeys:
y_pred.append (PredData[interaction_key])
#Create Key in Agg_results dict if not exists
if not interaction_key in Agg_results:
Agg_results[interaction_key] = 0 ;
#update Agg_results dict ...
Agg_results[interaction_key]+= PredData[interaction_key]
y_pred = np.array (y_pred , dtype=np.int16)
EvaluationResult = Evaluate (y_true, y_pred)
PredMetric.append (EvaluationResult["positive"]["f1-score"])
print "-"*20 , "\n" , len(PredFileList) , "MEAN : " , f_round (np.mean (PredMetric)) , "STD: " , f_round (np.std (PredMetric)) , "\n"
Agg_Thresholds_Evaluation = []
for threshold_level in range(1, len(PredFileList)+1):
y_pred = [] ;
for interaction_key in GoldKeys:
if Agg_results[interaction_key] >= threshold_level:
y_pred.append (1)
else:
y_pred.append (0)
y_pred = np.array (y_pred , dtype=np.int16)
EvaluationResult = Evaluate (y_true, y_pred)
print '{:2d}'.format(threshold_level), f_round (EvaluationResult["positive"]["recall"]), f_round(EvaluationResult["positive"]["precision"]), f_round(EvaluationResult["positive"]["f1-score"])
Agg_Thresholds_Evaluation.append ( (threshold_level, EvaluationResult["positive"]["recall"], EvaluationResult["positive"]["precision"], EvaluationResult["positive"]["f1-score"] ) )
Best_Threshold = sorted (Agg_Thresholds_Evaluation , key = lambda x: x[3] , reverse=True)[0]
print "\n" , f_round (np.mean (PredMetric)) + "\t" + f_round (np.std (PredMetric)) + "\t" + '{:2d}'.format(Best_Threshold[0]) + "\t" + f_round (Best_Threshold[1]) + "\t" + f_round (Best_Threshold[2]) + "\t"+ f_round (Best_Threshold[3])
FINAL_RES_STRING = f_round (np.mean (PredMetric)) + ";" + f_round (np.std (PredMetric)) + ";" + '{:2d}'.format(Best_Threshold[0]) + ";" + f_round (Best_Threshold[1]) + ";" + f_round (Best_Threshold[2]) + ";"+ f_round (Best_Threshold[3])
if ShouldWriteBackAggregation:
WRITTEN_Agg_Thresholds_Evaluation = []
for threshold_level in range(1, len(PredFileList)+1):
writeback_agg_results = {}
for interaction_key in GoldKeys:
if Agg_results[interaction_key] >= threshold_level:
writeback_agg_results[interaction_key] = 1
else:
writeback_agg_results[interaction_key] = 0
gold_tree = cp.deepcopy (GoldTree)
for interaction_element in gold_tree.findall (".//interaction"):
interaction_id = interaction_element.attrib["id"]
interaction_e1 = interaction_element.attrib["e1"]
interaction_e2 = interaction_element.attrib["e2"]
key = (interaction_id,interaction_e1,interaction_e2)
interaction_type = "Lives_In" if writeback_agg_results[key]==1 else "neg"
interaction_element.set('type',interaction_type)
wb_filename = "Aggr_Epoch_" + str(EpochFocus) + "_Threshold_" + str(threshold_level) + "_" ;
if PredFileList[0].split("/")[-1].startswith ("LinComb"):
wb_filename+= "LinComb"
else:
wb_filename+= "Forward"
wb_filename+= ".xml"
PredictedFileAddress = AggeragationFolder + wb_filename
LOADSAVE_save_xml(gold_tree, PredictedFileAddress)
EvaluationResult = CompareWithGold (GoldData, PredictedFileAddress)
WRITTEN_Agg_Thresholds_Evaluation.append ( (threshold_level, EvaluationResult["positive"]["recall"], EvaluationResult["positive"]["precision"], EvaluationResult["positive"]["f1-score"] ) )
print PredictedFileAddress, " " , '{:2d}'.format(threshold_level), f_round (EvaluationResult["positive"]["recall"]), f_round(EvaluationResult["positive"]["precision"]), f_round(EvaluationResult["positive"]["f1-score"])
assert Agg_Thresholds_Evaluation == WRITTEN_Agg_Thresholds_Evaluation
#import pdb ; pdb.set_trace()
return FINAL_RES_STRING
def ConfigFileVerification (self):
#1: Checking Class types
if not "CLASSES" in self.Configs:
self.PROGRAM_Halt ("Missing CLASSES dictionary in Config file.");
LOCAL_CLASSES = self.Configs["CLASSES"];
if not "Negative" in LOCAL_CLASSES:
self.PROGRAM_Halt ("Negative class should be defined in the config file.");
if not "Positive" in LOCAL_CLASSES:
self.PROGRAM_Halt ("Positive class should be defined in the config file.");
if len(LOCAL_CLASSES) <> 2:
self.PROGRAM_Halt ('There should be only Positive and Negatives classes. For Multiclass use e.g., "Positive": ["class1", "class2"]');
N , P = [] , [] ;
if isinstance (LOCAL_CLASSES["Negative"] , basestring):
N.append (LOCAL_CLASSES["Negative"]);
elif isinstance (LOCAL_CLASSES["Negative"] , list):
N.extend (LOCAL_CLASSES["Negative"]);
if len(N) != len(set(N)):
self.PROGRAM_Halt ("Negative class cannot have duplicate defined types.");
if isinstance (LOCAL_CLASSES["Positive"] , basestring):
P.append (LOCAL_CLASSES["Positive"]);
elif isinstance (LOCAL_CLASSES["Positive"] , list):
P.extend (LOCAL_CLASSES["Positive"]);
if len(P) != len(set(P)):
self.PROGRAM_Halt ("Positive class cannot have duplicate defined types.");
if len (set(N).intersection(set(P))) > 0:
self.PROGRAM_Halt ("Shared defined class type between Positive and Negative class.");
self.__DO_NOT_SET_ME["CLASSES"] = {"Negative": None, "Positive":None} ;
self.__DO_NOT_SET_ME["CLASSES"]["Negative"] = set (i.lower() for i in N); #<<<CRITICAL>>> LOWER CLASS NAMES
self.__DO_NOT_SET_ME["CLASSES"]["Positive"] = set (i.lower() for i in P); #<<<CRITICAL>>> LOWER CLASS NAMES
#What labels should be used when writing back prediction results into xmls:
self.__DO_NOT_SET_ME["WRITEBACK_CLASSES"] = {"Negative": None, "Positive":None} ;
self.__DO_NOT_SET_ME["WRITEBACK_CLASSES"]["Negative"] = set (i for i in N); #<<<CRITICAL>>> LOWER CLASS NAMES
self.__DO_NOT_SET_ME["WRITEBACK_CLASSES"]["Positive"] = set (i for i in P); #<<<CRITICAL>>> LOWER CLASS NAMES
#Renaming relations into other relations ...
#Good for lots of things, for example integrating many non-relevant relations into a single group
if not "RENAME_CLASSES" in self.Configs:
self.__DO_NOT_SET_ME["RENAME_CLASSES"] = {}
else:
if not isinstance (self.Configs["RENAME_CLASSES"], dict):
self.PROGRAM_Halt ("RENAME_CLASSES in the config file should be a dictionary or not given at all.");
temp_rename_class = {}
for key in self.Configs["RENAME_CLASSES"]:
if not isinstance(key, basestring):
self.PROGRAM_Halt ("Problem in RENAME_CLASSES in the config file. Each keys in dictionary should be string. Problematic: " + str(key))
if isinstance (self.Configs["RENAME_CLASSES"][key],basestring):
x = self.Configs["RENAME_CLASSES"][key].lower()
if (not x in self.__DO_NOT_SET_ME["CLASSES"]["Negative"]) and (not x in self.__DO_NOT_SET_ME["CLASSES"]["Positive"]):
self.PROGRAM_Halt ("Problem in RENAME_CLASSES in the config file. Value should belong to defined positive or negative classes. Problematic: " + x);
temp_rename_class[key.lower()] = x
elif isinstance(self.Configs["RENAME_CLASSES"][key],list):
allvalues = []
for x in self.Configs["RENAME_CLASSES"][key]:
x = x.lower()
if (not x in self.__DO_NOT_SET_ME["CLASSES"]["Negative"]) and (not x in self.__DO_NOT_SET_ME["CLASSES"]["Positive"]):
self.PROGRAM_Halt ("Problem in RENAME_CLASSES in the config file. Value should belong to defined positive or negative classes. Problematic: " + x);
allvalues.append(x)
temp_rename_class[key.lower()] = allvalues
else:
self.PROGRAM_Halt ("RENAME_CLASSES in the config file should be a dictionary, each value should be string or list of strings")
self.__DO_NOT_SET_ME["RENAME_CLASSES"] = temp_rename_class
self.lp (["-"*80 , "[WARNING]: RENAMING RELATIONS INTO OTHER RELATIONS:" , str(self.__DO_NOT_SET_ME["RENAME_CLASSES"]) , "-"*80])
#<<<CRITICAL>>>
"""
# - OneHotEncodingForMultiClass: is used for real classification and prediction and creating the softmax columns.
If we have 10 classes in total, and define 3 classes (i.e., "a","b","c") as Negative in the config file,
the other 7 will be regarded as positive, and will have 8 columns in the softmax. Index 0 in the softmax
will be always for negative label(S). This has the artificial name "negative".
1) Function: ProcessRoot in Preprocessing.py file:
----------------------------------------------------------
if pair_type in self.Configs["CLASSES"]["Negative"]:
positive=False;
class_tp=None ;
elif pair_type in self.Configs["CLASSES"]["Positive"]:
positive=True;
class_tp=pair_type ;
else:
self.PROGRAM_Halt ("Unknown class type for interaction:" + pair_type + "\n" + str(pair_attrib));
2) Function: Create_FeatureMatrix in NetworkInputGeneration.py file:
----------------------------------------------------------
HowManyColumnsForOneHotEncoding = len (self.Configs["OneHotEncodingForMultiClass"]);
y = np.zeros ((Total_Example_CNT, HowManyColumnsForOneHotEncoding),dtype=np.int16);
....
if self.Configs["ClassificationType"]== "binary":
y [seen_pair_count] = 1 if (pair["POSITIVE"]==True) else 0 ;
else:
if pair["POSITIVE"]==False:
y[seen_pair_count,0]=1;#index zero is always for negative class(ES)!!!
else:
OneHotIndex = self.Configs["OneHotEncodingForMultiClass"][pair["CLASS_TP"]];
y[seen_pair_count, OneHotIndex] = 1 ;
3) Function __Evaluate in RelationExtractionPipeline.py:
----------------------------------------------------------
USes for evaluation.
#- WRITEBACK_OneHotEncodingForMultiClass:
Like above, but this is used for writing back the prediction results into XML file.
<<<CRITICAL>>>:
ALL negative predictions will be written with the negative class label of index 0 in list of "Negative" in the config file.
So:
"CLASSES" :
{ "Negative" : ["neg" , "d"],
"Positive" : ["a","b",c"]
}
---> ALL NEGATIVE PREDICTIONS WILL GET "neg" class label.
"CLASSES" :
{ "Negative" : ["d" , "neg"],
"Positive" : ["a","b",c"]
}
---> ALL NEGATIVE PREDICTIONS WILL GET "d" class label.
"""
FIRSTneg = N[0]
OneHotEncodingForMultiClass = {u"negative":0};
WRITEBACK_OneHotEncodingForMultiClass = {FIRSTneg:0};
for i, j in enumerate (sorted(P)):
OneHotEncodingForMultiClass[j.lower()] = i+1 ;
WRITEBACK_OneHotEncodingForMultiClass[j] = i+1 ;
self.__DO_NOT_SET_ME["OneHotEncodingForMultiClass"] = OneHotEncodingForMultiClass ;
self.__DO_NOT_SET_ME["WRITEBACK_OneHotEncodingForMultiClass"] = WRITEBACK_OneHotEncodingForMultiClass ;
#Coarse class labels
#Example: Cause_Effect(e1,e2) and Cause_Effect(e2,e1) ==> Cause_Effect
#First get list, then use SET (to avoird duplicates), then list and sorted.
Coarse_Classes = sorted(list(set([class_tp.split("(e1,e2)")[0].split("(e2,e1)")[0] for class_tp in self.__DO_NOT_SET_ME["CLASSES"]["Positive"]])));
OneHotEncodingForCoarseMultiClass = {u"negative":0};
for i, j in enumerate(Coarse_Classes):
OneHotEncodingForCoarseMultiClass[j.lower()] = i+1 ;
self.__DO_NOT_SET_ME["OneHotEncodingForCoarseMultiClass"] = OneHotEncodingForCoarseMultiClass ;
#2: ClassificationType
if not "ClassificationType" in self.Configs:
self.PROGRAM_Halt ("Missing Section: ClassificationType (should be either 'Binary' or 'Multiclass'.");
if not self.Configs["ClassificationType"].lower() in ['binary','multiclass']:
self.PROGRAM_Halt ("Missing Section: ClassificationType (should be either 'Binary' or 'Multiclass'. GIVEN:" + str(self.Configs["ClassificationType"]));
self.__DO_NOT_SET_ME["ClassificationType"] = self.Configs["ClassificationType"].lower(); # <<<CRITICAL>>> LOWER CLASSIFICATION TYPE
#3: ExampleGeneration
if not "ExampleGeneration" in self.Configs:
self.PROGRAM_Halt ("Missing ExampleGeneration dictionary in the config file.");
"""
if not "HaltIfNoSDP" in self.Configs["ExampleGeneration"]:
self.PROGRAM_Halt ("Missing bool HaltIfNoSDP in ExampleGeneration dictionary in the config file.");
if not isinstance (self.Configs["ExampleGeneration"]["HaltIfNoSDP"], bool):
self.PROGRAM_Halt ("HaltIfNoSDP in the ExampleGeneration dictionary in the config file should be either true or false.");
if not "SDP_DIRECTION" in self.Configs["ExampleGeneration"]:
self.PROGRAM_Halt ("Missing string SDP_DIRECTION in ExampleGeneration dictionary in the config file.");
if not isinstance (self.Configs["ExampleGeneration"]["SDP_DIRECTION"], unicode):
self.PROGRAM_Halt ("SDP_DIRECTION in ExampleGeneration dictionary in the config file should be string.");
SDP_DIRECTION = self.Configs["ExampleGeneration"]["SDP_DIRECTION"].lower();
if not SDP_DIRECTION in ["from_e1value_to_e2value" , "from_e2value_to_e1value" , "from_firstoccurring_to_second" , "from_secondoccurring_to_first"]:
self.PROGRAM_Halt ("SDP_DIRECTION in ExampleGeneration dictionary in the config file should be one of 'from_e1value_to_e2value' , 'from_e2value_to_e1value' , 'from_firstoccurring_to_second' , 'from_secondoccurring_to_first'.");
self.__DO_NOT_SET_ME["ExampleGeneration"]["SDP_DIRECTION"] = SDP_DIRECTION ;
if not "Generate_Reversed_SDP_Features" in self.Configs["ExampleGeneration"]:
self.PROGRAM_Halt ("Missing boolean Generate_Reversed_SDP_Features in ExampleGeneration dictionary in the config file.");
if not isinstance (self.Configs["ExampleGeneration"]["Generate_Reversed_SDP_Features"], bool):
self.PROGRAM_Halt ("Generate_Reversed_SDP_Features in ExampleGeneration dictionary in the config file should be bool.");
if not "SDP_MAXLEN_BECAREFUL" in self.Configs["ExampleGeneration"]:
self.__DO_NOT_SET_ME["ExampleGeneration"]["SDP_MAXLEN_BECAREFUL"] = None ;
self.lp ("[INFO]: SDP_MAXLEN_BECAREFUL is ignored. SDP will have length as max of bags.")
else:
if (not isinstance(self.Configs["ExampleGeneration"]["SDP_MAXLEN_BECAREFUL"], int)) or (self.Configs["ExampleGeneration"]["SDP_MAXLEN_BECAREFUL"] <= 1):
self.PROGRAM_Halt ("SDP_MAXLEN_BECAREFUL in ExampleGeneration dictionary in the config file should be int and > 1.");
if not "Directional_Dependency_Types" in self.Configs["ExampleGeneration"]:
self.PROGRAM_Halt ("Missing boolean Directional_Dependency_Types in ExampleGeneration dictionary in the config file.");
else:
if (not isinstance(self.Configs["ExampleGeneration"]["Directional_Dependency_Types"], bool)):
self.PROGRAM_Halt ("Directional_Dependency_Types in ExampleGeneration dictionary in the config file should be either true or false.");
if not "Use_General_prep_prepc_conj_DT" in self.Configs["ExampleGeneration"]:
self.PROGRAM_Halt ("Missing boolean Use_General_prep_prepc_conj_DT in ExampleGeneration dictionary in the config file.");
else:
if (not isinstance(self.Configs["ExampleGeneration"]["Use_General_prep_prepc_conj_DT"], bool)):
self.PROGRAM_Halt ("Use_General_prep_prepc_conj_DT in ExampleGeneration dictionary in the config file should be either true or false.");
"""
#ActionON_CrossSentenceExamples: (1)Halt (2)Discard
if not "ActionON_CrossSentenceExamples" in self.Configs["ExampleGeneration"]:
self.PROGRAM_Halt ("Missing string ActionON_CrossSentenceExamples in ExampleGeneration dictionary in the config file.");
else:
if (not isinstance(self.Configs["ExampleGeneration"]["ActionON_CrossSentenceExamples"], unicode)):
self.PROGRAM_Halt ("ActionON_CrossSentenceExamples in ExampleGeneration dictionary in the config file should be STRING and either 'Halt' or 'Discard'.");
self.Configs["ExampleGeneration"]["ActionON_CrossSentenceExamples"] = self.Configs["ExampleGeneration"]["ActionON_CrossSentenceExamples"].upper()
if not (self.Configs["ExampleGeneration"]["ActionON_CrossSentenceExamples"] in ["HALT" , "DISCARD"]):
self.PROGRAM_Halt ("ActionON_CrossSentenceExamples in ExampleGeneration dictionary in the config file should be either 'Halt' or 'Discard'.");
if self.Configs["ExampleGeneration"]["ActionON_CrossSentenceExamples"] == "DISCARD":
self.lp (["*"*40 , "*"*40 ,"*"*40 , "[WARNING]: DISCARDING ALL CROSS-SENTECE RELATIONS IF THERE IS ANY !!!" , "*" *40, "*"*40 ,"*"*40]) ;
#ActionON_MissingRelations: (1)Halt (2)GenerateAsNegatives
if not "ActionON_MissingRelations" in self.Configs["ExampleGeneration"]:
self.PROGRAM_Halt ("Missing string ActionON_MissingRelations in ExampleGeneration dictionary in the config file.");
else:
if (not isinstance(self.Configs["ExampleGeneration"]["ActionON_MissingRelations"], unicode)):
self.PROGRAM_Halt ("ActionON_MissingRelations in ExampleGeneration dictionary in the config file should be STRING and either 'Halt' or 'GenerateAsNegatives'.");
self.Configs["ExampleGeneration"]["ActionON_MissingRelations"] = self.Configs["ExampleGeneration"]["ActionON_MissingRelations"].upper()
if not (self.Configs["ExampleGeneration"]["ActionON_MissingRelations"] in ["HALT" , "GENERATEASNEGATIVES"]):
self.PROGRAM_Halt ("ActionON_MissingRelations in ExampleGeneration dictionary in the config file should be either 'Halt' or 'GenerateAsNegatives'.");
if self.Configs["ExampleGeneration"]["ActionON_MissingRelations"] == "GENERATEASNEGATIVES":
self.lp (["*"*40,"*"*40,"*"*40, "[WARNING]: Missing Relations will be generated as Negatives !!!" ,"*"*40 ,"*"*40,"*" *40]);
#ActionON_DuplicateRelations
if not "ActionON_DuplicateRelations" in self.Configs["ExampleGeneration"]:
self.PROGRAM_Halt ("Missing string ActionON_DuplicateRelations in ExampleGeneration dictionary in the config file.");
else:
if (not isinstance(self.Configs["ExampleGeneration"]["ActionON_DuplicateRelations"], unicode)):
self.PROGRAM_Halt ("ActionON_DuplicateRelations in ExampleGeneration dictionary in the config file should be string and either 'Halt', 'Ignore', or 'Discard'.");
self.Configs["ExampleGeneration"]["ActionON_DuplicateRelations"] = self.Configs["ExampleGeneration"]["ActionON_DuplicateRelations"].upper()
if not (self.Configs["ExampleGeneration"]["ActionON_DuplicateRelations"] in ["HALT" , "IGNORE" , "DISCARD"]):
self.PROGRAM_Halt ("ActionON_MissingRelations in ExampleGeneration dictionary in the config file should be either 'Halt' or 'Ignore', or 'Discard'.");
if self.Configs["ExampleGeneration"]["ActionON_MissingRelations"] == "DISCARD":
self.lp (["*"*40,"*"*40,"*"*40, "[WARNING]: DISCARDING ANY DUPLICATE RELATIONS !!!" ,"*"*40 ,"*"*40,"*" *40]);
#4: ValidEnityTypesForRelations
if not "ValidEnityTypesForRelations" in self.Configs:
self.PROGRAM_Halt ("Missing ValidEnityTypesForRelations dictionary in the config file.");
self.__DO_NOT_SET_ME["ValidEnityTypesForRelations"] = set ([i.lower() for i in self.Configs["ValidEnityTypesForRelations"]]); #<<<CRITICAL>>> LOWER
OneHotEncodingForValidEnityTypesForRelations = {} ;
for i , e_tp in enumerate(self.Configs["ValidEnityTypesForRelations"]):
OneHotEncodingForValidEnityTypesForRelations[e_tp.lower()] = i ;
self.__DO_NOT_SET_ME["OneHotEncodingForValidEnityTypesForRelations"] = OneHotEncodingForValidEnityTypesForRelations ;
#5: InteractionElementName
if not "InteractionElementName" in self.Configs:
self.PROGRAM_Halt ("Missing InteractionElementName in the config file.");
if self.Configs["InteractionElementName"] == None:
self.PROGRAM_Halt ("InteractionElementName in the config file should be either interaction or pair.");
if not self.Configs["InteractionElementName"].lower() in ['interaction','pair']:
self.PROGRAM_Halt ("InteractionElementName in the config file should be either 'interaction' or 'pair'.");
self.__DO_NOT_SET_ME["InteractionElementName"] = self.Configs["InteractionElementName"].lower() ; #<<<CRITICAL>>> LOWER
#6: InteractionElementClassAttributeName : in which attribute, class for interaction is given
if not "InteractionElementClassAttributeName" in self.Configs:
self.PROGRAM_Halt ("Missing InteractionElementClassAttributeName in the config file.");
if self.Configs["InteractionElementClassAttributeName"] == None:
self.PROGRAM_Halt ("InteractionElementClassAttributeName in the config file should be either interaction or pair.");
if not self.Configs["InteractionElementClassAttributeName"].lower() in ['interaction','type']:
self.PROGRAM_Halt ("InteractionElementClassAttributeName in the config file should be either 'interaction' or 'type'.");
self.__DO_NOT_SET_ME["InteractionElementClassAttributeName"] = self.Configs["InteractionElementClassAttributeName"].lower() ;
#7: W2V
if not "W2V_Model" in self.Configs:
self.PROGRAM_Halt ("Missing W2V_Model dictionary in the config file.");
if not "Model_Address" in self.Configs["W2V_Model"]:
self.PROGRAM_Halt ("Missing Model_Address in W2V_Model dictionary in config file.");
if not "MaxWordsInMemory" in self.Configs["W2V_Model"]:
self.PROGRAM_Halt ("Missing MaxWordsInMemory in W2V_Model dictionary in config file.");
if not GF.FILE_CheckFileExists(self.Configs["W2V_Model"]["Model_Address"]):
self.PROGRAM_Halt ("File address for W2V_Model is not valid: file not exists.");
#8: Replace W2V vector of mentions with a better general W2V vector if mention not found in W2V model ...
if "ReplaceVectorForEntityTypeIfTokenNotFound" in self.Configs["ExampleGeneration"]:
VectorReplacementDict = {};
for Entity_Type , ReplacementVector in self.Configs["ExampleGeneration"]["ReplaceVectorForEntityTypeIfTokenNotFound"]:
VectorReplacementDict[Entity_Type.lower()] = ReplacementVector ;
if len (VectorReplacementDict) > 0:
self.lp (["REPLACING VECTORS FOR SOME ENTITIY TYPES IF NOT FOUND IN W2V Model" , str(VectorReplacementDict)]);
self.__DO_NOT_SET_ME["ExampleGeneration"]["ReplaceVectorForEntityTypeIfTokenNotFound_Dict"]= VectorReplacementDict ;
else:
self.__DO_NOT_SET_ME["ExampleGeneration"]["ReplaceVectorForEntityTypeIfTokenNotFound_Dict"]= {};
#9: Check if ValidInteractingPairEntityTypes is given ...
if not "ValidInteractingPairEntityTypes" in self.Configs:
self.PROGRAM_Halt ("Missing ValidInteractingPairEntityTypes list in the config file.");
if self.Configs["ValidInteractingPairEntityTypes"] == None:
self.PROGRAM_Halt ("ValidInteractingPairEntityTypes list is null in the config file.");
if len(self.Configs["ValidInteractingPairEntityTypes"]) < 1:
self.PROGRAM_Halt ("ValidInteractingPairEntityTypes in the config file should be list with at least one pair.");
L = self.Configs["ValidInteractingPairEntityTypes"] ;
ListOf_ValidInteractingPairEntityTypes = [] ;
for EntityTypePair in L:
if (not EntityTypePair[0].lower() in self.Configs["ValidEnityTypesForRelations"]) or \
(not EntityTypePair[1].lower() in self.Configs["ValidEnityTypesForRelations"]) :
self.PROGRAM_Halt ("Invalid entity pairs for section ValidInteractingPairEntityTypes:" + str(EntityTypePair));
ListOf_ValidInteractingPairEntityTypes.append ([EntityTypePair[0].lower(),EntityTypePair[1].lower()]);
#Important comment out:
# now we do not put both (e1tp,e2tp) and (e2tp,e1tp) here ...
# only preserve the original, so that we can create negatives according to this order for now ....
#ListOf_ValidInteractingPairEntityTypes.append ([EntityTypePair[1].lower(),EntityTypePair[0].lower()]);
self.__DO_NOT_SET_ME["ListOf_ValidInteractingPairEntityTypes"] = ListOf_ValidInteractingPairEntityTypes ;
#10: Can interact with itself: for example self-interacting proteins or drug ...
if (not "SelfInteractingEntities" in self.Configs):
self.PROGRAM_Halt ("Missing SelfInteractingEntities list in the config file. Set it to 'null' if you need nothing.");
if self.Configs["SelfInteractingEntities"] == None:
self.__DO_NOT_SET_ME["SelfInteractingEntities"] = [] ;
else:
L = [i.lower() for i in self.Configs["SelfInteractingEntities"]];
for i in L:
if not i in self.__DO_NOT_SET_ME["ValidEnityTypesForRelations"]:
self.PROGRAM_Halt ("A self-interacting entity type is defined in section SelfInteractingEntities, but not in ValidEnityTypesForRelations in Config file!");
if not [i,i] in self.__DO_NOT_SET_ME["ListOf_ValidInteractingPairEntityTypes"]:
self.PROGRAM_Halt ("A self-interacting entity type is defined in section SelfInteractingEntities, but no relation in ValidInteractingPairEntityTypes in Config file!");
self.__DO_NOT_SET_ME["SelfInteractingEntities"] = L ;
#11: RemoveSentenceIfNoParseExists
if (not "RemoveSentenceIfNoParseExists" in self.Configs):
self.PROGRAM_Halt ("Missing boolean RemoveSentenceIfNoParseExists in the config file.");
if (not isinstance (self.Configs["RemoveSentenceIfNoParseExists"] , bool)):
self.PROGRAM_Halt ("RemoveSentenceIfNoParseExists in the config file should be either true or false.");
#12: Max Sentence Length
if (not "MAX_SENTENCE_LENGTH" in self.Configs["ExampleGeneration"]):
self.PROGRAM_Halt ("MAX_SENTENCE_LENGTH in ExampleGeneration should be either null or an integer bigger than zero in the config file.");
if self.Configs["ExampleGeneration"]["MAX_SENTENCE_LENGTH"] == None:
self.Configs["ExampleGeneration"]["MAX_SENTENCE_LENGTH"] = -1 ;
else:
try:
if isinstance (self.Configs["ExampleGeneration"]["MAX_SENTENCE_LENGTH"] , bool): raise Exception ("") ;
if not isinstance (self.Configs["ExampleGeneration"]["MAX_SENTENCE_LENGTH"] , int): raise Exception ("") ;
L = int (self.Configs["ExampleGeneration"]["MAX_SENTENCE_LENGTH"]) ;
if not (L>0): raise Exception ("") ;
except:
self.PROGRAM_Halt ("MAX_SENTENCE_LENGTH in ExampleGeneration should be either null or an integer bigger than zero in the config file.");
#13: Runtime Parameters:
if (not "ExecutionParameters" in self.Configs):
self.PROGRAM_Halt ("Missing dictionary ExecutionParameters in the config file.");
if (not "DoNotAskAnyQuestions" in self.Configs["ExecutionParameters"]):
self.PROGRAM_Halt ("Missing boolean DoNotAskAnyQuestions in ExecutionParameters dictionary in the config file.");
if not isinstance (self.Configs["ExecutionParameters"]["DoNotAskAnyQuestions"], bool):
self.PROGRAM_Halt ("DoNotAskAnyQuestions in ExecutionParameters dictionary in the config file should be true/false.");
#14: Evaluation Parameters
if (not "EvaluationParameters" in self.Configs):
self.PROGRAM_Halt ("Missing dictionary EvaluationParameters in the config file.");
if (not "ExcludeClassLabelsList" in self.Configs["EvaluationParameters"]):
self.PROGRAM_Halt ("Missing list ExcludeClassLabelsList in EvaluationParameters dictionary in the config file.");
if not isinstance (self.Configs["EvaluationParameters"]["ExcludeClassLabelsList"], list):
self.PROGRAM_Halt ("ExcludeClassLabelsList in EvaluationParameters dictionary in the config file should be a list. Put [] if you don't want to exclude anything.");
self.Configs["EvaluationParameters"]["ExcludeClassLabelsList"] = [i.lower() for i in self.Configs["EvaluationParameters"]["ExcludeClassLabelsList"]];
if len (self.Configs["EvaluationParameters"]["ExcludeClassLabelsList"]) > 0:
ALL_CLASSES = self.Configs["CLASSES"]["Positive"] | self.Configs["CLASSES"]["Negative"] ;
for i in self.Configs["EvaluationParameters"]["ExcludeClassLabelsList"]:
if not i in ALL_CLASSES:
self.PROGRAM_Halt (i + " class label is defined in [EvaluationParameters][ExcludeClassLabelsList] in the config file but not defined as a valid class in the CLASSES dictionary.");
if (not "DecimalPoints" in self.Configs["EvaluationParameters"]):
self.PROGRAM_Halt ("Missing integer DecimalPoints in EvaluationParameters dictionary in the config file.");
if not isinstance (self.Configs["EvaluationParameters"]["DecimalPoints"], int):
self.PROGRAM_Halt ("DecimalPoints in EvaluationParameters dictionary in the config file should be a positive integer. Put 2 for default.");
if self.Configs["EvaluationParameters"]["DecimalPoints"] <=0 :
self.PROGRAM_Halt ("DecimalPoints in EvaluationParameters dictionary in the config file should be a positive integer. Put 2 for default.");
