Merge pull request #165 from sashafrey/master

Improved reporting of errors and exceptions

docs/ref/windows_distribution.txt

@@ -60,6 +60,8 @@ This chapter describes content of BigARTM distribution package for Windows, avai
 
                             * `example11_get_theta_matrix.py <https://raw.githubusercontent.com/bigartm/bigartm/master/src/python/examples/example11_get_theta_matrix.py>`_
 
+                            * `example12_get_topic_model.py <https://raw.githubusercontent.com/bigartm/bigartm/master/src/python/examples/example12_get_topic_model.py>`_
+
                             | Files ``docword.kos.txt`` and ``vocab.kos.txt`` represent
                             | a simple collection of text files in Bag-Of-Words format.
                             | The files are taken from UCI Machine Learning Repository

docs/tutorials/windows_basic.txt

@@ -134,6 +134,8 @@ Folder ``C:\BigARTM\python\examples`` contains several toy examples:
     * `example09_regularizers.py <https://raw.githubusercontent.com/bigartm/bigartm/master/src/python/examples/example09_regularizers.py>`_
     * `example10_multimodal.py <https://raw.githubusercontent.com/bigartm/bigartm/master/src/python/examples/example10_multimodal.py>`_
     * `example11_get_theta_matrix.py <https://raw.githubusercontent.com/bigartm/bigartm/master/src/python/examples/example11_get_theta_matrix.py>`_
+    * `example12_get_topic_model.py <https://raw.githubusercontent.com/bigartm/bigartm/master/src/python/examples/example12_get_topic_model.py>`_
+
 
 All examples does not have any parameters, and you may run them without arguments:
 

src/artm/core/common.h

@@ -116,7 +116,7 @@ inline bool make_rpcz_call(std::function<void()> f, const std::string& log_messa
       ss << "Network comminication timeout";
       if (!log_message.empty()) ss << " in " << log_message;
       LOG(ERROR) << ss.str();
-      if (!no_throw) throw artm::core::NetworkException(ss.str());
+      if (!no_throw) BOOST_THROW_EXCEPTION(artm::core::NetworkException(ss.str()));
       return false;
     }
 
@@ -126,15 +126,15 @@ inline bool make_rpcz_call(std::function<void()> f, const std::string& log_messa
       ss << ", code = " << error.get_application_error_code();
       if (!error.get_error_message().empty()) ss << ", error_message = " << error.get_error_message();
       LOG(ERROR) << ss.str();
-      if (!no_throw) throw artm::core::NetworkException(ss.str());
+      if (!no_throw) BOOST_THROW_EXCEPTION(artm::core::NetworkException(ss.str()));
       return false;
     }
 
     ss << "Network error";
     if (!log_message.empty()) ss << " in " << log_message;
     ss << ", rpcz_error_status = " << error.get_status();
     LOG(ERROR) << ss.str();
-    if (!no_throw) throw artm::core::NetworkException(ss.str());
+    if (!no_throw) BOOST_THROW_EXCEPTION(artm::core::NetworkException(ss.str()));
     return false;
   }
 

src/artm/core/data_loader.cc

@@ -6,6 +6,7 @@
 #include <vector>
 #include <fstream>  // NOLINT
 
+#include "boost/exception/diagnostic_information.hpp"
 #include "boost/lexical_cast.hpp"
 #include "boost/uuid/uuid_io.hpp"
 #include "boost/uuid/uuid_generators.hpp"
@@ -328,17 +329,8 @@ void LocalDataLoader::ThreadFunction() {
       DataLoader::PopulateDataStreams(*batch, pi.get());
       instance()->processor_queue()->push(pi);
     }
-  }
-  catch(boost::thread_interrupted&) {
-    LOG(WARNING) << "thread_interrupted exception in LocalDataLoader::ThreadFunction() function";
-    return;
-  }
-  catch (std::runtime_error& ex) {
-    LOG(ERROR) << ex.what();
-    throw;
   } catch(...) {
-    LOG(FATAL) << "Fatal exception in LocalDataLoader::ThreadFunction() function";
-    throw;
+    LOG(FATAL) << boost::current_exception_diagnostic_information();
   }
 }
 
@@ -386,7 +378,6 @@ void RemoteDataLoader::ThreadFunction() {
       int processor_queue_size = instance()->processor_queue()->size();
       int max_queue_size = config.processor_queue_max_size();
       if (processor_queue_size >= max_queue_size) {
-        // Sleep and check for interrupt.
         boost::this_thread::sleep(boost::posix_time::milliseconds(kIdleLoopFrequency));
         continue;
       }
@@ -444,13 +435,8 @@ void RemoteDataLoader::ThreadFunction() {
       }
     }
   }
-  catch(boost::thread_interrupted&) {
-    LOG(WARNING) << "thread_interrupted exception in RemoteDataLoader::ThreadFunction() function";
-    return;
-  }
-  catch(...) {
-    LOG(FATAL) << "Fatal exception in RemoteDataLoader::ThreadFunction() function";
-    throw;
+  catch (...) {
+    LOG(FATAL) << boost::current_exception_diagnostic_information();
   }
 }
 

src/artm/core/exceptions.h

@@ -19,6 +19,7 @@ try {
 #include <stdexcept>
 #include <string>
 
+#include "boost/exception/diagnostic_information.hpp"
 #include "boost/exception/get_error_info.hpp"
 #include "boost/lexical_cast.hpp"
 #include "boost/throw_exception.hpp"
@@ -97,12 +98,9 @@ catch (const ::artm::core::InternalError& e) {                                 \
 } catch (const rpcz::rpc_error& e) {                                           \
   set_last_error("rpc_error :  " + std::string(e.what()));                     \
   return ARTM_NETWORK_ERROR;                                                   \
-} catch (const std::runtime_error& e) {                                        \
-  set_last_error("InternalError :  " + std::string(e.what()));                 \
-  return ARTM_INTERNAL_ERROR;                                                  \
 } catch (...) {                                                                \
-  LOG(ERROR) << "unknown critical error.";                                     \
-  set_last_error("Unknown critical error. ");                                  \
+  LOG(ERROR) << boost::current_exception_diagnostic_information();             \
+  set_last_error(boost::current_exception_diagnostic_information());           \
   return ARTM_INTERNAL_ERROR;                                                  \
 }
 

src/artm/core/instance.cc

@@ -44,9 +44,9 @@
   }                                                                                           \
   if (!need_hot_reconfigure) {                                                                \
     regularizer.reset(new RegularizerType(regularizer_config));                               \
-    LOG(INFO) << "Regularizer '" + regularizer_name + "' was cold-reconfigured!";             \
+    LOG(INFO) << "Regularizer '" + regularizer_name + "' was cold-reconfigured";              \
   } else {                                                                                    \
-    LOG(INFO) << "Regularizer '" + regularizer_name + "' was hot-reconfigured!";              \
+    LOG(INFO) << "Regularizer '" + regularizer_name + "' was hot-reconfigured";               \
   }                                                                                           \
 }                                                                                             \
 
@@ -157,7 +157,7 @@ void Instance::CreateOrReconfigureRegularizer(const RegularizerConfig& config) {
   if (need_hot_reconfigure) {
     regularizer = schema_.get()->regularizer(regularizer_name);
   } else {
-    LOG(INFO) << "Regularizer '" + regularizer_name + "' will be created!";
+    LOG(INFO) << "Regularizer '" + regularizer_name + "' will be created";
   }
 
   std::string config_blob;  // Used by CREATE_OR_RECONFIGURE_REGULARIZER marco

src/artm/core/merger.cc

@@ -6,6 +6,7 @@
 #include <sstream>
 
 #include "boost/lexical_cast.hpp"
+#include "boost/exception/diagnostic_information.hpp"
 
 #include "glog/logging.h"
 
@@ -190,10 +191,6 @@ void Merger::ThreadFunction() {
         break;
       }
 
-      // Sleep and check for interrupt.
-      // To check for interrupt without sleep,
-      // use boost::this_thread::interruption_point()
-      // which also throws boost::thread_interrupted
       is_idle_ = true;
       boost::this_thread::sleep(boost::posix_time::milliseconds(kIdleLoopFrequency));
       is_idle_ = false;
@@ -264,12 +261,8 @@ void Merger::ThreadFunction() {
       }  // MAIN FOR LOOP
     }
   }
-  catch(boost::thread_interrupted&) {
-    LOG(WARNING) << "thread_interrupted exception in Merger::ThreadFunction() function";
-    return;
-  } catch(...) {
-    LOG(FATAL) << "Fatal exception in Merger::ThreadFunction() function";
-    throw;
+  catch(...) {
+    LOG(FATAL) << boost::current_exception_diagnostic_information();
   }
 }
 
@@ -555,6 +548,10 @@ void Merger::InitializeModel(const InitializeModelArgs& args) {
     BOOST_THROW_EXCEPTION(InvalidOperation(ss.str()));
   }
 
+  LOG(INFO) << "InitializeModel() with "
+            << model.topics_count() << " topics and "
+            << dict->size()  << " tokens";
+
   for (auto iter = dict->begin(); iter != dict->end(); ++iter) {
     ClassId class_id = iter->second.has_class_id() ? iter->second.class_id() : DefaultClass;
     new_ttm->AddToken(Token(class_id, iter->second.key_token()), true);

src/artm/core/processor.cc

@@ -8,6 +8,7 @@
 #include <string>
 #include <vector>
 
+#include "boost/exception/diagnostic_information.hpp"
 #include "boost/lexical_cast.hpp"
 #include "boost/uuid/uuid_generators.hpp"
 #include "boost/uuid/uuid_io.hpp"
@@ -640,10 +641,6 @@ void Processor::ThreadFunction() {
         pop_retries++;
         LOG_IF(INFO, pop_retries == pop_retries_max) << "No data in processing queue, waiting...";
 
-        // Sleep and check for interrupt.
-        // To check for interrupt without sleep,
-        // use boost::this_thread::interruption_point()
-        // which also throws boost::thread_interrupted
         boost::this_thread::sleep(boost::posix_time::milliseconds(kIdleLoopFrequency));
 
         continue;
@@ -652,7 +649,9 @@ void Processor::ThreadFunction() {
       LOG_IF(INFO, pop_retries >= pop_retries_max) << "Processing queue has data, processing started";
       pop_retries = 0;
 
-      CuckooWatch cuckoo("ProcessBatch");  // log time from now to destruction
+
+      // CuckooWatch logs time from now to destruction
+      CuckooWatch cuckoo(std::string("ProcessBatch(") + part->batch().id() + std::string(")"));
       total_processed_batches++;
 
       const Batch& batch = part->batch();
@@ -791,16 +790,8 @@ void Processor::ThreadFunction() {
       });
     }
   }
-  catch (boost::thread_interrupted&) {
-    LOG(WARNING) << "thread_interrupted exception in Processor::ThreadFunction() function";
-    return;
-  }
-  catch (std::runtime_error& ex) {
-    LOG(ERROR) << ex.what();
-    throw;
-  } catch(...) {
-    LOG(FATAL) << "Fatal exception in Processor::ThreadFunction() function";
-    throw;
+  catch (...) {
+    LOG(FATAL) << boost::current_exception_diagnostic_information();
   }
 }
 

src/artm/core/topic_model.cc

@@ -363,6 +363,10 @@ void TopicModel::RetrieveExternalTopicModel(
         topic_model, get_model_args.topic_name_size(), get_model_args.topic_name());
   }
 
+  LOG(INFO) << "RetrieveExternalTopicModel() with "
+            << (use_all_topics ? topic_size() : topics_to_use.size()) << " topics, "
+            << (use_all_tokens ? token_size() : tokens_to_use.size()) << " tokens";
+
   for (int token_index = 0; token_index < token_size(); ++token_index) {
     const Token& current_token = token_collection_.token(token_index);
     if (use_all_tokens ||

src/artm/utility/blas.h

@@ -6,7 +6,10 @@
 #include <assert.h>
 #include <memory>
 #include <vector>
+
+#include "boost/exception/diagnostic_information.hpp"
 #include "boost/utility.hpp"
+
 #include "glog/logging.h"
 
 #include "artm/utility/ice.h"
@@ -31,6 +34,13 @@ typedef void blas_scsr2csc_type(int m, int n, int nnz,
                             const float *csr_val, const int* csr_row_ptr, const int *csr_col_ind,
                                   float *csc_val,       int* csc_row_ind,       int* csc_col_ptr);
 
+#define CATCH_BIG_ALLOCATION(no_rows, no_cols)                                      \
+catch (...) {                                                                       \
+  LOG(ERROR) << "no_rows_ = " << no_rows << ", no_columns_ = " << no_cols << ". "   \
+    << boost::current_exception_diagnostic_information();                           \
+  throw;                                                                            \
+}
+
 namespace artm {
 namespace utility {
 
@@ -65,7 +75,9 @@ class DenseMatrix {
     store_by_rows_(store_by_rows),
     data_(nullptr) {
     if (no_rows > 0 && no_columns > 0) {
-      data_ = new T[no_rows_ * no_columns_];
+      try {
+        data_ = new T[no_rows_ * no_columns_];
+      } CATCH_BIG_ALLOCATION(no_rows_, no_columns_)
     }
   }
 
@@ -74,7 +86,10 @@ class DenseMatrix {
     no_columns_ = src_matrix.no_columns();
     store_by_rows_ = src_matrix.store_by_rows_;
     if (no_columns_ >0 && no_rows_ > 0) {
-      data_ = new T[no_rows_ * no_columns_];
+      try {
+        data_ = new T[no_rows_ * no_columns_];
+      } CATCH_BIG_ALLOCATION(no_rows_, no_columns_)
+
       for (int i = 0; i < no_rows_ * no_columns_; ++i) {
         data_[i] = src_matrix.get_data()[i];
       }
@@ -117,7 +132,10 @@ class DenseMatrix {
       delete[] data_;
     }
     if (no_columns_ >0 && no_rows_ > 0) {
-      data_ = new T[no_rows_ * no_columns_];
+      try {
+        data_ = new  T[no_rows_ * no_columns_];
+      } CATCH_BIG_ALLOCATION(no_rows_, no_columns_);
+
       for (int i = 0; i < no_rows_ * no_columns_; ++i) {
         data_[i] = src_matrix.get_data()[i];
       }
@@ -238,4 +256,6 @@ void AssignDenseMatrixByDivision(const DenseMatrix<T>& first_matrix,
 }  // namespace utility
 }  // namespace artm
 
+#undef CATCH_BIG_ALLOCATION
+
 #endif  // SRC_ARTM_UTILITY_BLAS_H_

src/artm_tests/rpcz_canary_test.cc

@@ -19,7 +19,7 @@
 // cd to /src folder and execute the following:
 // .\protoc.exe --cpp_out=. --rpcz_plugin_out=. .\artm_tests\rpcz_canary_service.proto
 
-const std::string error_message = "Some error had happened!";
+const std::string error_message = "Some error had happened";
 const int error_code = -999;
 
 class SearchServiceImpl : public SearchService {

src/python/examples/example12_get_topic_model.py

@@ -0,0 +1,32 @@
+# This example demonstrates various ways of retrieving Phi matrix from BigARTM
+# -*- coding: utf-8 -*-
+import glob
+
+import artm.messages_pb2
+import artm.library
+
+# Create master component and infer topic model
+batches_disk_path = 'kos'
+with artm.library.MasterComponent(disk_path=batches_disk_path) as master:
+    topic_names = []
+    for topic_index in range(0, 10):
+        topic_names.append("Topic" + str(topic_index))
+    model = master.CreateModel(topic_names=topic_names)
+
+    for iteration in range(0, 2):
+        master.InvokeIteration()
+        master.WaitIdle()  # wait for all batches are processed
+        model.Synchronize()  # synchronize model
+
+    # The following code retrieves one topic at a time.
+    # This avoids retrieving large topic models in a single protobuf message.
+    print "Output p(w|t) values for the first few tokens (alphabetically) in each topic:"
+    for topic_name in topic_names:
+        args = artm.messages_pb2.GetTopicModelArgs()
+        args.model_name = model.name()
+        args.topic_name.append(topic_name)
+        topic_model = master.GetTopicModel(args=args) # represents one column in Phi matrix
+        print topic_model.topic_name[0],
+        for i in range(0, 5):
+            print topic_model.token[i], "%.5f" % topic_model.token_weights[i].value[0],
+        print "..."