!3572 [MD] fix save pydoc and log

Merge pull request !3572 from liyong126/r0.6_fix_save_pydoc_log

mindspore/ccsrc/minddata/dataset/api/de_pipeline.cc

@@ -410,6 +410,7 @@ Status DEPipeline::SaveDataset(const std::vector<std::string> &file_names, const
       std::vector<std::string> index_fields;
       s = FetchMetaFromTensorRow(column_name_id_map, row, &mr_json, &index_fields);
       RETURN_IF_NOT_OK(s);
+      MS_LOG(DEBUG) << "Schema of saved mindrecord: " << mr_json.dump();
       if (mindrecord::SUCCESS !=
           mindrecord::ShardHeader::initialize(&mr_header, mr_json, index_fields, blob_fields, mr_schema_id)) {
         RETURN_STATUS_UNEXPECTED("Error: failed to initialize ShardHeader.");
@@ -569,6 +570,7 @@ Status DEPipeline::FetchMetaFromTensorRow(const std::unordered_map<std::string,
   if (column_name_id_map.empty()) {
     RETURN_STATUS_UNEXPECTED("Error: column not found.");
   }
+  json dataset_schema;
   for (auto &col : column_name_id_map) {
     auto idx = col.second;
     auto column_name = col.first;
@@ -580,6 +582,7 @@ Status DEPipeline::FetchMetaFromTensorRow(const std::unordered_map<std::string,
     auto shapes = column_shape.AsVector();
     std::vector<int> mr_shape(shapes.begin(), shapes.end());
     std::string el = column_type.ToString();
+    dataset_schema[column_name] = el;
     if (mindrecord::kTypesMap.find(el) == mindrecord::kTypesMap.end()) {
       std::string err_msg("Error: can not support data type: " + el);
       RETURN_STATUS_UNEXPECTED(err_msg);
@@ -605,6 +608,7 @@ Status DEPipeline::FetchMetaFromTensorRow(const std::unordered_map<std::string,
     if (mr_type == "bytes" || !mr_shape.empty()) continue;
     index_fields->emplace_back(column_name);  // candidate of index fields
   }
+  MS_LOG(DEBUG) << "Schema of dataset: " << dataset_schema.dump();
   return Status::OK();
 }
 Status DEPipeline::BuildMindrecordSamplerChain(const py::handle &handle,

mindspore/ccsrc/minddata/mindrecord/io/shard_writer.cc

@@ -83,7 +83,7 @@ MSRStatus ShardWriter::OpenDataFiles(bool append) {
       // if not append and mindrecord file exist, return FAILED
       fs->open(common::SafeCStr(file), std::ios::in | std::ios::binary);
       if (fs->good()) {
-        MS_LOG(ERROR) << "MindRecord file already existed.";
+        MS_LOG(ERROR) << "MindRecord file already existed, please delete file: " << common::SafeCStr(file);
         fs->close();
         return FAILED;
       }

mindspore/dataset/engine/datasets.py

@@ -1041,12 +1041,61 @@ class Dataset:
         """
         Save the dynamic data processed by dataset pipeline as common dataset format, support: mindrecord.
 
+        Implicit type casting exists when saving data as mindrecord. Table below shows how to do type casting.
+
+        .. list-table:: Implicit Type Casting of Saving as mindrecord
+           :widths: 25 25 50
+           :header-rows: 1
+
+           * - type in 'dataset'
+             - type in 'mindrecord'
+             - detail
+           * - DE_BOOL
+             - None
+             - Not support
+           * - DE_INT8
+             - int32
+             -
+           * - DE_UINT8
+             - bytes(1D uint8)
+             - Drop dimension
+           * - DE_INT16
+             - int32
+             -
+           * - DE_UINT16
+             - int32
+             -
+           * - DE_INT32
+             - int32
+             -
+           * - DE_UINT32
+             - int64
+             -
+           * - DE_INT64
+             - int64
+             -
+           * - DE_UINT64
+             - None
+             - Not support
+           * - DE_FLOAT16
+             - Not support
+             -
+           * - DE_FLOAT32
+             - float32
+             -
+           * - DE_FLOAT64
+             - float64
+             -
+           * - DE_STRING
+             - string
+             - Not support multi-dimensional DE_STRING
+
         Note:
             1. To save the samples in order, should set dataset's shuffle false and num_files 1.
             2. Before call the function, do not use batch, repeat operator or data augmentation operators
                with random attribute in map operator.
-            3. Mindreocrd do not support np.uint64, multi-dimensional np.uint8(drop dimension) and
-               multi-dimensional string.
+            3. Mindrecord does not support DE_UINT64, multi-dimensional DE_UINT8(drop dimension) and
+               multi-dimensional DE_STRING.
 
         Args:
             file_name (str): Path to dataset file.