i'm creating an application on C# that generates a hdf5 document with one row of the compound data being a timestamp (using unix timestamp), and i wish i could load this hdf5 in pandas and it load the timestamp column as a datetime64. I've tried to create the file from pandas and i saw it creates the dataset with a lot of attributes that i assume pandas read and do some logic to convert it while reading the file (the datetime inside the file created from pandas is a timestamp too), i can't figure it out... Here is my dataset creation code and some screenshots of what i had investigated public void CriaHDF5Customizado(PackingConfigFile pf) { // PopulaOPC(pf); H5FileId arquivoHDF5 = H5F.create("C:/Users/arthuro/Desktop/timestampteste.h5", H5F.CreateMode.ACC_TRUNC); H5GroupId datasetGroup = H5G.create(arquivoHDF5, "Datasets"); H5GroupId infosGroup = H5G.create(arquivoHDF5, "Informations"); H5G.close(infosGroup); opcSt opHelper = new opcSt(); opHelper.dt = (Int64)DateTime.Parse("1996-11-9").Subtract(new DateTime(1970, 1, 1,12,00,00, DateTimeKind.Utc)).TotalSeconds; opHelper.qlt = (Int64)DateTime.Now.Subtract(new DateTime(1970, 1, 1, 0, 0, 0, DateTimeKind.Utc)).TotalSeconds; opHelper.vl = 123456.7f; int structsize = Marshal.SizeOf(opHelper); H5DataTypeId myOPCType = H5T.create(H5T.CreateClass.COMPOUND, structsize); //H5T.insert(myOPCType, "TimeStamp", 0, new H5DataTypeId(H5T.H5Type.NATIVE_LONG)); H5T.insert(myOPCType, "Quality", 8, new H5DataTypeId(H5T.H5Type.NATIVE_LONG)); H5T.insert(myOPCType, "Value", 16, new H5DataTypeId(H5T.H5Type.NATIVE_FLOAT)); long[] ds = new long[1]; ds[0] = 1; H5DataTypeId dtDatetime = H5T.create_array(new H5DataTypeId(H5T.H5Type.NATIVE_LLONG), ds); H5T.insert(myOPCType, "TimeStamp", 0, dtDatetime); foreach (BasicVariable bv in pf.basicVariableList.bvList) { bv.PopulateOPCUA(); var bvnow = bv; var bvNro = pf.basicVariableList.bvList.IndexOf(bv); long[] dims = new long[1]; dims[0] = bv.bvData.Count; H5DataSpaceId myDataSpace = H5S.create_simple(1, dims); H5DataSetId bvDset = H5D.create(datasetGroup, bv.bvTag, myOPCType, myDataSpace); var arrayaux = new List<opcSt>(); foreach (OPC_UA opc in bv.bvData) { var aux = new opcSt(opc.timeStamp, opc.quality, (float)opc.data); arrayaux.Add(aux); } H5D.write(bvDset, myOPCType, new H5Array<opcSt>(arrayaux.ToArray())); string[] stringteste = { "datetime64" }; H5DataTypeId attrDt = H5T.copy(H5T.H5Type.C_S1); H5T.setSize(attrDt, stringteste[0].Length); var longsz = new long[] { 1 }; var enc = new System.Text.ASCIIEncoding(); var array1 = enc.GetBytes(stringteste[0]); var charArray = new byte[stringteste[0].Length + 1]; array1.CopyTo(charArray, 0); charArray[stringteste[0].Length] = 0; H5DataSpaceId atribDS = H5S.create_simple(1, longsz); H5AttributeId Attribs = H5A.create(bvDset, "TimeStamp_dtype", attrDt, atribDS); H5A.write(Attribs, attrDt, new H5Array<byte>(charArray)); stringteste[0] = "(lp0L0La."; H5T.setSize(attrDt, stringteste[0].Length); longsz = new long[] { 1 }; enc = new System.Text.ASCIIEncoding(); array1 = enc.GetBytes(stringteste[0]); charArray = new byte[stringteste[0].Length + 1]; array1.CopyTo(charArray, 0); charArray[stringteste[0].Length] = 0; atribDS = H5S.create_simple(1, longsz); H5AttributeId Attribs2 = H5A.create(bvDset, "TimeStamp_kind", attrDt, atribDS); H5A.write(Attribs2, attrDt, new H5Array<byte>(charArray)); stringteste[0] = "N."; H5T.setSize(attrDt, stringteste[0].Length); longsz = new long[] { 1 }; enc = new System.Text.ASCIIEncoding(); array1 = enc.GetBytes(stringteste[0]); charArray = new byte[stringteste[0].Length + 1]; array1.CopyTo(charArray, 0); charArray[stringteste[0].Length] = 0; atribDS = H5S.create_simple(1, longsz); H5AttributeId Attribs3 = H5A.create(bvDset, "TimeStamp_meta", attrDt, atribDS); H5A.write(Attribs3, attrDt, new H5Array<byte>(charArray)); H5S.close(atribDS); H5A.close(Attribs); H5A.close(Attribs2); H5A.close(Attribs3); H5D.close(bvDset); H5S.close(myDataSpace); bv.bvData = new List<OPC_UA>(); } H5G.close(datasetGroup); H5F.close(arquivoHDF5); } Pandas reading (timestamp and datetime): being 1- The value read inside jupyter notebook (processed by pandas) 2- The real value inside the HDF5 File https://i.stack.imgur.com/ZZJ3A.png <https://github.com/pandas-dev/pandas/issues/url> Ahd these are the attributes of the table created by pandas (that i've tried to recreate in my code) https://i.stack.imgur.com/U1Zom.png <https://github.com/pandas-dev/pandas/issues/url>