History#

current - 2022-03-10 - 0.00Mb#

  • issue377: Implements TreeEnsemble* for opsetml==3 (2022-03-10)

  • issue376: Avoids one circular import. (2022-03-07)

  • issue375: Adds code to turn onnx example into python unit test (2022-03-05)

  • issue374: Implements onnx backend with python runtime (2022-03-05)

  • issue372: Improves importing time (2022-03-05)

  • issue373: Adds support for Expand in python runtime (2022-03-04)

  • issue371: Support for ONNX functions (2022-03-04)

  • issue370: Refactors numpy API to use Xop API (2022-03-03)

  • issue369: Supports recursive display in onnx_simple_text_plot (2022-02-28)

  • issue368: Updates requirements, skl2onnx>=1.11 (2022-02-28)

  • issue367: Refactors results name in Xop API (2022-02-27)

  • issue366: Adds python runtime for CategoryMapper (2022-02-24)

  • issue365: Adds command line benchmark_doc (2022-02-24)

  • issue364: Runs onnx backend test with python runtime (2022-02-23)

  • issue363: Refactoring, moving files testing.experimental_c (2022-02-23)

  • issue362: Adds command line plot_onnx (2022-02-23)

  • issue361: Introduces __max_supported_opset__ and refactors the library (2022-02-23)

  • issue360: Xop API, adds class OnnxSubOnnx to insert ONNX graph (2022-02-22)

  • issue359: Supports domains in Xop API (2022-02-21)

  • issue358: Extends supported operator by OnnxShapeInference (2022-02-21)

  • issue357: Modifies OnnxShapeInference to deal with untyped outputs (2022-02-19)

  • issue356: Supports multiple affectations (xop) (2022-02-18)

  • issue355: Fixes for onnx==1.11 (2022-02-18)

  • issue353: Experimentations with a new API to create ONNX graphs (2022-02-18)

  • issue352: Supports for shape inference on unary operators (2022-02-14)

0.8.1697 - 2022-02-11 - 1.97Mb#

  • issue351: Adds name in ShapeResult, fixes zoo links (2022-02-11)

  • issue350: First version of runtime OnnxShapeInference (2022-02-09)

  • issue348: Moves OnnxMicroRuntime to onnxrt (2022-02-05)

  • issue346: Adds runtime for operator CastLike (2022-02-05)

  • issue347: numpy API for onnx: wrapped function can call other wrapped functions (2022-02-04)

  • issue345: Improves command line to measure latency for a model (2022-02-03)

  • issue344: Adds a method to_onnx to easily retrieve the onnx graph from numpy onnx function (2022-02-03)

  • issue343: Shows links in onnx_simple_text_plot (2022-02-03)

  • issue342: Displays small arrays in onnx_simple_text_plot (2022-01-22)

0.8.1674 - 2021-12-30 - 1.94Mb#

  • issue340: Implements tokenizer following scikit-learn’s API using onnxruntime-extensions (2021-12-29)

  • issue339: op_label_encoder support for keys_strings & values_floats (2) (replaces #335) (2021-12-29)

  • issue338: Updated to support key_strings and values_floats combo (2021-12-29)

  • issue335: op_label_encoder support for keys_strings & values_floats (2021-12-29)

  • issue322: Add tokenizers with onnxruntime-extensions (2021-12-29)

  • issue337: Supports operator Scan when exporting an onnx graph to onnx code (2021-12-21)

  • issue336: Enables GPU with OnnxInference and onnxruntime (2021-12-21)

0.7.1672 - 2021-12-19 - 1.95Mb#

  • issue334: update history (2021-12-19)

  • issue333: Adds command line latency to measure the latency of a runtime (2021-12-18)

  • issue332: Improves dot rendering, fixes disconnected subgraphs (2021-12-18)

  • issue331: Removes measure_time (2021-12-15)

  • issue330: Reduces verbosity when onnxruntime is used as a runtime for OnnxInference (2021-12-14)

  • issue329: Fixes type issue in shape inference for operator If (2021-12-14)

  • issue328: Extends command line onnx_stats (2021-12-14)

  • issue327: Adds runtime for operator LeakyRelu (2021-12-13)

  • issue326: Better error messages when name is shared with results and node name in onnx_simple_text_plot (2021-12-10)

0.7.1649 - 2021-12-09 - 1.95Mb#

  • issue325: Implements a simple text display for ONNX graph (2021-12-08)

  • issue324: Adds runtime for gradient operators YieldOp, BroadcastGradientArgs (2021-11-30)

  • issue323: Implements if with numpy API (2021-11-26)

  • issue320: Fix exporter to tf2onnx (2021-11-13)

  • issue319: Supports operator SequenceAt in OnnxInference (2021-11-09)

  • issue318: Disable onnxruntime optimisation on one particular graph (2021-11-04)

  • issue317: plot_onnx fails when node names contains ‘.’ (2021-10-28)

  • issue316: failed to use RandomForestRegressor ort in android studio (2021-10-28)

0.7.1626 - 2021-10-21 - 1.93Mb#

  • issue315: Fixes import issue for python 3.6 (2021-10-21)

0.7.1625 - 2021-10-12 - 15.57Mb#

  • issue314: Builds mlprodict for python 3.6 on linux (2021-10-11)

  • issue313: Fix a bug related to shapes when exporting a model to tf2onnx (2021-10-10)

  • issue312: Add more tests for einsum decomposition (2021-10-08)

0.7.1624 - 2021-10-02 - 15.19Mb#

  • issue311: Support opset 15 (onnx>=1.10) (2021-10-02)

  • issue310: Raise an exception when inplace and intermediate are True (OnnxInference.run) (2021-09-23)

0.7.1602 - 2021-09-21 - 22.30Mb#

  • issue309: Adds function insert_results_into_onnx to insert results into a graph to debug (2021-09-21)

  • issue308: Adds function to rename all results in ONNX graphs (2021-09-13)

  • issue307: Adds runtime for operator SequenceConstruct (2021-09-13)

  • issue305: Add option to split lightgbm converter into multipule TreeEnsemble (2021-09-10)

  • issue304: Add tree text visualization for TreeEnsemble (2021-09-01)

  • issue303: Implements a estimator speeding up the inference using ONNX (2021-08-31)

  • issue302: Removes unused nodes after changing the outputs. (2021-08-23)

  • issue298: Remove unused nodes after changing the outputs (2021-08-23)

  • issue301: Different build for manylinux on python 3.9 (2021-08-18)

  • issue300: Improves Lightgbm converter design + fix wrong prediction for TreeEnsemble with non contiguous arrays (2021-08-18)

  • issue297: Adds function to convert ONNX into numpy code. (2021-08-13)

  • issue296: Lightgbm + add function matmul to numpy API for ONNX (2021-08-07)

  • issue295: Implements runtime for operator FFT (2021-08-03)

  • issue291: Fixes infinite loop with operator loop, add support for static variables in Loop (2021-07-31)

  • issue294: Implements text representation of an ONNX graph (bigraph) (2021-07-30)

  • issue293: Add a tool to display an ONNX graph into text format (2021-07-30)

  • issue292: Adds operator AveragePool to the python runtime (2021-07-29)

  • issue290: Increases code coverage, add infer_size for Loop runtime (2021-07-28)

0.6.1522 - 2021-07-26 - 1.78Mb#

  • issue289: Avoids raising an exception when an optional parameter is not specified (2021-07-26)

  • issue288: Extends code coverage (2021-07-25)

  • issue287: Adds python runtime for operator Loop, SequenceInsert, ConcatFromSequence (2021-07-25)

  • issue286: Adds runtime for operator Range (2021-07-13)

0.6.1447 - 2021-07-12 - 2.56Mb#

  • issue285: Adds function cst to create constant with numpy API for ONNX (2021-07-12)

  • issue283: Commutative property (2021-07-12)

  • issue281: Infers temporary allocation needed while computing the outputs (2021-07-12)

  • issue284: Adds function transpose to numpy API for ONNX (2021-07-10)

  • issue282: Upgrade requirements to skl2onnx>=1.9.0 (2021-07-02)

  • issue280: More robustness for the python runtime (2021-07-01)

  • issue279: Implements method infer_types in OnnxInference (2021-06-28)

  • issue278: Adds operators ReduceSum, Max to OnnxMicroRuntime (2021-06-27)

  • issue277: Switch to python 3.9 in CI (2021-06-25)

  • issue276: Use openmp to parallelize QLinearConv (2021-06-25)

  • issue275: Adds new strategy to pick up the best einsum equation based on ML (2021-06-25)

  • issue274: Fixes issue raised with scipy 1.7.0 (2021-06-22)

  • issue273: Adds operator where, improves numpy api (x[x<0]= 2) (2021-06-18)

  • issue272: Explore custom implementation of operator add (2021-06-18)

  • issue271: Updates default opset from 13 to 14 (2021-06-17)

  • issue270: Adds more tests for QLinearConv runtime (2021-06-16)

  • issue269: Adds runtime for operator QLinearConv (2021-06-04)

  • issue268: Adds function to prepare data for onnxruntime_perf_test (2021-05-17)

  • issue267: Moves onnxruntime code inside a wrapper to reduce logs (2021-05-14)

  • issue266: Optimizes einsum even if not decomposed (2021-05-13)

  • issue265: Refactoring, moves files to onnx_tools (2021-05-12)

  • issue264: Support SessionOptions for runtime onnxruntime2 (2021-05-12)

  • issue263: Refactor einsum files (2021-05-06)

  • issue262: Refactoring, moving files into onnx_tools (2021-05-06)

  • issue261: Improves einsum decomposition by using gemm and removing a transpose (2021-05-05)

  • issue260: New command line to benchmark einsum decomposition (2021-05-03)

  • issue259: Minor changes to Einsum decomposition (2021-05-02)

  • issue258: Decomposes Einsum into simple matrix operations (2021-04-30)

  • issue257: Fixes #256, add method to validate input data in numpy API for ONNX (2021-04-20)

  • issue256: Add virtual method to validate input before predictions in numpy API for ONNX (2021-04-20)

0.5.1447 - 2021-04-17 - 0.38Mb#

  • issue255: Supports any embedded estimator with numpy API (2021-04-17)

  • issue254: Adds python runtime for operator ReduceL1 (2021-04-16)

  • issue253: Adds runtime for operator ReduceL2 (2021-04-14)

  • issue252: Implements an experimental version of reducesum for the case RK (2021-04-07)

  • issue251: Increases code coverage (2021-04-07)

  • issue250: Increases code coverage of unit tests (2021-04-03)

  • issue248: Adds implementation of BatchNormalization opset 14 (2021-03-29)

  • issue247: Introduces FctVersion to fix issue with optional arguments (2021-03-29)

  • issue246: Extends example on ReduceSum benchmark (2021-03-26)

  • issue244: Supports embedded models, complete tutorial on numpy API for ONNX (2021-03-26)

  • issue243: Add decorator to wrap converter for clustering (numpy API) (2021-03-17)

  • issue242: Add decorator to wrap converter for classifier (numpy API) (2021-03-17)

  • issue241: Add decorator to register scikit-learn classes with numpy API for ONNX (2021-03-14)

  • issue240: Add decorator to wrap converter for regressor (numpy API) (2021-03-14)

  • issue239: Add runtime empty (2021-03-13)

  • issue238: Use numpy API for ONNX to write custom converters (2021-03-13)

  • issue237: Add a unit test to check an exception (2021-03-10)

  • issue236: Implements __setitem__ for one dimension array (2021-03-08)

  • issue235: Supports profiling for runtime onnxruntime1 (2021-03-04)

  • issue233: Extend documentation about numpy API for ONNX (2021-03-04)

  • issue234: Add parameter overwrite to select_model_inputs_outputs (2021-03-03)

  • issue232: Implements pickling for functions used in numpy API for ONNX (2021-03-03)

  • issue231: Supports different inputs in select_model_inputs_outputs (2021-03-03)

  • issue230: Add unsqueeze, squeeze, expand_dims to numpy API for ONNX (2021-03-02)

  • issue229: Add method flatten, function pad to numpy API for ONNX (2021-03-01)

  • issue228: Improves numpy API for ONNX: type constraints (2021-03-01)

  • issue227: Add functions arange, cumsum, compress to numpy API for ONNX (2021-03-01)

  • issue226: Add function Einsum to numpy API for ONNX (2021-02-28)

  • issue225: Adds function Clip to numpy API for ONNX (2021-02-28)

  • issue224: Adds functions ceil, round to numpy API for onnx (2021-02-27)

  • issue223: Test numpy API against onnxruntime (2021-02-27)

  • issue222: Add hyperbolic function, prod, mean, argmin, argmax (2021-02-26)

  • issue221: Add many simple functions to numpy API for ONNX (2021-02-26)

  • issue220: Tutorial on numpy API for ONNX (2021-02-26)

  • issue219: Simplifies onnxfication of FunctionTransformer (2021-02-23)

  • issue218: Implements __setitem__ for class OnnxVar (2021-02-21)

  • issue217: Move custom operator to a specific method easier to maintain (2021-02-21)

  • issue216: Fix crash with Gather, TopK when k=0 or indices is empty. (2021-02-20)

  • issue215: Implements __getitem__ for OnnxVar (onnxnumpy) (2021-02-20)

  • issue214: Implements numpy functions with onnx (2021-02-19)

  • issue213: Add parameter show to plot_onnx. (2021-02-11)

  • issue212: Fixes #210, check first models from zoo, fix operator conv when B is not null (2021-02-05)

  • issue210: Investigate models from ONNX zoo (2021-02-05)

  • issue211: numpy 1.20 does not allow nan values in int64 arrays any more, fix a unit test about imputer (2021-02-02)

  • issue208: Add try catch around import in asv benchmark (2021-01-30)

  • issue207: Reduces greater batch size to 10.000 instead of 100.000. (2021-01-29)

  • issue205: Fixes asv configuration (2021-01-18)

  • issue206: Build wheel for all many platforms in CI (2021-01-17)

0.5.1360 - 2021-01-04 - 0.35Mb#

  • issue203: Enable Python 3.9, enable opset 13, upgrade version number (2021-01-04)

  • issue202: Enable opset 13 (ONNX) (2021-01-04)

  • issue201: Fixes #200, add support for float16 (2020-12-30)

  • issue200: Add support for bfloat16 (2020-12-30)

  • issue199: Fix unit tests recently failing due to onnxruntime update. (2020-12-15)

0.4.1352 - 2020-12-11 - 1.42Mb#

  • issue196: Fixes operator Slice for opset 9 (2020-12-11)

  • issue198: Fixes #197, add function to plot onnx graph with matplotlib (2020-12-09)

  • issue197: Add a function to plot an onnx graph into matplotlib (2020-12-09)

  • issue195: Fixes #194, add function to add an operator in the graph (2020-12-08)

  • issue194: Add a function to insert a cast operator between two nodes (2020-12-08)

  • issue193: Improves notebook coverage, update CI (2020-11-29)

  • issue192: Fixes #191, improves performance of TreeEnsemble (2020-11-28)

  • issue191: Improves performance of TreeEnsemble (2020-11-28)

  • issue190: Fixes #189, parallelization of Einsum (2020-11-17)

  • issue189: Introduce parallelization in experimental einsum implementation (2020-11-17)

  • issue188: Fixes #187, custom implementation for operator Einsum (2020-11-15)

  • issue187: Custom implementation for operator Einsum (2020-11-15)

  • issue186: Fixes #185, add operator LessOrEqual (2020-11-15)

  • issue185: Add operator LessOrEqual (2020-11-15)

  • issue181: Fix converter xgboost when ntree_limit is set up (2020-11-14)

  • issue184: Fixes #183, fix missing parameter black_op in OnnxPipeline (2020-11-07)

  • issue183: Fix error in OnnxPipeline, parameter black_op not found (2020-11-07)

  • issue182: Fixes #178, fix xgboost issue with ntree_limit (2020-11-07)

  • issue178: Fixes unit test testing OnnxConv (issue with shapes) (2020-11-07)

  • issue180: Fixes #179, fix guess_schema_from_data for categories (2020-11-03)

  • issue179: guess_schema_data_type fails with category in dataframe (2020-11-03)

  • issue176: Fixes #175, add operator dropout (2020-09-29)

  • issue175: Add operator Dropout (2020-09-29)

  • issue174: Add support for ReduceSum >= 13 (2020-09-21)

  • issue173: Fixes #172, add runtime for operator MaxPool (2020-09-16)

  • issue172: Add runtime for operator MaxPool (2020-09-16)

  • issue171: Fixes #170, add operator Pad (2020-09-10)

  • issue170: Add runtime for operator Pad (2020-09-10)

0.4.1259 - 2020-09-03 - 1.32Mb#

  • issue169: fix compiling issue with ubuntu 16.04 (2020-09-03)

  • issue167: Add runtime for Operator Or (2020-08-25)

  • issue166: Add runtime for operator And (2020-08-25)

  • issue165: Add runtime for operator GreaterOrEqual (2020-08-25)

  • issue164: Add runtime for operator If (2020-08-25)

  • issue163: Add runtime for operator Unsqueeze (2020-08-25)

  • issue162: Add runtime for operator Split (2020-08-25)

  • issue161: Add support for disable_optimisation (2020-08-12)

  • issue160: Fixes #159, add operator ConvTranspose, refactoring. (2020-08-07)

  • issue159: Implements runtime for ConvTranspose (2020-08-07)

  • issue158: Fixes benchmark import issues (2020-08-03)

  • issue157: Simplify scenarios, reduce time for benchmark. (2020-08-02)

  • issue156: Fixes #155, improves documentation (2020-08-02)

  • issue155: Fixes API on documentation (2020-08-02)

  • issue154: Fixes y_train dtype for most of the problems. Fixes subproblems with GridSearchCV (2020-07-31)

  • issue153: Fixes #152, set set n_jobs to the number of CPU (2020-07-31)

  • issue152: Set n_jobs to the number of core - 1 when doing benchmark (2020-07-31)

  • issue151: Force operator Conv to use continuous array (2020-07-30)

  • issue150: Fixes nan issue in operator conv (2020-07-29)

  • issue147: Fixes #145, #150, shape inference for operator Conv (2020-07-29)

  • issue145: Fixes missing shape inference for operator conv (2020-07-29)

  • issue149: Fixes #148, add operator Atan (2020-07-22)

  • issue148: Add operator atan (2020-07-22)

  • issue146: Fixes #144, add operator GlobalAveragePool (2020-07-21)

  • issue144: Implements operator GlobalAveragePool (2020-07-21)

  • issue143: Fixes #142, add operator BatchNormalization (2020-07-21)

  • issue142: Implement python runtime for operator BatchNormalization (2020-07-21)

  • issue141: Fixes #140, add runtime for QuantizeLinear, DequantizeLinear (2020-07-20)

  • issue140: Implement runtime for QuantizeLinear, DequantizeLinear (2020-07-20)

0.4.1204 - 2020-07-09 - 0.31Mb#

  • issue139: Add runtime for operator EyeLike (2020-07-08)

  • issue138: Add code to register custom python operator (2020-07-08)

  • issue137: Remove parameter dtype (onnx conversion) (2020-07-08)

  • issue136: Add parameter reshape to OnnxTransformer (2020-07-03)

  • issue135: Add a function to change the first dimension output (ONNX). (2020-07-03)

  • issue133: Implements runtime for operator Gather (ONNX) (2020-06-18)

  • issue132: Add operator StringNormalizer, Tokenizer, TfidfVectorizer (ONNX) (2020-06-15)

  • issue131: Add custom operator solve (2020-06-12)

  • issue130: Add operator Erf (ONNX) (2020-06-11)

  • issue129: Add operator Einsum (ONNX) (2020-06-11)

  • issue128: Fixes #127, implements OnnxPipeline, train, convert at each step (2020-06-08)

  • issue127: Implements a pipeline which replaces early stages by onnx (2020-06-08)

0.3.1129 - 2020-06-04 - 0.29Mb#

  • issue123: Enables opset 12 (ONNX) (2020-06-04)

  • issue117: Support for op_version in onnx grammar (2020-06-04)

0.3.1108 - 2020-05-20 - 0.29Mb#

  • issue126: Fix xgboost converter for xgboost >= 1.0 (2020-05-18)

  • issue125: Refactor rewritten sklearn operators (2020-05-18)

  • issue124: Fixes #122, capture standard C ouptput with dump_data_model, first step for #123 (2020-05-16)

  • issue122: Captures C output when calling dump_data_and_model (2020-05-16)

0.3.1082 - 2020-05-01 - 2.84Mb#

  • issue121: Add function to convert array to bytes and bytes to array (onnx tensor) (2020-04-30)

  • issue120: Fix discrepencies for SVM classifier (ONNX) (2020-04-30)

  • issue119: Keep order in topk implementation (2020-04-17)

  • issue118: opset is not propagated in OnnxTransformer (2020-04-09)

0.3.1070 - 2020-04-07 - 0.29Mb#

  • issue115: Add a function to replay a benchmark when this one was dumped (more accurate) (2020-04-06)

  • issue116: Makes ZipMapDictionary picklable (2020-03-30)

  • issue114: Add more parameters to specify benchmark time (2020-03-30)

  • issue113: Add operators for opset 12 (2020-03-26)

  • issue112: Number of feature is wrong for problem num-tr-clus (2020-03-20)

0.3.1029 - 2020-03-17 - 0.28Mb#

  • issue111: Reduce the number of allocation in TreeEnsemble when it is parallelized (cache) (2020-03-13)

  • issue110: Implements runtime for operator Constant-12 (2020-03-06)

  • issue109: Generate a benchmark with asv to compare different runtime. Update modules in asv. (2020-03-06)

  • issue108: Add a function to reduce the memory footprint (2020-02-25)

  • issue106: Add operator Neg (2020-02-25)

  • issue101: Fix DecisionTreeClassifier disappearance on the benchmark graph (2020-02-25)

  • issue107: Add operator IsNaN (2020-02-24)

  • issue105: Support string labels for Linear, TreeEnsemble, SVM classifiers. (2020-02-24)

  • issue104: Enable / disable parallelisation in topk (2020-02-23)

  • issue103: Implements plot benchmark ratio depending on two parameters (2020-02-22)

  • issue102: Fix conversion for xgboost 1.0 (2020-02-21)

  • issue100: add notebook on TreeEnsemble (2020-02-19)

  • issue99: Fixes #93, use same code for TreeEnsembleClassifier and TreeEnsembleRegression (2020-02-19)

  • issue93: Use pointer for TreeClassifier (2020-02-19)

  • issue98: mlprodict i broken after onnxruntime, skl2onnx update (2020-02-15)

  • issue97: Add runtime for operator Conv (2020-01-24)

  • issue96: Fixes #97, add runtime for operator Conv (2020-01-24)

  • issue95: Fix OnnxInference where an output and an operator share the same name (2020-01-15)

  • issue94: Raw scores are always positive for TreeEnsembleClassifier (binary) (2020-01-13)

  • issue90: Implements a C++ runtime for topk (2019-12-17)

  • issue86: Use pointers to replace treeindex in tree ensemble cpp runtime (2019-12-17)

  • issue92: Implements a C++ version of ArrayFeatureExtractor (2019-12-14)

  • issue89: Implements a function which extracts some informations on the models (2019-12-14)

  • issue88: Fix bug in runtime of GatherElements (2019-12-14)

  • issue87: Add converter for HistGradientBoostRegressor (2019-12-09)

  • issue85: Implements a precompiled run method in OnnxInference (runtime=’python_compiled’) (2019-12-07)

  • issue84: Automatically creates files to profile time_predict function in the benchmark with py-spy (2019-12-04)

  • issue83: ONNX: includes experimental operators in the benchmark (2019-12-04)

  • issue82: Function translate_fct2onnx: use of opset_version (2019-12-04)

  • issue81: ONNX benchmark: track_score returns scores equal to 0 or 1 (unexpected) (2019-12-04)

  • issue80: ONNX: extend benchmark to decision_function for some models (2019-12-03)

  • issue77: Improves ONNX benchmark to measure zipmap impact. (2019-12-03)

  • issue76: Implements ArgMax 12, ArgMax 12 (python onnx runtime) (2019-11-27)

  • issue75: ONNX: fix random_state whevever it is available when running benchmark (2019-11-27)

  • issue59: ONNX: Investigate kmeans and opset availability. (2019-11-21)

  • issue66: ONNX: improves speed of python runtime for decision trees (2019-11-19)

  • issue74: Function _modify_dimension should return the same dataset if called the same parameter (even if it uses random functions) (2019-11-15)

  • issue73: ONNX: fix links on benchmark page (opset is missing) (2019-11-07)

  • issue72: ONNX: support of sparse tensor for a unary and binary python operators (2019-11-06)

  • issue71: ONNX: add operator Constant (2019-11-06)

  • issue67: ONNX: improves speed of svm regressor (2019-11-06)

  • issue70: ONNX: write tools to test convervsion for models in scikit-learn examples (2019-10-29)

  • issue65: ONNX: investigate discrepencies for k-NN (2019-10-28)

  • issue69: ONNX: side by side should work by name and not by positions (2019-10-23)

  • issue68: ONNX: improves speed of SGDClassifier (2019-10-23)

  • issue61: Implements a function to create a benchmark based on asv (ONNX) (2019-10-17)

  • issue63: Export asv results to csv (ONNX) + command line (2019-10-11)

  • issue64: Add an example with lightgbm and categorical variables (ONNX) (2019-10-07)

  • issue62: Implements command line for the asv benchmark (ONNX) (2019-10-04)

  • issue60: Improve lightgbm converter (ONNX) (2019-09-30)

  • issue58: Fix table checking model, merge is wrong in documentation (2019-09-20)

  • issue57: ONNX: handles dataframe when converting a model (2019-09-15)

  • issue56: ONNX: implements cdist operator (2019-09-12)

  • issue54: ONNX: fix summary, it produces multiple row when model are different when opset is different (2019-09-12)

  • issue51: ONNX: measure the time performance obtained by using optimization (2019-09-11)

  • issue52: ONNC-cli: add a command line to optimize an onnx model (2019-09-10)

  • issue49: ONNX optimization: remove redundant subparts of a graph (2019-09-09)

  • issue48: ONNX optimization: reduce the number of Identity nodes (2019-09-09)

  • issue47: Implements statistics on onnx graph and sklearn models, add them to the documentation (2019-09-06)

  • issue46: Implements KNearestNeibhorsRegressor supporting batch mode (ONNX) (2019-08-31)

  • issue45: KNearestNeighborsRegressor (2019-08-30)

  • issue44: Add an example to look into the performance of every node for a particular dataset (2019-08-30)

  • issue43: LGBMClassifier has wrong shape (2019-08-29)

  • issue42: Adds a graph which visually summarize the validating benchmark (ONNX). (2019-08-27)

  • issue41: Enables to test multiple number of features at the same time (ONNX) (2019-08-27)

  • issue40: Add a parameter to change the number of featuress when validating a model (ONNX). (2019-08-26)

  • issue39: Add a parameter to dump all models even if they don’t produce errors when being validated (ONNX) (2019-08-26)

  • issue24: support double for TreeEnsembleClassifier (python runtime ONNX) (2019-08-23)

  • issue38: See issue on onnxmltools. https://github.com/onnx/onnxmltools/issues/321 (2019-08-19)

  • issue35: Supports parameter time_kwargs in the command line (ONNX) (2019-08-09)

  • issue34: Add intervals when measuring time ratios between scikit-learn and onnx (ONNX) (2019-08-09)

  • issue31: Implements shape inference for the python runtime (ONNX) (2019-08-06)

  • issue15: Tells operator if the execution can be done inplace for unary operators (ONNX). (2019-08-06)

  • issue27: Bug fix (2019-08-02)

  • issue23: support double for TreeEnsembleRegressor (python runtime ONNX) (2019-08-02)

  • issue26: Tests all converters in separate processeses to make it easier to catch crashes (2019-08-01)

  • issue25: Ensures operator clip returns an array of the same type (ONNX Python Runtime) (2019-07-30)

  • issue22: Implements a function to shake an ONNX model and test float32 conversion (2019-07-28)

  • issue21: Add customized converters (2019-07-28)

  • issue20: Enables support for TreeEnsemble operators in python runtime (ONNX). (2019-07-28)

  • issue19: Enables support for SVM operators in python runtime (ONNX). (2019-07-28)

  • issue16: fix documentation, visual graph are not being rendered in notebooks (2019-07-23)

  • issue18: implements python runtime for SVM (2019-07-20)

  • issue17: add a mechanism to use ONNX with double computation (2019-07-15)

  • issue13: add automated benchmark of every scikit-learn operator in the documentation (2019-07-05)

  • issue12: implements a way to measure time for each node of the ONNX graph (2019-07-05)

  • issue11: implements a better ZipMap node based on dedicated container (2019-07-05)

  • issue8: implements runtime for decision tree (2019-07-05)

  • issue7: implement python runtime for scaler, pca, knn, kmeans (2019-07-05)

  • issue10: implements full runtime with onnxruntime not node by node (2019-06-16)

  • issue9: implements a onnxruntime runtime (2019-06-16)

  • issue6: first draft of a python runtime for onnx (2019-06-15)

  • issue5: change style highlight-ipython3 (2018-01-05)

previous

Tricky detail when converting a random forest from scikit-learn into ONNX