-
Notifications
You must be signed in to change notification settings - Fork 4
/
CHANGES
216 lines (135 loc) · 9.79 KB
/
CHANGES
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
* 2020.08.19 * [FGG] added module:paper
* 2019.10.25 * added support for versioning in experiments
* 2018.11.03 * https://github.com/ctuning/ck-analytics/pull/9
* 2018.05.10 * added CK model.species for AlexNet (for distributed inference workflow from ReQuEST submission)
* 2018.04.07 * added CK model.species for MobileNetsV2 and ResNet 18
* 2018.04.03 * improved experiment selector for dashboard (for ReQuEST)
* 2018.03.21 * added ResNet50 and Inception v3 model species (for ReQuEST)
* 2018.03.19 * added MobileNet model.species (for ReQuEST)
* 2018.03.16 * improved D3 scatter visualization (xerr)
* 2018.03.13 * improved result visualization (for ReQuEST)
* 2018.02.25 * changed logic in ck replay (no reusing deps by default and rebuild packages)
* 2018.02.21 * added dummy "model.species" to identify equivalent models from different AI engines
* 2018.02.19 * improving experiment search pruning for ReQuEST tournaments
* 2018.02.02 * removed ck-tensorflow repo dependency. CK will ask user to install it if needed.
* 2017.11.23 * added unified access to TensorFlow DNN Classifier,
DNN Linear Combined Classifier
and Linear Classifier (model.tf)
* fixed a bug when buiding very large decision trees via CK scikit-learn
* 2017.10.10 * added support to rebuild deps during replay from another machine using package UOA
* 2017.10.06 * major update to the universal experiment selector
* 2017.10.05 * added unified experiment selector (useful for DNN crowd-tuning)
* 2017.09.23 * improved html table with experiments
* improved speed of replay output comparison
* 2017.09.09 * added support in D3 graps to change size and color of individual points
(useful to visualize DNN co-design results)
* 2017.08.29 * added reduce_bug option to localize bugs during compiler fuzzing (reducing complexity)
* 2017.08.28 * fixing ck replay experiment (proper rebuilding of choices)!
* 2017.08.26 * added customization of y axis labels
* 2017.08.25 * added another heatmap example (useful for ML-based autotuning)
* 2017.07.30 * added "process_plus_minus math.variation" to round results
* added universal HTML & TEX table generator ("prepare table") -
useful for interactive reports
* 2017.07.27 * added individual interactive graph customization (JSON meta)
* 2017.07.26 * improvements in interactive D3 graphs (multiple in the same report)
* 2017.07.24 * minor fix in interactive graphs (if multiple in one CK entry)
* 2017.07.15 * updated replay function (iteratively reducing complexity)
* 2017.07.11 * added further configuration of 2D scatter graphs "point_style" (for Pareto and errorbar, etc):
"capsize" - size of error bars
"leave_only_permanent"
"remove_permanent"
* 2017.07.03 * added plot customization
* 2017.05.14 * added support for Caffe2 unified classification
* 2017.05.05 * added unified CK AI JSON API for image classification
* 2017.04.30 * started improving collaborative AI API
* 2017.04.08 * improved interactive and reproducible report module
* 2017.04.02 * improved MPL scatter graphs (connected_lines and labels)
* 2017.02.24 * added support for x_ticks_period in D3 interactive graphs
* 2017.01.14 * improved replay support (for OpenCL kernels from Lift/CLBlast during crowd-tuning)
* 2016.12.13 * added '--prune' to 'ck replay experiment'
* 2016.12.12 * added automatic detection of views when browsing experiments
* added 'ck browse experiment:{UOA}' to view local experiments in a browser
* 2016.11.07 * prune experiment crowdsourcing scenarios if "skip_from_cmd"
* 2016.09.15 * adding partial support for MediaWiki format in CK-powered interactive reports
* 2016.09.01 * added experiment.raw module
* 2016.08.23 * added multi-object classification using decision trees from skiearn
* 2016.07.26 * added geometric mean
* 2016.07.20 * added jnotebook module to preserve, clean outputs, and run (ipython) jupyter notebooks
* 2016.05.30 * support for new CK web server
* 2016.05.27 * added drawing of variation in D3 scatter
(used to show empirical results during crowd-tuning)
* 2016.05.24 * starting speedup function to compare 2 lists of samples,
perform t-test, check significance, etc.
* 2016.05.21 * possibility to connect lines in scatter graph
* 2016.05.20 * improving D3 visualization (lines + customized URLs)
* 2016.02.26 * adding logging functions during experiment crowdsourcing
* 2016.02.25 * preparing experiment crowdsourcing via mobile devices
(such as collaborative program optimization using Android mobile phones)
* 2016.02.05 * various small fixes and enhancements in experiment replay
* 2016.02.04 * adding d3_2d_bars to draw speedups for all solutions in HTML during crowdtuning
* 2016.02.03 * adding ck.type_long for Python2/3 compatibility
* 2016.02.02 * slightly changing the logic of conditions for multi-objective optimization
* fixing a minor bug in experiment processing
* 2016.02.01 * added mode to skip recording pipeline/desc info to experiment
(to avoid sending huge files between servers during optimization crowdsourcing)
* added support to pack experiments (to be able to send via Internet
(during experiment crowdsourcing such as program crowdtuning)
* 2016.01.31 * fixing calculation of improvements during multi-objective autotuning
* adding more conditions "==" & "!="
* 2016.01.29 * general improvements
* adding calculation of multi-dimensional improvements
* adding universal conditions on solutions during auto-tuning
(such as improve energy without sacrificing execution time,
or improve execution time without increasing code size ...)
* 2016.01.28 * adding a few functions to automatically detect scenario
(such as multi-objective autotuning) and pick up parameters
when plotting graphs
* 2016.01.27 * improving experiment crowdsourcing (making it universal)
* 2016.01.20 * fixing a stupid bug in cmd processing (thanks to Andrei Lascu)
* 2016.01.15 * Adding margins when comparing values on Pareto frontier
(used to monitor only significant speedups when crowdsourcing autotuning)
* 2016.01.11 * Adding ck crowdsource experiment - universal experiment crowdsourcing engine
(to support our collaborative OpenCL/CUDA/compiler flag program optimization)
* 2015.12.18 * When reproducing a given experiment, allow to compare results with another experiment
(for example with default one) using --compare_data_uoa
* When reproducing experiment in Linux and specifying dimensions to check from CMD,
add possibility to substitute ^ with # (to check long flat keys)
* 2015.12.15 * adding support to prune experimental vectors by threshold (when retrieving results):
"vector_thresholds" and "vector_thresholds_conditions" - useful to keep or skip
points with high variation, for example.
* adding "force_pipeline_update" key to properly update pipelines (workflows) during replay
* 2015.12.14 * fixing bug in replay (by accident removed update of choices in pipeline)
* 2015.12.13 * improving reproducibility (experiment replay):
* adding '--deps' flag to reuse all SW deps when reproducing
experiment on the same machine;
* adding '--dims' to compare only one flat key (or use wildcards)
* fixing bug with statistical repetitions
(now taking the correct number from a given experiment)
* 2015.12.07 * taking statistical repetition number from the entry when replaying a given experiment
* passing 'local_platform' and 'skip_clean_after' to program pipeline when
replaying such experiments ...
* fixing experiment dictionary output comparison during replay
* 2015.10.09 * added "zmin" and "zmax" params for 3D graphs
* added "invert_x_axis", "invert_y_axis" and "invert_z_axis" to 3D graphs
* added "x_ticks_scale", "y_ticks_scale", "z_ticks_scale" to 3D graphs
* added "color_dict" to 3D graphs
* 2015.10.08 * added 'fmt' parameter to graphs
* added 'font' parameters for graphs (size, weight, family)
* 2015.09.19 * added 'skip' flag to skip comparison of results during experiment 'replay'
* 2015.09.11 * added demo/ml-decision-tree - simple decision tree maker (suggested by Sergey Yakushkin)
* added 'use' in 'model' to perform predictions
* added model recording to CK entry and using model from CK entry
(to share and collaboratively improve predictive models)
* 2015.09.10 * added 'advice' module (about experiments, bugs, models, features, optimizations, adaptation, community remarks ...)
* 2015.09.09 * added 'skip_stat_analysis' to compact recorded data when stat analysis is not needed
(useful for CLSmith)
* 2015.09.03 * fixed bug in showing +- % when 0 (noticed by Anton)
* 2015.09.02 * added extension to keys when retrieving experiments
* 2015.08.12 * added possibility to reverse key order in Pareto frontier filter during multi-objective autotuning
(useful for FPS in slambench, for example)
* improving interactive graphs for interactive articles
* 2015.08.11 * added output of decision trees to png (for interactive reports and articles via CK)
* added horizontal and vertical lines for graphs (used in graphs for adaptive scheduling (predictive modeling and active learning))
* 2015.08.09 * added extra dict when recording experiments (useful to record name and subview_uoa, for example)
* fixing a few bugs in experiment viewing
* 2015.08.07 added reverse sorting in experiment tables