Permalink
Switch branches/tags
Nothing to show
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
276 lines (206 sloc) 6.8 KB
submission 11 - 0.68505
=============
This is submission 10 + an morphological close operation (see postprocess_morpho notebook)
submission 13 - 0.66688
=============
19/train_fccn_relu
submission 17 - 0.66056
=============
22_restart_from_sub11/0_train_fcnn_maxout.ipynb
submission 18 - 0.63071
=============
22_restart_from_sub11/1_train_fcnn_relu.sub18.ipynb
submission 19 - 0.65681
=============
22_restart_from_sub11/2_train_fcnn_maxout_bn.sub19.ipynb
submission 20 - 0.63599
=======================
22_restart_from_sub11/2_train_fcnn_maxout_bn.sub20.ipynb
submission 22,23,24 - 0.63611
=======================
22_restart_from_sub11/2_train_fcnn_maxout_nobn.sub21.ipynb - without opencv morphology
0.63611 with bin_thresh = 0.95 (=sub 21)
0.65601 with bin_thresh = 0.5 (=sub 22)
0.65426 with bin_thresh = 0 (=sub 24)
submission 25 - 0.65099
=======================
23_fcnn_sub25/1_train_fcnn_maxout_nobn.ipynb
submission 26 - 0.66759
=======================
24_fcnn_fulltrain/1_train_fcnn_maxout.sub26.ipynb
submission 27 - 0.67807
=======================
26_ensemble/0_ensemble_predict.sub27.ipynb
submission 28 - 0.65849
=======================
25_fcnn_fulltrain/1_train_fcnn_relu_nobn.ipynb
submission 29 - 0.65486
=======================
26_ensemble/0_ensemble_predict.sub27.ipynb
submission 30 - 0.68691
=======================
outmap, outmap4, outmap8 and outbin
27_fcnn_auxiliary.sub30/1_train_fcnn_auxiliary.sub30.ipynb
submission 31 - 0.65596
=======================
This has outmap and outmap4
31_fcnn_auxiliary_clean2.sub31/1_train_fcnn_auxiliary.sub31.ipynb
submission 32 - 0.66525
=======================
33_fcnn_auxiliary.relu/1_train_fcnn_auxiliary_relu_zscore.ipynb
Includes z-score normalisation of data
Opencv morphology post-processing
Optimize for 40% mask in kaggle's test
submission 33 - 0.66406
=======================
34_fcnn_auxiliary/1_train_fcnn_auxiliary.sub33.ipynb
Like submission 30, but trained on data_clean2_128_train.npz
Opencv morphology post-processing
Optimize for 40% mask in kaggle's test
submission 34 - 0.68211
=======================
train/test split based on patient ids
35_fcnn_auxiliary.sub34/1_train_fcnn_auxiliary.sub34.ipynb
submission 35 - 0.68223
=======================
this is submission 34 + notebooks/_ellipse_clean
submission 36 - 0.68547
=======================
this is submission 30 + notebooks/_ellipse_clean
submission 37 - 0.68579
=======================
this is the notebook of 36_fcnn_auxiliary.sub37
this is with map_thresh=0.4, bin_thresh=0.35 (39% with masks)
submission 38 - 0.68719
=======================
37_fcnn_auxiliary.vary_thresholds/1_predict_from_sub37model.ipynb
same model as submission 37, but with
map_thresh = 0.4, bin_thresh = 0.5 (37.78% with masks)
submission 39 - 0.68414
=======================
37_fcnn_auxiliary.vary_thresholds/1_predict_from_sub37model.ipynb
same model as submission 37, but with
map_thresh = 0.4, bin_thresh = 0.6 (33.71% with masks)
submission 40 - 0.68675
=======================
37_fcnn_auxiliary.vary_thresholds/1_predict_from_sub37model.ipynb
same model as submission 37, but with
map_thresh = 0.3, bin_thresh = 0.5 (39.16% with masks)
submission 41 - 0.67794
=======================
37_fcnn_auxiliary.vary_thresholds/1_predict_from_sub37model.ipynb
same model as submission 37, but with
map_thresh = 0.5, bin_thresh = 0.5 (34.69% with masks)
submission 42 - 0.69086
=======================
38_merge_submissions/merge_11_30_34_38.sub42.ipynb (35.42% with masks)
submission 43 - 0.67686
=======================
37_fcnn_auxiliary.vary_thresholds/1_predict_from_sub37model.ipynb
same model as submission 37, but with
map_thresh = 0.3, bin_thresh = 0.0 (49.22% with masks)
submission 44 - 0.68870
=======================
this is submission 42 with notebooks/ellipse_clean post-processing
NOTE
====
There is an error in the fcnn code used for a lot of previous submission :
# upsample the classif map
x = merge([
UpSampling2D((2, 2))(outmap4),
UpSampling2D((4, 4))(outmap5),
pool3
], mode='concat', concat_axis=1)
x = Convolution2D(16, 3, 3, border_mode='same')(pool3)
The conv layer should take x, not only pool3 as input
submission 45 - 0.68420
=======================
41_fcnn_auxiliary.correct_topo/1_train_fcnn_auxiliary.ipynb
submission 46 - 0.67410
=======================
42_fcnn_auxiliary.correct_topo.pool_upsampling/1_train_fcnn_auxiliary.ipynb
with
map_thresh = 0.4
bin_thresh = 0.35
submission 47 - 0.67609
=======================
42_fcnn_auxiliary.correct_topo.pool_upsampling/1_train_fcnn_auxiliary.ipynb
with
map_thresh = 0.5
bin_thresh = 0.5
submission 48 - 0.69055
=======================
42_fcnn_auxiliary.correct_topo.pool_upsampling/1_train_fcnn_auxiliary.ipynb
modified to remove internal sigmoid activations
with
map_thresh = 0.3
bin_thresh = 0.5 (37% masks)
submission 49 - 0.69167
=======================
submission 48 + masks_pca_clean
using 20 principal components
submission 50 - 0.68229
=======================
43_fcnn_auxiliary.correct_topo.vary_threshold/1_run_sub_48.ipynb
map_thresh = 0.3
bin_thresh = 0.3 (45% masks)
+ masks_pca_clean
using 20 principal components
submission 51 - 0.66727
=======================
44_fcnn_auxiliary.correct_topo.bigger.all_clean_train/1_train_fcnn_auxiliary.ipynb
map_thresh = 0.5
bin_thresh = 0.5 (43.10% with masks)
submission 52 - 0.67014
=======================
sub 51 + pca clean with 20 principal components
submission 53 - 0.66890
=======================
44_fcnn_auxiliary.correct_topo.bigger.all_clean_train/1_train_fcnn_auxiliary.ipynb
map_thresh = 0.5
bin_thresh = 0.8 (38.60% with masks)
+ pca clean with 20 principal components
submission 54 - 0.69237
=======================
this is submission 42 + mask pca
submission 55 - 0.68870
=======================
45_merge_submissions/merge_8_subs.sub55.ipynb
submission 56 - 0.69025
=======================
this is submission 55 + mask pca
submission 57 - 0.67852
=======================
46_merge_submissions.sub57/merge_3_keep_all.ipynb
+ clean by mask pca
submission 58 - 0.67519
=======================
47_fcnn_auxiliary.pool_upsampling/1_train_fcnn_auxiliary.ipynb
with
map_thresh = 0.4
bin_thresh = 0.4 (39.40% with masks)
+ pca clean
submission 59 - 0.67653
=======================
like sub58 but with different thresholds
map_thresh = 0.3
bin_thresh = 0.3 (44.90% with masks)
+ pca clean
submission 60 - 0.67240
=======================
similar to sub58 and sub59, but with shorter training (25 epochs)
48_fcnn_auxiliary.pool_upsampling.25epochs/1_train_fcnn_auxiliary.ipynb
map_thresh = 0.45
bin_thresh = 0.7 (35.38% with masks)
submission 61 - 0.68374
=======================
sub60 but with
map_thresh = 0.45
bin_thresh = 0.5 (41.90% with masks)
+ pca clean
submission 62 - 0.66106
=======================
49_fcnn_auxiliary.pool_upsampling.small/1_train_fcnn_auxiliary.ipynb
map_thresh = 0.4
bin_thresh = 0.4 (37.55% with masks)
+ pca clean