{
  "id": 96885,
  "title": "EDA and more in C++",
  "url": "/competitions/champs-scalar-coupling/discussion/96885",
  "author_name": "rilesdg3",
  "post_date": "2019-06-24T09:02:00.009000",
  "votes": 8,
  "comment_count": 17,
  "views": 0,
  "content": "<p>After working with Python for some time I have decided to try data science/machine learning with C++. I found python to be very slow and take up way to much memory. For this competition Python would probably work just fine, but I am building out my libraries for C++.</p>\n\n<p>The code, charts, and write up will live <a href=\"https://github.com/rilesdg3/Kaggle-CHAMPS\">here</a>.  The write up live here as well.</p>\n\n<p>I will update this page and the Github pages as I progress.</p>\n\n<p><strong>Update 07 25 2019</strong> I put new code up on GitHub It is not the greatest but will build a neural network using Caffe2 api</p>\n\n<p>All scalar coupling\nmax 204.88 mean 15.9216 median 2.27058 min -36.2186 range 241.099 sd 34.942 skewness 2.02425 var 1220.94 </p>\n\n<p>Simple statistics for scalar coupling grouped by <em>type</em>.\n<strong>1JHC</strong>\nmax 204.88 mean 94.9762 median 88.0455 min 66.6008 range 138.279 sd 18.2772 skewness 2.72468 var 334.057 \n<strong>1JHN</strong>\nmax 80.4187 mean 47.4799 median 47.8237 min 24.3222 range 56.0965 sd 10.9222 skewness 0.0724719 var 119.294 \n<strong>2JHC</strong>\nmax 42.8192 mean -0.270624 median -0.941388 min -36.2186 range 79.0378 sd 4.52361 skewness 2.91453 var 20.4631 \n<strong>2JHH</strong>\nmax 11.8542 mean -10.2866 median -11.344 min -35.1761 range 47.0303 sd 3.97961 skewness 1.62319 var 15.8373 \n<strong>2JHN</strong>\nmax 17.7436 mean 3.12475 median 2.03526 min -2.62085 range 20.3645 sd 3.67347 skewness 1.10857 var 13.4944 \n<strong>3JHC</strong>\nmax 76.0437 mean 3.68847 median 2.89505 min -18.5821 range 94.6258 sd 3.07091 skewness 1.16998 var 9.43047 \n<strong>3JHH</strong>\nmax 17.4841 mean 4.77102 median 3.71061 min -3.02046 range 20.5046 sd 3.70498 skewness 0.645849 var 13.7269 \n<strong>3JHN</strong>\nmax 10.9712 mean 0.99073 median 0.662068 min -3.17241 range 14.1436 sd 1.31539 skewness 1.64438 var 1.73026 </p>\n\n<p>Simple Stats for scalar coupling contributions grouped by <em>type</em>. A chart of the histograms will be posted to the Github page</p>\n\n<p><strong>1JHC</strong>    dso max 1.70161 mean    0.937555    median  0.95363 min 0.27201 range   1.4296  sd  0.195684    skewness    0.0195439   var 0.0382921\n<strong>1JHN</strong>    dso max 0.417811    mean    0.26821 median  0.253365    min 0.067922    range   0.349889    sd  0.0505772   skewness    0.425214    var 0.00255805\n<strong>2JHC</strong>    dso max 0.591145    mean    -0.0789262  median  -0.0506959  min -1.24942    range   1.84057 sd  0.186024    skewness    -1.82861    var 0.034605\n<strong>2JHH</strong>    dso max -0.722259   mean    -2.2633 median  -2.2634 min -6.85679    range   6.13453 sd  0.513059    skewness    -3.13265    var 0.26323\n<strong>2JHN</strong>    dso max 0.107937    mean    -0.0404772  median  -0.0365436  min -0.437965   range   0.545902    sd  0.0479925   skewness    -0.670977   var 0.00230328\n<strong>3JHC</strong>    dso max 0.728981    mean    -0.147708   median  -0.188094   min -0.927072   range   1.65605 sd  0.265384    skewness    0.0260897   var 0.0704285\n<strong>3JHH</strong>    dso max 1.53109 mean    -0.984241   median  -0.490005   min -4.85343    range   6.38452 sd  1.185   skewness    -0.671356   var 1.40423\n<strong>3JHN</strong>    dso max 0.15681 mean    -0.060585   median  -0.0821114  min -0.271433   range   0.428243    sd  0.082979    skewness    0.146923    var 0.00688551\n<strong>1JHC</strong>    fc  max 205.517 mean    93.4526 median  86.6223 min 64.7082 range   140.809 sd  18.4211 skewness    2.73049 var 339.336\n<strong>1JHN</strong>    fc  max 79.0258 mean    45.7    median  45.8877 min 23.6274 range   55.3984 sd  10.8646 skewness    0.105357    var 118.039\n<strong>2JHC</strong>    fc  max 42.9676 mean    -0.306388   median  -1.00951    min -37.224 range   80.1916 sd  4.44987 skewness    2.86531 var 19.8014\n<strong>2JHH</strong>    fc  max 12.058  mean    -10.4573    median  -11.5036    min -35.3676    range   47.4256 sd  4.05017 skewness    1.63082 var 16.4039\n<strong>2JHN</strong>    fc  max 18.4756 mean    3.16404 median  1.95665 min -2.13019    range   20.6058 sd  3.83202 skewness    1.18234 var 14.6844\n<strong>3JHC</strong>    fc  max 76.4528 mean    3.76597 median  2.95253 min -18.1236    range   94.5764 sd  3.1088  skewness    1.16408 var 9.66467\n<strong>3JHH</strong>    fc  max 18.1148 mean    4.92025 median  3.69917 min -3.21961    range   21.3344 sd  3.86353 skewness    0.803262    var 14.9268\n<strong>3JHN</strong>    fc  max 11.0317 mean    1.05408 median  0.681106    min -3.08096    range   14.1127 sd  1.31317 skewness    1.71729 var 1.72441\n<strong>1JHC</strong>    pso max 1.3303  mean    0.387471    median  0.4927  min -2.38873    range   3.71903 sd  0.460782    skewness    -0.819143   var 0.21232\n<strong>1JHN</strong>    pso max 2.15194 mean    1.4268  median  1.45848 min -0.331303   range   2.48324 sd  0.237667    skewness    -0.971366   var 0.0564855\n<strong>2JHC</strong>    pso max 5.3263  mean    0.0394564   median  0.0433233   min -3.28811    range   8.61441 sd  0.553037    skewness    5.77545 var 0.30585\n<strong>2JHH</strong>    pso max 8.20153 mean    2.06026 median  2.04252 min 0.663344    range   7.53819 sd  0.472996    skewness    4.30584 var 0.223725\n<strong>2JHN</strong>    pso max 2.34189 mean    -0.00310926 median  0.0891185   min -1.22959    range   3.57148 sd  0.245205    skewness    -1.78983    var 0.0601254\n<strong>3JHC</strong>    pso max 1.04693 mean    0.0855547   median  0.0975408   min -1.09216    range   2.13909 sd  0.193157    skewness    -0.109494   var 0.0373095\n<strong>3JHH</strong>    pso max 3.69055 mean    0.773651    median  0.433427    min -1.02461    range   4.71516 sd  0.827177    skewness    0.744086    var 0.684223\n<strong>3JHN</strong>    pso max 0.555222    mean    0.00705348  median  0.0137612   min -0.654522   range   1.20974 sd  0.103286    skewness    -0.800263   var 0.0106681\n<strong>1JHC</strong>    sd  max 2.11149 mean    0.198496    median  0.169054    min -0.808604   range   2.92009 sd  0.127312    skewness    2.18674 var 0.0162082\n<strong>1JHN</strong>    sd  max 0.226941    mean    0.0848519   median  0.0808529   min -0.164922   range   0.391863    sd  0.0558663   skewness    -0.179632   var 0.00312104\n<strong>2JHC</strong>    sd  max 2.29669 mean    0.0752328   median  0.0759824   min -3.18491    range   5.4816  sd  0.109143    skewness    3.13352 var 0.0119121\n<strong>2JHH</strong>    sd  max 0.775729    mean    0.373714    median  0.369694    min -0.215389   range   0.991118    sd  0.0376742   skewness    -0.0199843  var 0.00141934\n<strong>2JHN</strong>    sd  max 0.537983    mean    0.00430365  median  0.00584075  min -0.746569   range   1.28455 sd  0.0356897   skewness    0.0726665   var 0.00127376\n<strong>3JHC</strong>    sd  max 1.16598 mean    -0.0153426  median  -0.0225856  min -1.4553 range   2.62128 sd  0.0417047   skewness    0.514652    var 0.00173928\n<strong>3JHH</strong>    sd  max 1.61011 mean    0.0613592   median  0.0629828   min -1.08987    range   2.69998 sd  0.0634342   skewness    -0.343884   var 0.00402389\n<strong>3JHN</strong>    sd  max 0.169883    mean    -0.00981557 median  -0.00855217 min -0.68493    range   0.854813    sd  0.0245621   skewness    -0.928331   var 0.000603296</p>",
  "messages": [
    {
      "id": 559549,
      "postDate": "2019-06-24T09:02:00.010Z",
      "content": "<p>After working with Python for some time I have decided to try data science/machine learning with C++. I found python to be very slow and take up way to much memory. For this competition Python would probably work just fine, but I am building out my libraries for C++.</p>\n\n<p>The code, charts, and write up will live <a href=\"https://github.com/rilesdg3/Kaggle-CHAMPS\">here</a>.  The write up live here as well.</p>\n\n<p>I will update this page and the Github pages as I progress.</p>\n\n<p><strong>Update 07 25 2019</strong> I put new code up on GitHub It is not the greatest but will build a neural network using Caffe2 api</p>\n\n<p>All scalar coupling\nmax 204.88 mean 15.9216 median 2.27058 min -36.2186 range 241.099 sd 34.942 skewness 2.02425 var 1220.94 </p>\n\n<p>Simple statistics for scalar coupling grouped by <em>type</em>.\n<strong>1JHC</strong>\nmax 204.88 mean 94.9762 median 88.0455 min 66.6008 range 138.279 sd 18.2772 skewness 2.72468 var 334.057 \n<strong>1JHN</strong>\nmax 80.4187 mean 47.4799 median 47.8237 min 24.3222 range 56.0965 sd 10.9222 skewness 0.0724719 var 119.294 \n<strong>2JHC</strong>\nmax 42.8192 mean -0.270624 median -0.941388 min -36.2186 range 79.0378 sd 4.52361 skewness 2.91453 var 20.4631 \n<strong>2JHH</strong>\nmax 11.8542 mean -10.2866 median -11.344 min -35.1761 range 47.0303 sd 3.97961 skewness 1.62319 var 15.8373 \n<strong>2JHN</strong>\nmax 17.7436 mean 3.12475 median 2.03526 min -2.62085 range 20.3645 sd 3.67347 skewness 1.10857 var 13.4944 \n<strong>3JHC</strong>\nmax 76.0437 mean 3.68847 median 2.89505 min -18.5821 range 94.6258 sd 3.07091 skewness 1.16998 var 9.43047 \n<strong>3JHH</strong>\nmax 17.4841 mean 4.77102 median 3.71061 min -3.02046 range 20.5046 sd 3.70498 skewness 0.645849 var 13.7269 \n<strong>3JHN</strong>\nmax 10.9712 mean 0.99073 median 0.662068 min -3.17241 range 14.1436 sd 1.31539 skewness 1.64438 var 1.73026 </p>\n\n<p>Simple Stats for scalar coupling contributions grouped by <em>type</em>. A chart of the histograms will be posted to the Github page</p>\n\n<p><strong>1JHC</strong>    dso max 1.70161 mean    0.937555    median  0.95363 min 0.27201 range   1.4296  sd  0.195684    skewness    0.0195439   var 0.0382921\n<strong>1JHN</strong>    dso max 0.417811    mean    0.26821 median  0.253365    min 0.067922    range   0.349889    sd  0.0505772   skewness    0.425214    var 0.00255805\n<strong>2JHC</strong>    dso max 0.591145    mean    -0.0789262  median  -0.0506959  min -1.24942    range   1.84057 sd  0.186024    skewness    -1.82861    var 0.034605\n<strong>2JHH</strong>    dso max -0.722259   mean    -2.2633 median  -2.2634 min -6.85679    range   6.13453 sd  0.513059    skewness    -3.13265    var 0.26323\n<strong>2JHN</strong>    dso max 0.107937    mean    -0.0404772  median  -0.0365436  min -0.437965   range   0.545902    sd  0.0479925   skewness    -0.670977   var 0.00230328\n<strong>3JHC</strong>    dso max 0.728981    mean    -0.147708   median  -0.188094   min -0.927072   range   1.65605 sd  0.265384    skewness    0.0260897   var 0.0704285\n<strong>3JHH</strong>    dso max 1.53109 mean    -0.984241   median  -0.490005   min -4.85343    range   6.38452 sd  1.185   skewness    -0.671356   var 1.40423\n<strong>3JHN</strong>    dso max 0.15681 mean    -0.060585   median  -0.0821114  min -0.271433   range   0.428243    sd  0.082979    skewness    0.146923    var 0.00688551\n<strong>1JHC</strong>    fc  max 205.517 mean    93.4526 median  86.6223 min 64.7082 range   140.809 sd  18.4211 skewness    2.73049 var 339.336\n<strong>1JHN</strong>    fc  max 79.0258 mean    45.7    median  45.8877 min 23.6274 range   55.3984 sd  10.8646 skewness    0.105357    var 118.039\n<strong>2JHC</strong>    fc  max 42.9676 mean    -0.306388   median  -1.00951    min -37.224 range   80.1916 sd  4.44987 skewness    2.86531 var 19.8014\n<strong>2JHH</strong>    fc  max 12.058  mean    -10.4573    median  -11.5036    min -35.3676    range   47.4256 sd  4.05017 skewness    1.63082 var 16.4039\n<strong>2JHN</strong>    fc  max 18.4756 mean    3.16404 median  1.95665 min -2.13019    range   20.6058 sd  3.83202 skewness    1.18234 var 14.6844\n<strong>3JHC</strong>    fc  max 76.4528 mean    3.76597 median  2.95253 min -18.1236    range   94.5764 sd  3.1088  skewness    1.16408 var 9.66467\n<strong>3JHH</strong>    fc  max 18.1148 mean    4.92025 median  3.69917 min -3.21961    range   21.3344 sd  3.86353 skewness    0.803262    var 14.9268\n<strong>3JHN</strong>    fc  max 11.0317 mean    1.05408 median  0.681106    min -3.08096    range   14.1127 sd  1.31317 skewness    1.71729 var 1.72441\n<strong>1JHC</strong>    pso max 1.3303  mean    0.387471    median  0.4927  min -2.38873    range   3.71903 sd  0.460782    skewness    -0.819143   var 0.21232\n<strong>1JHN</strong>    pso max 2.15194 mean    1.4268  median  1.45848 min -0.331303   range   2.48324 sd  0.237667    skewness    -0.971366   var 0.0564855\n<strong>2JHC</strong>    pso max 5.3263  mean    0.0394564   median  0.0433233   min -3.28811    range   8.61441 sd  0.553037    skewness    5.77545 var 0.30585\n<strong>2JHH</strong>    pso max 8.20153 mean    2.06026 median  2.04252 min 0.663344    range   7.53819 sd  0.472996    skewness    4.30584 var 0.223725\n<strong>2JHN</strong>    pso max 2.34189 mean    -0.00310926 median  0.0891185   min -1.22959    range   3.57148 sd  0.245205    skewness    -1.78983    var 0.0601254\n<strong>3JHC</strong>    pso max 1.04693 mean    0.0855547   median  0.0975408   min -1.09216    range   2.13909 sd  0.193157    skewness    -0.109494   var 0.0373095\n<strong>3JHH</strong>    pso max 3.69055 mean    0.773651    median  0.433427    min -1.02461    range   4.71516 sd  0.827177    skewness    0.744086    var 0.684223\n<strong>3JHN</strong>    pso max 0.555222    mean    0.00705348  median  0.0137612   min -0.654522   range   1.20974 sd  0.103286    skewness    -0.800263   var 0.0106681\n<strong>1JHC</strong>    sd  max 2.11149 mean    0.198496    median  0.169054    min -0.808604   range   2.92009 sd  0.127312    skewness    2.18674 var 0.0162082\n<strong>1JHN</strong>    sd  max 0.226941    mean    0.0848519   median  0.0808529   min -0.164922   range   0.391863    sd  0.0558663   skewness    -0.179632   var 0.00312104\n<strong>2JHC</strong>    sd  max 2.29669 mean    0.0752328   median  0.0759824   min -3.18491    range   5.4816  sd  0.109143    skewness    3.13352 var 0.0119121\n<strong>2JHH</strong>    sd  max 0.775729    mean    0.373714    median  0.369694    min -0.215389   range   0.991118    sd  0.0376742   skewness    -0.0199843  var 0.00141934\n<strong>2JHN</strong>    sd  max 0.537983    mean    0.00430365  median  0.00584075  min -0.746569   range   1.28455 sd  0.0356897   skewness    0.0726665   var 0.00127376\n<strong>3JHC</strong>    sd  max 1.16598 mean    -0.0153426  median  -0.0225856  min -1.4553 range   2.62128 sd  0.0417047   skewness    0.514652    var 0.00173928\n<strong>3JHH</strong>    sd  max 1.61011 mean    0.0613592   median  0.0629828   min -1.08987    range   2.69998 sd  0.0634342   skewness    -0.343884   var 0.00402389\n<strong>3JHN</strong>    sd  max 0.169883    mean    -0.00981557 median  -0.00855217 min -0.68493    range   0.854813    sd  0.0245621   skewness    -0.928331   var 0.000603296</p>",
      "rawMarkdown": "After working with Python for some time I have decided to try data science/machine learning with C++. I found python to be very slow and take up way to much memory. For this competition Python would probably work just fine, but I am building out my libraries for C++.\n\nThe code, charts, and write up will live [here](https://github.com/rilesdg3/Kaggle-CHAMPS).  The write up live here as well.\n\nI will update this page and the Github pages as I progress.\n\n**Update 07 25 2019** I put new code up on GitHub It is not the greatest but will build a neural network using Caffe2 api\n\nAll scalar coupling\nmax 204.88 mean 15.9216 median 2.27058 min -36.2186 range 241.099 sd 34.942 skewness 2.02425 var 1220.94 \n\nSimple statistics for scalar coupling grouped by *type*.\n**1JHC**\nmax 204.88 mean 94.9762 median 88.0455 min 66.6008 range 138.279 sd 18.2772 skewness 2.72468 var 334.057 \n**1JHN**\nmax 80.4187 mean 47.4799 median 47.8237 min 24.3222 range 56.0965 sd 10.9222 skewness 0.0724719 var 119.294 \n**2JHC**\nmax 42.8192 mean -0.270624 median -0.941388 min -36.2186 range 79.0378 sd 4.52361 skewness 2.91453 var 20.4631 \n**2JHH**\nmax 11.8542 mean -10.2866 median -11.344 min -35.1761 range 47.0303 sd 3.97961 skewness 1.62319 var 15.8373 \n**2JHN**\nmax 17.7436 mean 3.12475 median 2.03526 min -2.62085 range 20.3645 sd 3.67347 skewness 1.10857 var 13.4944 \n**3JHC**\nmax 76.0437 mean 3.68847 median 2.89505 min -18.5821 range 94.6258 sd 3.07091 skewness 1.16998 var 9.43047 \n**3JHH**\nmax 17.4841 mean 4.77102 median 3.71061 min -3.02046 range 20.5046 sd 3.70498 skewness 0.645849 var 13.7269 \n**3JHN**\nmax 10.9712 mean 0.99073 median 0.662068 min -3.17241 range 14.1436 sd 1.31539 skewness 1.64438 var 1.73026 \n\n\n\n\n\nSimple Stats for scalar coupling contributions grouped by *type*. A chart of the histograms will be posted to the Github page\n\n**1JHC**\tdso\tmax\t1.70161\tmean\t0.937555\tmedian\t0.95363\tmin\t0.27201\trange\t1.4296\tsd\t0.195684\tskewness\t0.0195439\tvar\t0.0382921\n**1JHN**\tdso\tmax\t0.417811\tmean\t0.26821\tmedian\t0.253365\tmin\t0.067922\trange\t0.349889\tsd\t0.0505772\tskewness\t0.425214\tvar\t0.00255805\n**2JHC**\tdso\tmax\t0.591145\tmean\t-0.0789262\tmedian\t-0.0506959\tmin\t-1.24942\trange\t1.84057\tsd\t0.186024\tskewness\t-1.82861\tvar\t0.034605\n**2JHH**\tdso\tmax\t-0.722259\tmean\t-2.2633\tmedian\t-2.2634\tmin\t-6.85679\trange\t6.13453\tsd\t0.513059\tskewness\t-3.13265\tvar\t0.26323\n**2JHN**\tdso\tmax\t0.107937\tmean\t-0.0404772\tmedian\t-0.0365436\tmin\t-0.437965\trange\t0.545902\tsd\t0.0479925\tskewness\t-0.670977\tvar\t0.00230328\n**3JHC**\tdso\tmax\t0.728981\tmean\t-0.147708\tmedian\t-0.188094\tmin\t-0.927072\trange\t1.65605\tsd\t0.265384\tskewness\t0.0260897\tvar\t0.0704285\n**3JHH**\tdso\tmax\t1.53109\tmean\t-0.984241\tmedian\t-0.490005\tmin\t-4.85343\trange\t6.38452\tsd\t1.185\tskewness\t-0.671356\tvar\t1.40423\n**3JHN**\tdso\tmax\t0.15681\tmean\t-0.060585\tmedian\t-0.0821114\tmin\t-0.271433\trange\t0.428243\tsd\t0.082979\tskewness\t0.146923\tvar\t0.00688551\n**1JHC**\tfc\tmax\t205.517\tmean\t93.4526\tmedian\t86.6223\tmin\t64.7082\trange\t140.809\tsd\t18.4211\tskewness\t2.73049\tvar\t339.336\n**1JHN**\tfc\tmax\t79.0258\tmean\t45.7\tmedian\t45.8877\tmin\t23.6274\trange\t55.3984\tsd\t10.8646\tskewness\t0.105357\tvar\t118.039\n**2JHC**\tfc\tmax\t42.9676\tmean\t-0.306388\tmedian\t-1.00951\tmin\t-37.224\trange\t80.1916\tsd\t4.44987\tskewness\t2.86531\tvar\t19.8014\n**2JHH**\tfc\tmax\t12.058\tmean\t-10.4573\tmedian\t-11.5036\tmin\t-35.3676\trange\t47.4256\tsd\t4.05017\tskewness\t1.63082\tvar\t16.4039\n**2JHN**\tfc\tmax\t18.4756\tmean\t3.16404\tmedian\t1.95665\tmin\t-2.13019\trange\t20.6058\tsd\t3.83202\tskewness\t1.18234\tvar\t14.6844\n**3JHC**\tfc\tmax\t76.4528\tmean\t3.76597\tmedian\t2.95253\tmin\t-18.1236\trange\t94.5764\tsd\t3.1088\tskewness\t1.16408\tvar\t9.66467\n**3JHH**\tfc\tmax\t18.1148\tmean\t4.92025\tmedian\t3.69917\tmin\t-3.21961\trange\t21.3344\tsd\t3.86353\tskewness\t0.803262\tvar\t14.9268\n**3JHN**\tfc\tmax\t11.0317\tmean\t1.05408\tmedian\t0.681106\tmin\t-3.08096\trange\t14.1127\tsd\t1.31317\tskewness\t1.71729\tvar\t1.72441\n**1JHC**\tpso\tmax\t1.3303\tmean\t0.387471\tmedian\t0.4927\tmin\t-2.38873\trange\t3.71903\tsd\t0.460782\tskewness\t-0.819143\tvar\t0.21232\n**1JHN**\tpso\tmax\t2.15194\tmean\t1.4268\tmedian\t1.45848\tmin\t-0.331303\trange\t2.48324\tsd\t0.237667\tskewness\t-0.971366\tvar\t0.0564855\n**2JHC**\tpso\tmax\t5.3263\tmean\t0.0394564\tmedian\t0.0433233\tmin\t-3.28811\trange\t8.61441\tsd\t0.553037\tskewness\t5.77545\tvar\t0.30585\n**2JHH**\tpso\tmax\t8.20153\tmean\t2.06026\tmedian\t2.04252\tmin\t0.663344\trange\t7.53819\tsd\t0.472996\tskewness\t4.30584\tvar\t0.223725\n**2JHN**\tpso\tmax\t2.34189\tmean\t-0.00310926\tmedian\t0.0891185\tmin\t-1.22959\trange\t3.57148\tsd\t0.245205\tskewness\t-1.78983\tvar\t0.0601254\n**3JHC**\tpso\tmax\t1.04693\tmean\t0.0855547\tmedian\t0.0975408\tmin\t-1.09216\trange\t2.13909\tsd\t0.193157\tskewness\t-0.109494\tvar\t0.0373095\n**3JHH**\tpso\tmax\t3.69055\tmean\t0.773651\tmedian\t0.433427\tmin\t-1.02461\trange\t4.71516\tsd\t0.827177\tskewness\t0.744086\tvar\t0.684223\n**3JHN**\tpso\tmax\t0.555222\tmean\t0.00705348\tmedian\t0.0137612\tmin\t-0.654522\trange\t1.20974\tsd\t0.103286\tskewness\t-0.800263\tvar\t0.0106681\n**1JHC**\tsd\tmax\t2.11149\tmean\t0.198496\tmedian\t0.169054\tmin\t-0.808604\trange\t2.92009\tsd\t0.127312\tskewness\t2.18674\tvar\t0.0162082\n**1JHN**\tsd\tmax\t0.226941\tmean\t0.0848519\tmedian\t0.0808529\tmin\t-0.164922\trange\t0.391863\tsd\t0.0558663\tskewness\t-0.179632\tvar\t0.00312104\n**2JHC**\tsd\tmax\t2.29669\tmean\t0.0752328\tmedian\t0.0759824\tmin\t-3.18491\trange\t5.4816\tsd\t0.109143\tskewness\t3.13352\tvar\t0.0119121\n**2JHH**\tsd\tmax\t0.775729\tmean\t0.373714\tmedian\t0.369694\tmin\t-0.215389\trange\t0.991118\tsd\t0.0376742\tskewness\t-0.0199843\tvar\t0.00141934\n**2JHN**\tsd\tmax\t0.537983\tmean\t0.00430365\tmedian\t0.00584075\tmin\t-0.746569\trange\t1.28455\tsd\t0.0356897\tskewness\t0.0726665\tvar\t0.00127376\n**3JHC**\tsd\tmax\t1.16598\tmean\t-0.0153426\tmedian\t-0.0225856\tmin\t-1.4553\trange\t2.62128\tsd\t0.0417047\tskewness\t0.514652\tvar\t0.00173928\n**3JHH**\tsd\tmax\t1.61011\tmean\t0.0613592\tmedian\t0.0629828\tmin\t-1.08987\trange\t2.69998\tsd\t0.0634342\tskewness\t-0.343884\tvar\t0.00402389\n**3JHN**\tsd\tmax\t0.169883\tmean\t-0.00981557\tmedian\t-0.00855217\tmin\t-0.68493\trange\t0.854813\tsd\t0.0245621\tskewness\t-0.928331\tvar\t0.000603296\n",
      "votes": 7
    },
    {
      "id": 560462,
      "postDate": "2019-06-25T13:11:21.293Z",
      "content": "<p>fyi and if I am not mistaken:</p>\n\n<p>Michael Jahrer won the Porto Seguro’s Safe Driver Prediction with a neural network written in C++ of his own.</p>\n\n<p>The winners, more specifically, Medrr, in the Santander Customer Satisfaction had part of their solution written in customized C.</p>",
      "rawMarkdown": "fyi and if I am not mistaken:\n\nMichael Jahrer won the Porto Seguro’s Safe Driver Prediction with a neural network written in C++ of his own.\n\nThe winners, more specifically, Medrr, in the Santander Customer Satisfaction had part of their solution written in customized C.",
      "votes": 2,
      "replies": [
        {
          "id": 561283,
          "postDate": "2019-06-26T09:53:32.810Z",
          "content": "<p>Right, our preprocessing here + some modelling is end-to-end in C++ too</p>",
          "rawMarkdown": "Right, our preprocessing here + some modelling is end-to-end in C++ too",
          "votes": 3
        },
        {
          "id": 571195,
          "postDate": "2019-07-09T09:36:54.750Z",
          "content": "<p>Michael, thanks for confirming you are using your NN C++ code here ;)</p>",
          "rawMarkdown": "Michael, thanks for confirming you are using your NN C++ code here ;)",
          "votes": 1
        },
        {
          "id": 584802,
          "postDate": "2019-07-26T13:59:03.967Z",
          "content": "<p>He also won Home Credit where his C++ code was a crucial component.  ;)</p>",
          "rawMarkdown": "He also won Home Credit where his C++ code was a crucial component.  ;)",
          "votes": 1
        }
      ]
    },
    {
      "id": 559587,
      "postDate": "2019-06-24T10:03:37.240Z",
      "content": "<p>Very interesting! How faster is C++ comapred to Python? I recently had a conversation with colleagues: we were wondering why C++ is not used more often.</p>",
      "rawMarkdown": "Very interesting! How faster is C++ comapred to Python? I recently had a conversation with colleagues: we were wondering why C++ is not used more often.",
      "votes": 1,
      "replies": [
        {
          "id": 559674,
          "postDate": "2019-06-24T12:08:16.060Z",
          "content": "<p>With they way I coded I used Python3.4 with Tensorflow and switched to C++ and Caffe2 and what use to take a day to run will now run in an hour or so and use like 1-10% of the memory. In short I can now do things that I was unable to do using python, like building a larger network. There are things you can do in python to use less memory but it seems like when you did that the code became just as verbose as C++;</p>\n\n<p>I think C++ is not used more often because it is not as easy to get started with and there is a lack of easy to use libraries for doing data science in C++. If you need something in python you just pip install the library and you are all set to go.  Also C++ requires you to declare the variable type like int, string, float…., however this is no longer the case with the use of auto but I know a lot of C++ programmers who get tired of declaring variables. With C++17 and C++20 I think it is getting easier to use. </p>\n\n<p>As I develop my code base and plan my project a little better C++ is getting easier and easier to use for these types of tasks. So far I am very glad I made the switch.</p>",
          "rawMarkdown": "With they way I coded I used Python3.4 with Tensorflow and switched to C++ and Caffe2 and what use to take a day to run will now run in an hour or so and use like 1-10% of the memory. In short I can now do things that I was unable to do using python, like building a larger network. There are things you can do in python to use less memory but it seems like when you did that the code became just as verbose as C++;\n\nI think C++ is not used more often because it is not as easy to get started with and there is a lack of easy to use libraries for doing data science in C++. If you need something in python you just pip install the library and you are all set to go.  Also C++ requires you to declare the variable type like int, string, float…., however this is no longer the case with the use of auto but I know a lot of C++ programmers who get tired of declaring variables. With C++17 and C++20 I think it is getting easier to use. \n\nAs I develop my code base and plan my project a little better C++ is getting easier and easier to use for these types of tasks. So far I am very glad I made the switch.\n",
          "votes": 6
        },
        {
          "id": 560327,
          "postDate": "2019-06-25T09:56:51.590Z",
          "content": "<p>Thank you for your well detailed answer. </p>",
          "rawMarkdown": "Thank you for your well detailed answer. "
        },
        {
          "id": 584179,
          "postDate": "2019-07-25T14:22:26.657Z",
          "content": "<p><a href=\"/rilesdg3\">@rilesdg3</a> very interesting work!</p>\n\n<p>Can you maybe point me to some examples of doing the same thing in C++(plus Caffe2) vs python (plus TF) that show TF to be slower?  </p>\n\n<p>If this is Caffe2 vs TF, I would expect to see roughly the same speedup in PyTorch. </p>\n\n<p>Do you use Caffe2 directly, or have you tried using the PyTorch C++ API? </p>",
          "rawMarkdown": "@rilesdg3 very interesting work!\n\nCan you maybe point me to some examples of doing the same thing in C++(plus Caffe2) vs python (plus TF) that show TF to be slower?  \n\nIf this is Caffe2 vs TF, I would expect to see roughly the same speedup in PyTorch. \n\nDo you use Caffe2 directly, or have you tried using the PyTorch C++ API? \n"
        },
        {
          "id": 584557,
          "postDate": "2019-07-26T06:44:31.137Z",
          "content": "<p><a href=\"/fnands\">@fnands</a> </p>\n\n<p><strong>Can you maybe point me to some examples of doing the same thing in C++(plus Caffe2) vs python (plus TF) that show TF to be slower?</strong></p>\n\n<p>Unfortunately, I do not have any examples and at the time I made the switch I didn't think to document the differences.  I could not really find anything showing the speed differences between Caffe2 and TF</p>\n\n<p><strong>If this is Caffe2 vs TF, I would expect to see roughly the same speedup in PyTorch.</strong></p>\n\n<p>I don't think this is just Caffe2 vs TF I think it is the combination of Caffe2 and C++. </p>\n\n<p><strong>Do you use Caffe2 directly, or have you tried using the PyTorch C++ API?</strong></p>\n\n<p>I use Caffe2 directly and I have not used the PyTorch C++ api. Facebook is in the process of <a href=\"https://caffe2.ai/blog/2018/05/02/Caffe2_PyTorch_1_0.html\">merging</a> Caffe2 and PyTorch. So in the future they might just be the same.</p>",
          "rawMarkdown": "@fnands \n\n**Can you maybe point me to some examples of doing the same thing in C++(plus Caffe2) vs python (plus TF) that show TF to be slower?**\n\n Unfortunately, I do not have any examples and at the time I made the switch I didn't think to document the differences.  I could not really find anything showing the speed differences between Caffe2 and TF\n\n**If this is Caffe2 vs TF, I would expect to see roughly the same speedup in PyTorch.**\n\nI don't think this is just Caffe2 vs TF I think it is the combination of Caffe2 and C++. \n\n**Do you use Caffe2 directly, or have you tried using the PyTorch C++ API?**\n\nI use Caffe2 directly and I have not used the PyTorch C++ api. Facebook is in the process of [merging](https://caffe2.ai/blog/2018/05/02/Caffe2_PyTorch_1_0.html) Caffe2 and PyTorch. So in the future they might just be the same."
        }
      ]
    },
    {
      "id": 561185,
      "postDate": "2019-06-26T08:05:04.927Z",
      "content": "<p>Thanks for sharing. Using opencv and cvplot for plotting looks very interesting. Most of the time I use matplotlib via the python/C interface, but I have to give cvplot a try.</p>\n\n<p>I am a big fan of C++ for machine learning myself. Specially for loop based code you have a big speed advantage over python.</p>",
      "rawMarkdown": "Thanks for sharing. Using opencv and cvplot for plotting looks very interesting. Most of the time I use matplotlib via the python/C interface, but I have to give cvplot a try.\n\nI am a big fan of C++ for machine learning myself. Specially for loop based code you have a big speed advantage over python.",
      "votes": 1,
      "replies": [
        {
          "id": 563284,
          "postDate": "2019-06-28T05:07:29.780Z",
          "content": "<p>This is what I used to get started with cvplot in c++ <a href=\"https://github.com/leonardvandriel/cvplot\">https://github.com/leonardvandriel/cvplot</a></p>\n\n<p>A big part of the reason I went to c++ was because I felt I was doing a lot of looping that I could not use some python function that was optimized in c or c++.</p>",
          "rawMarkdown": "This is what I used to get started with cvplot in c++ https://github.com/leonardvandriel/cvplot\n\nA big part of the reason I went to c++ was because I felt I was doing a lot of looping that I could not use some python function that was optimized in c or c++.",
          "votes": 1
        }
      ]
    },
    {
      "id": 583947,
      "postDate": "2019-07-25T08:27:14.747Z",
      "content": "<p>I put new code up on <a href=\"https://github.com/rilesdg3/Kaggle-CHAMPS\">GitHub</a> It is not the greatest but will build a neural network using Caffe2 api</p>",
      "rawMarkdown": "I put new code up on [GitHub](https://github.com/rilesdg3/Kaggle-CHAMPS) It is not the greatest but will build a neural network using Caffe2 api"
    },
    {
      "id": 560287,
      "postDate": "2019-06-25T09:00:54.507Z",
      "content": "<p>Julia promises better speed than Python with the same readability - have you considered using that?</p>",
      "rawMarkdown": "Julia promises better speed than Python with the same readability - have you considered using that?",
      "replies": [
        {
          "id": 560402,
          "postDate": "2019-06-25T11:37:06.477Z",
          "content": "<p>I have never looked at Julia, I will have to check it out</p>",
          "rawMarkdown": "I have never looked at Julia, I will have to check it out"
        },
        {
          "id": 563342,
          "postDate": "2019-06-28T07:05:31.997Z",
          "content": "<p>Cython and Numba can be helpful in speeding up python, also Julia looks really interesting!</p>",
          "rawMarkdown": "Cython and Numba can be helpful in speeding up python, also Julia looks really interesting!"
        }
      ]
    },
    {
      "id": 559640,
      "postDate": "2019-06-24T11:10:46.783Z",
      "rawMarkdown": "",
      "votes": -5,
      "isDeleted": true
    },
    {
      "id": 564986,
      "postDate": "2019-06-30T09:37:01.017Z",
      "content": "<p>Thanks for sharing </p>",
      "rawMarkdown": "Thanks for sharing "
    }
  ],
  "comments": [
    {
      "id": 560462,
      "author_name": "eagle4",
      "author_url": "",
      "post_date": "2019-06-25T13:11:21.293000",
      "content": "<p>fyi and if I am not mistaken:</p>\n\n<p>Michael Jahrer won the Porto Seguro’s Safe Driver Prediction with a neural network written in C++ of his own.</p>\n\n<p>The winners, more specifically, Medrr, in the Santander Customer Satisfaction had part of their solution written in customized C.</p>",
      "votes": 2,
      "replies": [
        {
          "id": 561283,
          "author_name": "Michael Jahrer",
          "author_url": "",
          "post_date": "2019-06-26T09:53:32.810000",
          "content": "<p>Right, our preprocessing here + some modelling is end-to-end in C++ too</p>",
          "votes": 3,
          "replies": []
        },
        {
          "id": 571195,
          "author_name": "CPMP",
          "author_url": "",
          "post_date": "2019-07-09T09:36:54.750000",
          "content": "<p>Michael, thanks for confirming you are using your NN C++ code here ;)</p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 584802,
          "author_name": "Bojan Tunguz",
          "author_url": "",
          "post_date": "2019-07-26T13:59:03.967000",
          "content": "<p>He also won Home Credit where his C++ code was a crucial component.  ;)</p>",
          "votes": 1,
          "replies": []
        }
      ]
    },
    {
      "id": 559587,
      "author_name": "Jean-Philippe MARY",
      "author_url": "",
      "post_date": "2019-06-24T10:03:37.240000",
      "content": "<p>Very interesting! How faster is C++ comapred to Python? I recently had a conversation with colleagues: we were wondering why C++ is not used more often.</p>",
      "votes": 1,
      "replies": [
        {
          "id": 559674,
          "author_name": "rilesdg3",
          "author_url": "",
          "post_date": "2019-06-24T12:08:16.060000",
          "content": "<p>With they way I coded I used Python3.4 with Tensorflow and switched to C++ and Caffe2 and what use to take a day to run will now run in an hour or so and use like 1-10% of the memory. In short I can now do things that I was unable to do using python, like building a larger network. There are things you can do in python to use less memory but it seems like when you did that the code became just as verbose as C++;</p>\n\n<p>I think C++ is not used more often because it is not as easy to get started with and there is a lack of easy to use libraries for doing data science in C++. If you need something in python you just pip install the library and you are all set to go.  Also C++ requires you to declare the variable type like int, string, float…., however this is no longer the case with the use of auto but I know a lot of C++ programmers who get tired of declaring variables. With C++17 and C++20 I think it is getting easier to use. </p>\n\n<p>As I develop my code base and plan my project a little better C++ is getting easier and easier to use for these types of tasks. So far I am very glad I made the switch.</p>",
          "votes": 6,
          "replies": []
        },
        {
          "id": 560327,
          "author_name": "Jean-Philippe MARY",
          "author_url": "",
          "post_date": "2019-06-25T09:56:51.590000",
          "content": "<p>Thank you for your well detailed answer. </p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 584179,
          "author_name": "fnands",
          "author_url": "",
          "post_date": "2019-07-25T14:22:26.657000",
          "content": "<p><a href=\"/rilesdg3\">@rilesdg3</a> very interesting work!</p>\n\n<p>Can you maybe point me to some examples of doing the same thing in C++(plus Caffe2) vs python (plus TF) that show TF to be slower?  </p>\n\n<p>If this is Caffe2 vs TF, I would expect to see roughly the same speedup in PyTorch. </p>\n\n<p>Do you use Caffe2 directly, or have you tried using the PyTorch C++ API? </p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 584557,
          "author_name": "rilesdg3",
          "author_url": "",
          "post_date": "2019-07-26T06:44:31.137000",
          "content": "<p><a href=\"/fnands\">@fnands</a> </p>\n\n<p><strong>Can you maybe point me to some examples of doing the same thing in C++(plus Caffe2) vs python (plus TF) that show TF to be slower?</strong></p>\n\n<p>Unfortunately, I do not have any examples and at the time I made the switch I didn't think to document the differences.  I could not really find anything showing the speed differences between Caffe2 and TF</p>\n\n<p><strong>If this is Caffe2 vs TF, I would expect to see roughly the same speedup in PyTorch.</strong></p>\n\n<p>I don't think this is just Caffe2 vs TF I think it is the combination of Caffe2 and C++. </p>\n\n<p><strong>Do you use Caffe2 directly, or have you tried using the PyTorch C++ API?</strong></p>\n\n<p>I use Caffe2 directly and I have not used the PyTorch C++ api. Facebook is in the process of <a href=\"https://caffe2.ai/blog/2018/05/02/Caffe2_PyTorch_1_0.html\">merging</a> Caffe2 and PyTorch. So in the future they might just be the same.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 561185,
      "author_name": "AndreasToescher",
      "author_url": "",
      "post_date": "2019-06-26T08:05:04.927000",
      "content": "<p>Thanks for sharing. Using opencv and cvplot for plotting looks very interesting. Most of the time I use matplotlib via the python/C interface, but I have to give cvplot a try.</p>\n\n<p>I am a big fan of C++ for machine learning myself. Specially for loop based code you have a big speed advantage over python.</p>",
      "votes": 1,
      "replies": [
        {
          "id": 563284,
          "author_name": "rilesdg3",
          "author_url": "",
          "post_date": "2019-06-28T05:07:29.780000",
          "content": "<p>This is what I used to get started with cvplot in c++ <a href=\"https://github.com/leonardvandriel/cvplot\">https://github.com/leonardvandriel/cvplot</a></p>\n\n<p>A big part of the reason I went to c++ was because I felt I was doing a lot of looping that I could not use some python function that was optimized in c or c++.</p>",
          "votes": 1,
          "replies": []
        }
      ]
    },
    {
      "id": 583947,
      "author_name": "rilesdg3",
      "author_url": "",
      "post_date": "2019-07-25T08:27:14.747000",
      "content": "<p>I put new code up on <a href=\"https://github.com/rilesdg3/Kaggle-CHAMPS\">GitHub</a> It is not the greatest but will build a neural network using Caffe2 api</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 560287,
      "author_name": "BradSShaw",
      "author_url": "",
      "post_date": "2019-06-25T09:00:54.507000",
      "content": "<p>Julia promises better speed than Python with the same readability - have you considered using that?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 560402,
          "author_name": "rilesdg3",
          "author_url": "",
          "post_date": "2019-06-25T11:37:06.477000",
          "content": "<p>I have never looked at Julia, I will have to check it out</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 563342,
          "author_name": "Winter",
          "author_url": "",
          "post_date": "2019-06-28T07:05:31.997000",
          "content": "<p>Cython and Numba can be helpful in speeding up python, also Julia looks really interesting!</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 559640,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-06-24T11:10:46.783000",
      "content": "",
      "votes": -5,
      "replies": []
    },
    {
      "id": 564986,
      "author_name": "Madhur Sabherwal",
      "author_url": "",
      "post_date": "2019-06-30T09:37:01.017000",
      "content": "<p>Thanks for sharing </p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "559549": "After working with Python for some time I have decided to try data science/machine learning with C++. I found python to be very slow and take up way to much memory. For this competition Python would probably work just fine, but I am building out my libraries for C++.\n\nThe code, charts, and write up will live [here](https://github.com/rilesdg3/Kaggle-CHAMPS).  The write up live here as well.\n\nI will update this page and the Github pages as I progress.\n\n**Update 07 25 2019** I put new code up on GitHub It is not the greatest but will build a neural network using Caffe2 api\n\nAll scalar coupling\nmax 204.88 mean 15.9216 median 2.27058 min -36.2186 range 241.099 sd 34.942 skewness 2.02425 var 1220.94 \n\nSimple statistics for scalar coupling grouped by *type*.\n**1JHC**\nmax 204.88 mean 94.9762 median 88.0455 min 66.6008 range 138.279 sd 18.2772 skewness 2.72468 var 334.057 \n**1JHN**\nmax 80.4187 mean 47.4799 median 47.8237 min 24.3222 range 56.0965 sd 10.9222 skewness 0.0724719 var 119.294 \n**2JHC**\nmax 42.8192 mean -0.270624 median -0.941388 min -36.2186 range 79.0378 sd 4.52361 skewness 2.91453 var 20.4631 \n**2JHH**\nmax 11.8542 mean -10.2866 median -11.344 min -35.1761 range 47.0303 sd 3.97961 skewness 1.62319 var 15.8373 \n**2JHN**\nmax 17.7436 mean 3.12475 median 2.03526 min -2.62085 range 20.3645 sd 3.67347 skewness 1.10857 var 13.4944 \n**3JHC**\nmax 76.0437 mean 3.68847 median 2.89505 min -18.5821 range 94.6258 sd 3.07091 skewness 1.16998 var 9.43047 \n**3JHH**\nmax 17.4841 mean 4.77102 median 3.71061 min -3.02046 range 20.5046 sd 3.70498 skewness 0.645849 var 13.7269 \n**3JHN**\nmax 10.9712 mean 0.99073 median 0.662068 min -3.17241 range 14.1436 sd 1.31539 skewness 1.64438 var 1.73026 \n\n\n\n\n\nSimple Stats for scalar coupling contributions grouped by *type*. A chart of the histograms will be posted to the Github page\n\n**1JHC**\tdso\tmax\t1.70161\tmean\t0.937555\tmedian\t0.95363\tmin\t0.27201\trange\t1.4296\tsd\t0.195684\tskewness\t0.0195439\tvar\t0.0382921\n**1JHN**\tdso\tmax\t0.417811\tmean\t0.26821\tmedian\t0.253365\tmin\t0.067922\trange\t0.349889\tsd\t0.0505772\tskewness\t0.425214\tvar\t0.00255805\n**2JHC**\tdso\tmax\t0.591145\tmean\t-0.0789262\tmedian\t-0.0506959\tmin\t-1.24942\trange\t1.84057\tsd\t0.186024\tskewness\t-1.82861\tvar\t0.034605\n**2JHH**\tdso\tmax\t-0.722259\tmean\t-2.2633\tmedian\t-2.2634\tmin\t-6.85679\trange\t6.13453\tsd\t0.513059\tskewness\t-3.13265\tvar\t0.26323\n**2JHN**\tdso\tmax\t0.107937\tmean\t-0.0404772\tmedian\t-0.0365436\tmin\t-0.437965\trange\t0.545902\tsd\t0.0479925\tskewness\t-0.670977\tvar\t0.00230328\n**3JHC**\tdso\tmax\t0.728981\tmean\t-0.147708\tmedian\t-0.188094\tmin\t-0.927072\trange\t1.65605\tsd\t0.265384\tskewness\t0.0260897\tvar\t0.0704285\n**3JHH**\tdso\tmax\t1.53109\tmean\t-0.984241\tmedian\t-0.490005\tmin\t-4.85343\trange\t6.38452\tsd\t1.185\tskewness\t-0.671356\tvar\t1.40423\n**3JHN**\tdso\tmax\t0.15681\tmean\t-0.060585\tmedian\t-0.0821114\tmin\t-0.271433\trange\t0.428243\tsd\t0.082979\tskewness\t0.146923\tvar\t0.00688551\n**1JHC**\tfc\tmax\t205.517\tmean\t93.4526\tmedian\t86.6223\tmin\t64.7082\trange\t140.809\tsd\t18.4211\tskewness\t2.73049\tvar\t339.336\n**1JHN**\tfc\tmax\t79.0258\tmean\t45.7\tmedian\t45.8877\tmin\t23.6274\trange\t55.3984\tsd\t10.8646\tskewness\t0.105357\tvar\t118.039\n**2JHC**\tfc\tmax\t42.9676\tmean\t-0.306388\tmedian\t-1.00951\tmin\t-37.224\trange\t80.1916\tsd\t4.44987\tskewness\t2.86531\tvar\t19.8014\n**2JHH**\tfc\tmax\t12.058\tmean\t-10.4573\tmedian\t-11.5036\tmin\t-35.3676\trange\t47.4256\tsd\t4.05017\tskewness\t1.63082\tvar\t16.4039\n**2JHN**\tfc\tmax\t18.4756\tmean\t3.16404\tmedian\t1.95665\tmin\t-2.13019\trange\t20.6058\tsd\t3.83202\tskewness\t1.18234\tvar\t14.6844\n**3JHC**\tfc\tmax\t76.4528\tmean\t3.76597\tmedian\t2.95253\tmin\t-18.1236\trange\t94.5764\tsd\t3.1088\tskewness\t1.16408\tvar\t9.66467\n**3JHH**\tfc\tmax\t18.1148\tmean\t4.92025\tmedian\t3.69917\tmin\t-3.21961\trange\t21.3344\tsd\t3.86353\tskewness\t0.803262\tvar\t14.9268\n**3JHN**\tfc\tmax\t11.0317\tmean\t1.05408\tmedian\t0.681106\tmin\t-3.08096\trange\t14.1127\tsd\t1.31317\tskewness\t1.71729\tvar\t1.72441\n**1JHC**\tpso\tmax\t1.3303\tmean\t0.387471\tmedian\t0.4927\tmin\t-2.38873\trange\t3.71903\tsd\t0.460782\tskewness\t-0.819143\tvar\t0.21232\n**1JHN**\tpso\tmax\t2.15194\tmean\t1.4268\tmedian\t1.45848\tmin\t-0.331303\trange\t2.48324\tsd\t0.237667\tskewness\t-0.971366\tvar\t0.0564855\n**2JHC**\tpso\tmax\t5.3263\tmean\t0.0394564\tmedian\t0.0433233\tmin\t-3.28811\trange\t8.61441\tsd\t0.553037\tskewness\t5.77545\tvar\t0.30585\n**2JHH**\tpso\tmax\t8.20153\tmean\t2.06026\tmedian\t2.04252\tmin\t0.663344\trange\t7.53819\tsd\t0.472996\tskewness\t4.30584\tvar\t0.223725\n**2JHN**\tpso\tmax\t2.34189\tmean\t-0.00310926\tmedian\t0.0891185\tmin\t-1.22959\trange\t3.57148\tsd\t0.245205\tskewness\t-1.78983\tvar\t0.0601254\n**3JHC**\tpso\tmax\t1.04693\tmean\t0.0855547\tmedian\t0.0975408\tmin\t-1.09216\trange\t2.13909\tsd\t0.193157\tskewness\t-0.109494\tvar\t0.0373095\n**3JHH**\tpso\tmax\t3.69055\tmean\t0.773651\tmedian\t0.433427\tmin\t-1.02461\trange\t4.71516\tsd\t0.827177\tskewness\t0.744086\tvar\t0.684223\n**3JHN**\tpso\tmax\t0.555222\tmean\t0.00705348\tmedian\t0.0137612\tmin\t-0.654522\trange\t1.20974\tsd\t0.103286\tskewness\t-0.800263\tvar\t0.0106681\n**1JHC**\tsd\tmax\t2.11149\tmean\t0.198496\tmedian\t0.169054\tmin\t-0.808604\trange\t2.92009\tsd\t0.127312\tskewness\t2.18674\tvar\t0.0162082\n**1JHN**\tsd\tmax\t0.226941\tmean\t0.0848519\tmedian\t0.0808529\tmin\t-0.164922\trange\t0.391863\tsd\t0.0558663\tskewness\t-0.179632\tvar\t0.00312104\n**2JHC**\tsd\tmax\t2.29669\tmean\t0.0752328\tmedian\t0.0759824\tmin\t-3.18491\trange\t5.4816\tsd\t0.109143\tskewness\t3.13352\tvar\t0.0119121\n**2JHH**\tsd\tmax\t0.775729\tmean\t0.373714\tmedian\t0.369694\tmin\t-0.215389\trange\t0.991118\tsd\t0.0376742\tskewness\t-0.0199843\tvar\t0.00141934\n**2JHN**\tsd\tmax\t0.537983\tmean\t0.00430365\tmedian\t0.00584075\tmin\t-0.746569\trange\t1.28455\tsd\t0.0356897\tskewness\t0.0726665\tvar\t0.00127376\n**3JHC**\tsd\tmax\t1.16598\tmean\t-0.0153426\tmedian\t-0.0225856\tmin\t-1.4553\trange\t2.62128\tsd\t0.0417047\tskewness\t0.514652\tvar\t0.00173928\n**3JHH**\tsd\tmax\t1.61011\tmean\t0.0613592\tmedian\t0.0629828\tmin\t-1.08987\trange\t2.69998\tsd\t0.0634342\tskewness\t-0.343884\tvar\t0.00402389\n**3JHN**\tsd\tmax\t0.169883\tmean\t-0.00981557\tmedian\t-0.00855217\tmin\t-0.68493\trange\t0.854813\tsd\t0.0245621\tskewness\t-0.928331\tvar\t0.000603296\n",
    "560462": "fyi and if I am not mistaken:\n\nMichael Jahrer won the Porto Seguro’s Safe Driver Prediction with a neural network written in C++ of his own.\n\nThe winners, more specifically, Medrr, in the Santander Customer Satisfaction had part of their solution written in customized C.",
    "559587": "Very interesting! How faster is C++ comapred to Python? I recently had a conversation with colleagues: we were wondering why C++ is not used more often.",
    "561185": "Thanks for sharing. Using opencv and cvplot for plotting looks very interesting. Most of the time I use matplotlib via the python/C interface, but I have to give cvplot a try.\n\nI am a big fan of C++ for machine learning myself. Specially for loop based code you have a big speed advantage over python.",
    "583947": "I put new code up on [GitHub](https://github.com/rilesdg3/Kaggle-CHAMPS) It is not the greatest but will build a neural network using Caffe2 api",
    "560287": "Julia promises better speed than Python with the same readability - have you considered using that?",
    "559640": "",
    "564986": "Thanks for sharing "
  }
}