adding tfi competition code

FindingELO/main.py

@@ -2,8 +2,7 @@
 from extras import *
 from features import *
 from sklearn.cross_validation import cross_val_score
-from sklearn.linear_model import LinearRegression
-from sklearn.ensemble import RandomForestRegressor,GradientBoostingRegressor
+from sklearn.linear_model import LogisticRegression
 from sklearn.base import clone
 import warnings
 import sys
@@ -12,7 +11,7 @@ def clf_generator():
     clfs = [
         # RandomForestRegressor(n_estimators=3000,verbose=1,n_jobs=1),
         # GradientBoostingRegressor(n_estimators=5000,verbose=1,learning_rate=0.01),
-        LinearRegression(),
+        LogisticRegression(),
     ]
     for clf in clfs:
         yield clf,clone(clf)
@@ -26,20 +25,20 @@ def main():
     for clf_w,clf_b in clf_generator():
         print list(X.columns.values)
         print clf_w
-        print cross_val_score(estimator=clf_w,
-                              scoring="mean_absolute_error",
-                              cv=4,
-                              X=X,
-                              y=yw,
-                              n_jobs=-1)
-        print cross_val_score(estimator=clf_b,
-                              scoring="mean_absolute_error",
-                              cv=4,
-                              X=X,
-                              y=yb,
-                              n_jobs=-1)
-        print "\n"
-        continue
+        # print cross_val_score(estimator=clf_w,
+        #                       scoring="mean_absolute_error",
+        #                       cv=4,
+        #                       X=X,
+        #                       y=yw,
+        #                       n_jobs=-1)
+        # print cross_val_score(estimator=clf_b,
+        #                       scoring="mean_absolute_error",
+        #                       cv=4,
+        #                       X=X,
+        #                       y=yb,
+        #                       n_jobs=-1)
+        # print "\n"
+        # continue
         print "Training ..."
         clf_w.fit(X,yw)
         clf_b.fit(X,yb)

FindingELO/settings.json

@@ -1,5 +1,5 @@
 {
-    "data_dir" : "../data/FindingELO",
-    "cache_dir" : "../data/FindingELO/cache",
+    "data_dir" : "../../data/FindingELO",
+    "cache_dir" : "../../data/FindingELO/cache",
     "num_partitions" : 3
 }

InriaBCIChallenge/4foldcvlogs.txt

@@ -0,0 +1,59 @@
+Training
+__KeepChannel_[28, 29, 30]__ButterworthFilter_0.1_15_4_200_epoched__EEGConcatExtracter__Downsampler_dec_8
+Fitting 4 folds for each of 1 candidates, totalling 4 fits
+[Parallel(n_jobs=-1)]: Done   1 jobs       | elapsed:   10.7s
+[Parallel(n_jobs=-1)]: Done   4 out of   4 | elapsed:   11.0s finished
+Cross Validation Stats
+Best parameters set:
+GradientBoostingClassifier(init=None, learning_rate=0.05, loss='deviance',
+              max_depth=2, max_features=0.25, max_leaf_nodes=None,
+              min_samples_leaf=1, min_samples_split=2, n_estimators=500,
+              random_state=None, subsample=1.0, verbose=0,
+              warm_start=False)
+Grid scores on development set:
+0.661 (+/-0.025) for {'max_features': 0.25, 'n_estimators': 500, 'learning_rate': 0.05, 'max_depth': 2}
+Training
+____KeepChannel_[28, 29, 30]__ButterworthFilter_0.1_15_4_200_epoched__EEGConcatExtracter__Downsampler_dec_12__KeepChannel_[28, 29, 30]__ButterworthFilter_0.1_15_4_200_epoched__<transforms.ActualWaveletTransform instance at 0x7f6c07a37fc8>__EEGConcatExtracter__Downsampler_dec_12
+(5440, 260, 3)
+Fitting 4 folds for each of 1 candidates, totalling 4 fits
+[Parallel(n_jobs=-1)]: Done   1 jobs       | elapsed:   10.5s
+[Parallel(n_jobs=-1)]: Done   4 out of   4 | elapsed:   10.9s finished
+Cross Validation Stats
+Best parameters set:
+GradientBoostingClassifier(init=None, learning_rate=0.05, loss='deviance',
+              max_depth=2, max_features=0.25, max_leaf_nodes=None,
+              min_samples_leaf=1, min_samples_split=2, n_estimators=500,
+              random_state=None, subsample=1.0, verbose=0,
+              warm_start=False)
+Grid scores on development set:
+0.661 (+/-0.022) for {'max_features': 0.25, 'n_estimators': 500, 'learning_rate': 0.05, 'max_depth': 2}
+Training
+____KeepChannel_[28, 29, 30]__ButterworthFilter_0.1_15_4_200_epoched__EEGConcatExtracter__Downsampler_dec_24__KeepChannel_[28, 29, 30]__ButterworthFilter_0.1_15_4_200_epoched__<transforms.ActualWaveletTransform instance at 0x7f6c07a3e4d0>__EEGConcatExtracter__Downsampler_dec_24
+(5440, 260, 3)
+Fitting 4 folds for each of 1 candidates, totalling 4 fits
+[Parallel(n_jobs=-1)]: Done   1 jobs       | elapsed:    5.6s
+[Parallel(n_jobs=-1)]: Done   4 out of   4 | elapsed:    5.9s finished
+Cross Validation Stats
+Best parameters set:
+GradientBoostingClassifier(init=None, learning_rate=0.05, loss='deviance',
+              max_depth=2, max_features=0.25, max_leaf_nodes=None,
+              min_samples_leaf=1, min_samples_split=2, n_estimators=500,
+              random_state=None, subsample=1.0, verbose=0,
+              warm_start=False)
+Grid scores on development set:
+0.658 (+/-0.025) for {'max_features': 0.25, 'n_estimators': 500, 'learning_rate': 0.05, 'max_depth': 2}
+Training
+____KeepChannel_[28, 29, 30]__ButterworthFilter_0.1_15_4_200_epoched__EEGConcatExtracter__Downsampler_dec_12__KeepChannel_[28, 29, 30]__ButterworthFilter_0.1_15_4_200_epoched__<transforms.ActualWaveletTransform instance at 0x7f6c07a3e998>__EEGConcatExtracter__Downsampler_dec_24
+(5440, 260, 3)
+Fitting 4 folds for each of 1 candidates, totalling 4 fits
+[Parallel(n_jobs=-1)]: Done   1 out of   4 | elapsed:    9.0s remaining:   26.9s
+[Parallel(n_jobs=-1)]: Done   4 out of   4 | elapsed:    9.2s finished
+Cross Validation Stats
+Best parameters set:
+GradientBoostingClassifier(init=None, learning_rate=0.05, loss='deviance',
+              max_depth=2, max_features=0.25, max_leaf_nodes=None,
+              min_samples_leaf=1, min_samples_split=2, n_estimators=500,
+              random_state=None, subsample=1.0, verbose=0,
+              warm_start=False)
+Grid scores on development set:
+0.657 (+/-0.025) for {'max_features': 0.25, 'n_estimators': 500, 'learning_rate': 0.05, 'max_depth': 2}

InriaBCIChallenge/batchio.py

@@ -0,0 +1,69 @@
+from pandas import read_csv
+import os
+from extras import *
+import numpy as np
+from model_gen import get_preproc
+
+def read_metadata(params):
+    """
+    Return metadata about channels and extra associated information given in 'ChannelsLocation.csv'
+    """
+    metadata_fname = "ChannelsLocation.csv"
+    return read_csv(os.path.join(params.data_dir,metadata_fname))
+
+def read_maindata(params,type):
+    """
+    Read training/testing set files and extract the essential X,y matrices
+    which can be processed further. To do this we first get list of files
+    corresponding to subjects given in 'settings.json'. Then parse these to
+    get a final list of dataframes with every item corresponding to a single
+    feedback event and mapped y label
+    :param params:'settings.json' paramters
+    :param type:'train' or 'test'
+    :return: if 'train' then returns X,y
+             if 'test' then returns X
+    """
+    assert (type == "train" or type == "test"),"type field must be either train or test"
+    flist,subjsess_list = get_filelist(params,type)
+    X = []
+    if type == "train":
+        train_labels = read_csv(os.path.join(params.data_dir,"TrainLabels.csv"))
+        y = []
+        for findex in range(len(flist)):
+            X.extend(get_x(read_csv(flist[findex]),0,260))
+            y.extend(train_labels[
+                train_labels["IdFeedBack"].str.contains(
+                    subjsess_list[findex])]["Prediction"])
+        assert (len(X) == len(y)),"Training and Prediction Set values dont match"
+        return (np.array(X),y)
+    elif type == "test":
+        for findex in range(len(flist)):
+            X.extend(get_x(read_csv(flist[findex]),0,260))
+        return (np.array(X))
+
+def get_x(my_df,start_offset,end_offset):
+    """
+    Every loaded csv file has EEG + EOG + FeedbackEvent info
+    This function takes a loaded csv, extracts the necessary feedback events
+    Finally depending on provided indices here it extracts required time periods
+    of data for every feedback event
+    :param my_df: csv read as dataframe
+    :return: list of dataframes where each dataframe corresponds
+    to data for a given time period corresponding to a FeedbackEvent
+    """
+    fb_list = []
+    fb_indices = my_df[my_df["FeedBackEvent"] == 1].index.tolist()
+    my_df = my_df.drop('FeedBackEvent', axis = 1).drop('Time',axis=1).drop('EOG',axis=1).as_matrix()
+    # my_df = my_df.drop('FeedBackEvent', axis = 1).as_matrix()
+    my_df = get_preproc().transform(my_df)
+    for fb_ind in fb_indices:
+        fb_list.append(my_df[fb_ind + start_offset:fb_ind + end_offset,:])
+    return fb_list
+
+
+
+
+
+
+
+