def kfold_train(gamma=5):
# with open('../data/node_dict.pkl','r') as f:
# graph_nodes = cPickle.load( f )
graph_nodes = build_graph()
lat_lng_df = pd.read_pickle('../data/all_intersections_df.pkl')
df_train = pd.read_pickle('../data/train_df.pkl')
X = df_train.loc[:, ['X', 'Y']].values
y = df_train.loc[:, 'Category'].values
gc = GraphClassifier(graph_nodes, lat_lng_df, gamma=gamma)
gc.fit(X, y)
with open('../data/graph_class_gamma-{0}.pkl'.format(gamma),'w') as f:
cPickle.dump( gc, f, -1 )
def yst_train(gamma=20):
# with open('../data/node_dict.pkl','r') as f:
# graph_nodes = cPickle.load( f )
graph_nodes = build_graph()
lat_lng_df = pd.read_pickle('../data/all_intersections_df.pkl')
df_train = pd.read_pickle('../data/train_df_yst.pkl')
X = df_train.loc[:, ['X', 'Y']].values
y = df_train.loc[:, 'Category'].values
# yst_columns = [ 'year_group', 'season', 'time_of_day', 'weekend' ]
yst_columns = [ 'year', 'month', 'DayOfWeek' ]
yst_column_values = [ df_train[col].values for col in yst_columns ]
yst_dict = { col: val for col, val in izip(yst_columns, yst_column_values) }
gc = GraphClassifier(graph_nodes, lat_lng_df, gamma=gamma)
gc.fit(X, y, yst_dict)
with open('../data/graph_class_yst_gamma-{0}.pkl'.format(gamma),'w') as f:
cPickle.dump( gc, f, -1 )
def train_model(predict=True, use_resampling=False, num_resamples=10):
graph_nodes = build_graph()
with open('../data/node_dict.pkl','w') as f:
cPickle.dump( graph_nodes, f, -1 )
lat_lng_df = pd.read_pickle('../data/all_intersections_df.pkl')
df_train = pd.read_csv('../data/train.csv', parse_dates=['Dates'])
X = df_train.loc[:, ['X', 'Y']].values
y = df_train.loc[:, 'Category'].values
if use_resampling:
resamples = num_resamples
else:
resamples = None
gc = GraphClassifier(graph_nodes, lat_lng_df, num_resamples=resamples)
gc.fit(X, y)
# save fit model
if use_resampling:
with open('../data/graph_class_resample_{0}.pkl'.format(num_resamples),'w') as f:
cPickle.dump( gc, f, -1 )
else:
with open('../data/graph_class_normal.pkl','w') as f:
cPickle.dump( gc, f, -1 )
if predict:
predict_df = gc.predict(X)
print "\nsaving submission..."
with gzip.open('../data/submissions/submission_training_data.csv.gz', 'w') as f:
predict_df.to_csv(f, index=False)
print "Done!"
