def calc_feature(self, org_train, org_test):
if self.datatype == "train":
df = org_train
df = df.loc[df.installation_id.isin(
self.train_labels.installation_id.unique())]
else:
# 直前までのnum_correct/incorrectを取得する
df = org_test
c_ass_idx = ((df.type == "Assessment")
& (df["event_data"].str.contains("true"))) | \
((df.type == "Assessment")
& (df["event_data"].str.contains("true")))
inc_ass_idx = ((df.type == "Assessment")
& (df["event_data"].str.contains("false"))) | \
((df.type == "Assessment")
& (df["event_data"].str.contains("false")))
df.loc[c_ass_idx, 'num_correct'] = 1
df.loc[inc_ass_idx, 'num_incorrect'] = 1
ret = applyParallel(df.groupby("installation_id"), self._calc_features)
self.format_and_save_feats(ret)
return ret
def calc_feature(self, org_train, org_test):
if self.datatype == "train":
df = org_train
df = df.loc[df.installation_id.isin(
self.train_labels.installation_id.unique())]
else:
# 直前までのnum_correct/incorrectを取得する
df = org_test
ret = applyParallel(df.groupby("installation_id"),
self.ins_id_sessions)
ret_col = [
c for c in list(ret.columns) if c not in [
"accuracy", "accuracy_group", "cum_accuracy", "game_session",
"installation_id", "title", "type"
]
]
# self.format_and_save_feats(ret)
use_cols = [
c for c in list(ret.columns) if c not in [
"accuracy", "accuracy_group", "cum_accuracy", "title", "type",
"event_code", "gs_max_time"
]
]
self.format_and_save_feats(ret[use_cols])
return ret[use_cols]
def calc_feature(self, org_train, org_test):
if self.datatype == "train":
df = org_train
assess_user = df.loc[df.type ==
"Assessment"].installation_id.unique()
df = df.loc[df.installation_id.isin(assess_user)]
else:
# 直前までのnum_correct/incorrectを取得する
org_test.loc[(org_test.event_code.isin([4100, 4110])) &
(org_test["event_data"].str.contains("true")),
'num_correct'] = 1
org_test.loc[(org_test.event_code.isin([4100, 4110])) &
(org_test["event_data"].str.contains("false")),
'num_incorrect'] = 1
df = org_test
# get encodings informations
self.get_encoder(org_train, org_test)
ret = applyParallel(df.groupby("installation_id"),
self.ins_id_sessions)
use_cols = [
c for c in list(ret.columns) if c not in [
"accuracy", "accuracy_group", "cum_accuracy", "title", "type",
"event_code", "gs_max_time"
]
]
self.format_and_save_feats(ret[use_cols])
return ret[use_cols]
def calc_feature(self, org_train, org_test):
if self.datatype == "train":
df = org_train
assess_user = df.loc[df.type ==
"Assessment"].installation_id.unique()
df = df.loc[df.installation_id.isin(assess_user)]
else:
# 直前までのnum_correct/incorrectを取得する
df = org_test
ret = applyParallel(df.groupby("installation_id"), self.count_sessions)
ret_col = [
c for c in list(ret.columns)
if c not in ["game_session", "installation_id", "title", "type"]
]
ret[ret_col] = ret[ret_col].fillna(0).astype("int32")
use_cols = [
c for c in list(ret.columns)
if c not in ["title", "type", "event_code", "gs_max_time"]
]
self.format_and_save_feats(ret[use_cols])
return ret[use_cols]
def calc_feature(self, df):
target_c_ass_idx = ((df.event_code == 4100)
& (df.title != "Bird Measurer (Assessment)")
& (df.type == "Assessment")
& (df["event_data"].str.contains("true"))) | \
((df.event_code == 4110)
& (df.title == "Bird Measurer (Assessment)")
& (df.type == "Assessment")
& (df["event_data"].str.contains("true")))
target_inc_ass_idx = ((df.event_code == 4100)
& (df.title != "Bird Measurer (Assessment)")
& (df.type == "Assessment")
& (df["event_data"].str.contains("false"))) | \
((df.event_code == 4110)
& (df.title == "Bird Measurer (Assessment)")
& (df.type == "Assessment")
& (df["event_data"].str.contains("false")))
df.loc[target_c_ass_idx, 'num_correct'] = 1
df.loc[target_inc_ass_idx, 'num_incorrect'] = 1
df = df[(df.type == 'Assessment')
& (((df.event_code == 4100)
& (df.title != 'Bird Measurer (Assessment)'))
| ((df.event_code == 4110)
& (df.title == 'Bird Measurer (Assessment)')))]
print(df.shape)
ret = applyParallel(df.groupby("installation_id"), self._calc_features)
self.format_and_save_feats(ret)
return ret
def calc_feature(self, org_train, org_test):
if self.datatype == "train":
df = org_train
df = df.loc[df.installation_id.isin(
self.train_labels.installation_id.unique())]
else:
df = org_test
c_ass_idx = (((df.event_code == 4100)
& (df.title != "Bird Measurer (Assessment)")
& (df["event_data"].str.contains("true"))) |
((df.event_code == 4110)
& (df.title == "Bird Measurer (Assessment)")
& (df["event_data"].str.contains("true"))) & (df["type"] == "Assessment"))
inc_ass_idx = (((df.event_code == 4100)
& (df.title != "Bird Measurer (Assessment)")
& (df["event_data"].str.contains("false"))) |
((df.event_code == 4110)
& (df.title == "Bird Measurer (Assessment)")
& (df["event_data"].str.contains("false"))) & (df["type"] == "Assessment"))
df.loc[c_ass_idx, 'num_correct'] = 1
df.loc[inc_ass_idx, 'num_incorrect'] = 1
ret = applyParallel(
df.groupby("installation_id"),
self.ins_id_sessions)
ret_col = [c for c in list(ret.columns) if c not in ["accuracy", "accuracy_group", "cum_accuracy",
"game_session", "installation_id", "title",
"type"
]]
use_cols = [c for c in list(ret.columns) if "Assessment" not in c]
del ret["accum_acc_gr_-99"], ret["prev_acc_gr_-99"]
fill_cols = [
c for c in list(
ret.columns) if c not in [
"cum_accuracy",
"cum_accuracy",
"prev_num_corrects",
"prev_num_incorrects"]]
ret[fill_cols] = ret[fill_cols].fillna(0)
if self.datatype == "train":
ret = pd.merge(
ret, self.train_labels, how="inner", on=[
"installation_id", "game_session"])
self.format_and_save_feats(ret)
return ret
def calc_feature(self, org_train, org_test):
if self.datatype == "train":
df = org_train
df = df.loc[df.installation_id.isin(
self.train_labels.installation_id.unique())]
else:
# 直前までのnum_correct/incorrectを取得する
df = org_test
ret = applyParallel(df.groupby("installation_id"), self._calc_features)
self.format_and_save_feats(ret)
return ret
def calc_feature(self, org_train, org_test):
# 直前までのnum_correct/incorrectを取得する
if self.datatype == "train":
df = org_train
else:
df = org_test
c_ass_idx = ((df.event_code == 4100)
& (df.type == "Assessment")
& (df.title != "Bird Measurer (Assessment)")
& (df["event_data"].str.contains("true"))) | \
((df.event_code == 4110)
& (df.type == "Assessment")
& (df.title == "Bird Measurer (Assessment)")
& (df["event_data"].str.contains("true")))
inc_ass_idx = ((df.event_code == 4100)
& (df.type == "Assessment")
& (df.title != "Bird Measurer (Assessment)")
& (df["event_data"].str.contains("false"))) | \
((df.event_code == 4110)
& (df.type == "Assessment")
& (df.title == "Bird Measurer (Assessment)")
& (df["event_data"].str.contains("false")))
df.loc[c_ass_idx, 'num_correct'] = 1
df.loc[inc_ass_idx, 'num_incorrect'] = 1
ret = applyParallel(df.groupby("installation_id"),
self.ins_id_sessions)
#_ret = []
#for grp_id, grp_df in tqdm(df.groupby("installation_id")):
# _ret.append(self.ins_id_sessions(grp_df))
#ret = pd.concat(_ret)
ret_col = [
c for c in list(ret.columns) if c not in [
"accuracy", "accuracy_group", "cum_accuracy", "game_session",
"installation_id", "title", "type"
]
]
ret[ret_col] = ret[ret_col].fillna(0).astype("int32")
self.format_and_save_feats(ret)
use_cols = [c for c in list(ret.columns) if "Assessment" not in c]
return ret[use_cols]
def calc_feature(self, org_train, org_test):
if self.datatype == "train":
df = org_train
df = df.loc[df.installation_id.isin(
self.train_labels.installation_id.unique())]
else:
# 直前までのnum_correct/incorrectを取得する
df = org_test
c_ass_idx = ((df.event_code == 4100)
& (df.title != "Bird Measurer (Assessment)")
& (df["event_data"].str.contains("true"))) | \
((df.event_code == 4110)
& (df.title == "Bird Measurer (Assessment)")
& (df["event_data"].str.contains("true")))
inc_ass_idx = ((df.event_code == 4100)
& (df.title != "Bird Measurer (Assessment)")
& (df["event_data"].str.contains("false"))) | \
((df.event_code == 4110)
& (df.title == "Bird Measurer (Assessment)")
& (df["event_data"].str.contains("false")))
df.loc[c_ass_idx, 'num_correct'] = 1
df.loc[inc_ass_idx, 'num_incorrect'] = 1
ret = applyParallel(df.groupby("installation_id"),
self.ins_id_sessions)
ret_col = [
c for c in list(ret.columns) if c not in [
"accuracy", "accuracy_group", "cum_accuracy", "game_session",
"installation_id", "title", "type"
]
]
ret[ret_col] = ret[ret_col].fillna(0).astype("int32")
use_cols = [c for c in list(ret.columns) if "Assessment" not in c]
ret = ret[use_cols]
self.format_and_save_feats(ret)
return ret
