liuyu/qlib
1
0
Fork 0
mirror of https://github.com/microsoft/qlib.git synced 2026-07-13 07:46:53 +08:00
Code Activity

format

Browse Source
This commit is contained in:
Yuchen Fang
2021-01-28 00:41:02 +08:00
parent 98086e4fdc
commit a03b08bb4c
21 changed files with 154 additions and 563 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
examples/trade/observation/obs_rule.py
View File

@@ -60,9 +60,7 @@ class RuleObs(BaseObs):
prediction = [df_list[i].reshape(-1) for i in range(len(df_list))]
for i, p in enumerate(prediction):
if len(p) < interval_num:
prediction[i] = np.concatenate(
(p, np.zeros(interval_num - len(p))), axis=-1
)
prediction[i] = np.concatenate((p, np.zeros(interval_num - len(p))), axis=-1)
# res = np.stack(prediction).transpose().reshape(-1)
return np.concatenate(prediction)
for i in range(len(self.features)):
@@ -73,9 +71,7 @@ class RuleObs(BaseObs):
if time == -1:
predictions += [0.0] * size
else:
predictions += (
df.iloc[size * time : size * (time + 1)].reshape(-1).tolist()
)
predictions += df.iloc[size * time : size * (time + 1)].reshape(-1).tolist()
elif feature["type"] == "daily":
predictions += df.reshape(-1)[:size].tolist()
elif feature["type"] == "range":
@@ -86,35 +82,19 @@ class RuleObs(BaseObs):
else:
predictions += df.iloc[time : size + time].reshape(-1).tolist()
elif feature["type"] == "interval":
if (
len(df.iloc[interval * size : (interval + 1) * size].reshape(-1))
== size
):
predictions += (
df.iloc[interval * size : (interval + 1) * size]
.reshape(-1)
.tolist()
)
if len(df.iloc[interval * size : (interval + 1) * size].reshape(-1)) == size:
predictions += df.iloc[interval * size : (interval + 1) * size].reshape(-1).tolist()
else:
predictions += [0.0] * size
elif feature["type"] == "step":
if (
len(df.iloc[size * (time + 1) : size * (time + 2)].reshape(-1))
== size
):
predictions += (
df.iloc[size * (time + 1) : size * (time + 2)]
.reshape(-1)
.tolist()
)
if len(df.iloc[size * (time + 1) : size * (time + 2)].reshape(-1)) == size:
predictions += df.iloc[size * (time + 1) : size * (time + 2)].reshape(-1).tolist()
else:
predictions += [0.0] * size
return np.array(predictions)
def get_obs(
self, raw_df, feature_dfs, t, interval, position, target, is_buy, *args, **kargs
):
def get_obs(self, raw_df, feature_dfs, t, interval, position, target, is_buy, *args, **kargs):
private_state = np.array([position, target, t, self.max_step_num])
prediction_state = self.get_feature_res(feature_dfs, t, interval)
return {

17
examples/trade/observation/ppo_obs.py
View File

@@ -11,17 +11,7 @@ class PPOObs(RuleObs):
"""The observation defined in IJCAI 2020. The action of previous state is included in private state"""
def get_obs(
self,
raw_df,
feature_dfs,
t,
interval,
position,
target,
is_buy,
max_step_num,
interval_num,
action=0,
self, raw_df, feature_dfs, t, interval, position, target, is_buy, max_step_num, interval_num, action=0,
):
if t == -1:
self.private_states = []
@@ -32,10 +22,7 @@ class PPOObs(RuleObs):
self.private_states.append(private_state)
list_private_state = np.concatenate(self.private_states)
list_private_state = np.concatenate(
(
list_private_state,
[0.0] * 3 * (interval_num + 1 - len(self.private_states)),
)
(list_private_state, [0.0] * 3 * (interval_num + 1 - len(self.private_states)),)
)
seqlen = np.array([interval])
return np.concatenate((public_state, list_private_state, seqlen))

48
examples/trade/observation/teacher_obs.py
View File

@@ -16,18 +16,7 @@ class TeacherObs(RuleObs):
"""
def get_obs(
self,
raw_df,
feature_dfs,
t,
interval,
position,
target,
is_buy,
max_step_num,
interval_num,
*args,
**kargs,
self, raw_df, feature_dfs, t, interval, position, target, is_buy, max_step_num, interval_num, *args, **kargs,
):
if t == -1:
self.private_states = []
@@ -36,18 +25,13 @@ class TeacherObs(RuleObs):
self.private_states.append(private_state)
list_private_state = np.concatenate(self.private_states)
list_private_state = np.concatenate(
(
list_private_state,
[0.0] * 2 * (interval_num + 1 - len(self.private_states)),
)
(list_private_state, [0.0] * 2 * (interval_num + 1 - len(self.private_states)),)
)
seqlen = np.array([interval])
assert not (
np.isnan(list_private_state).any() | np.isinf(list_private_state).any()
), f"{private_state}, {target}"
assert not (
np.isnan(public_state).any() | np.isinf(public_state).any()
), f"{public_state}"
assert not (np.isnan(public_state).any() | np.isinf(public_state).any()), f"{public_state}"
return np.concatenate((public_state, list_private_state, seqlen))
@@ -55,35 +39,17 @@ class RuleTeacher(RuleObs):
""" """
def get_obs(
self,
raw_df,
feature_dfs,
t,
interval,
position,
target,
is_buy,
max_step_num,
interval_num,
*args,
**kargs,
self, raw_df, feature_dfs, t, interval, position, target, is_buy, max_step_num, interval_num, *args, **kargs,
):
if t == -1:
self.private_states = []
public_state = feature_dfs[0].reshape(-1)[: 6 * 240]
private_state = np.array([position / target, (t + 1) / max_step_num])
teacher_action = self.get_feature_res(feature_dfs, t, interval)[
-self.features[1]["size"] :
]
teacher_action = self.get_feature_res(feature_dfs, t, interval)[-self.features[1]["size"] :]
self.private_states.append(private_state)
list_private_state = np.concatenate(self.private_states)
list_private_state = np.concatenate(
(
list_private_state,
[0.0] * 2 * (interval_num + 1 - len(self.private_states)),
)
(list_private_state, [0.0] * 2 * (interval_num + 1 - len(self.private_states)),)
)
seqlen = np.array([interval])
return np.concatenate(
(teacher_action, public_state, list_private_state, seqlen)
)
return np.concatenate((teacher_action, public_state, list_private_state, seqlen))