Fix bugs in the new models

This commit is contained in:
D-X-Y 2021-03-24 21:17:55 +08:00
parent 15dda79e3b
commit e7467fd474
4 changed files with 67 additions and 60 deletions

@ -1 +1 @@
Subproject commit 419629e4d2eefed52ceb207afb887a47aac732ca Subproject commit ba56e4071efd1c08003eaf7e23978aaf81376dd1

View File

@ -47,13 +47,12 @@ task:
net_config: net_config:
name: basic name: basic
d_feat: 6 d_feat: 6
stem_dim: 48 embed_dim: 48
embed_dims: [48, 48, 48, 48, 48]
num_heads: [4, 4, 4, 4, 4] num_heads: [4, 4, 4, 4, 4]
mlp_hidden_multipliers: [4, 4, 4, 4, 4] mlp_hidden_multipliers: [4, 4, 4, 4, 4]
qkv_bias: True qkv_bias: True
pos_drop: 0.1 pos_drop: 0.1
other_drop: 0.1 other_drop: 0
opt_config: opt_config:
loss: mse loss: mse
GPU: 0 GPU: 0

View File

@ -45,7 +45,6 @@ DEFAULT_NET_CONFIG = None
_default_max_depth = 5 _default_max_depth = 5
DefaultSearchSpace = dict( DefaultSearchSpace = dict(
d_feat=6, d_feat=6,
stem_dim=spaces.Categorical(*_get_list_mul(8, 16)),
embed_dim=spaces.Categorical(*_get_list_mul(8, 16)), embed_dim=spaces.Categorical(*_get_list_mul(8, 16)),
num_heads=_get_mul_specs((1, 2, 4, 8), _default_max_depth), num_heads=_get_mul_specs((1, 2, 4, 8), _default_max_depth),
mlp_hidden_multipliers=_get_mul_specs((0.5, 1, 2, 4, 8), _default_max_depth), mlp_hidden_multipliers=_get_mul_specs((0.5, 1, 2, 4, 8), _default_max_depth),
@ -61,7 +60,6 @@ class SuperTransformer(super_core.SuperModule):
def __init__( def __init__(
self, self,
d_feat: int = 6, d_feat: int = 6,
stem_dim: super_core.IntSpaceType = DefaultSearchSpace["stem_dim"],
embed_dim: List[super_core.IntSpaceType] = DefaultSearchSpace["embed_dim"], embed_dim: List[super_core.IntSpaceType] = DefaultSearchSpace["embed_dim"],
num_heads: List[super_core.IntSpaceType] = DefaultSearchSpace["num_heads"], num_heads: List[super_core.IntSpaceType] = DefaultSearchSpace["num_heads"],
mlp_hidden_multipliers: List[super_core.IntSpaceType] = DefaultSearchSpace[ mlp_hidden_multipliers: List[super_core.IntSpaceType] = DefaultSearchSpace[
@ -74,15 +72,14 @@ class SuperTransformer(super_core.SuperModule):
): ):
super(SuperTransformer, self).__init__() super(SuperTransformer, self).__init__()
self._embed_dim = embed_dim self._embed_dim = embed_dim
self._stem_dim = stem_dim
self._num_heads = num_heads self._num_heads = num_heads
self._mlp_hidden_multipliers = mlp_hidden_multipliers self._mlp_hidden_multipliers = mlp_hidden_multipliers
# the stem part # the stem part
self.input_embed = super_core.SuperAlphaEBDv1(d_feat, stem_dim) self.input_embed = super_core.SuperAlphaEBDv1(d_feat, embed_dim)
self.cls_token = nn.Parameter(torch.zeros(1, 1, self.stem_dim)) self.cls_token = nn.Parameter(torch.zeros(1, 1, self.embed_dim))
self.pos_embed = super_core.SuperPositionalEncoder( self.pos_embed = super_core.SuperPositionalEncoder(
d_model=stem_dim, max_seq_len=max_seq_len, dropout=pos_drop d_model=embed_dim, max_seq_len=max_seq_len, dropout=pos_drop
) )
# build the transformer encode layers -->> check params # build the transformer encode layers -->> check params
_assert_types(num_heads, (tuple, list)) _assert_types(num_heads, (tuple, list))
@ -111,15 +108,13 @@ class SuperTransformer(super_core.SuperModule):
self.apply(self._init_weights) self.apply(self._init_weights)
@property @property
def stem_dim(self): def embed_dim(self):
return spaces.get_max(self._stem_dim) return spaces.get_max(self._embed_dim)
@property @property
def abstract_search_space(self): def abstract_search_space(self):
root_node = spaces.VirtualNode(id(self)) root_node = spaces.VirtualNode(id(self))
if not spaces.is_determined(self._stem_dim): if not spaces.is_determined(self._embed_dim):
root_node.append("_stem_dim", self._stem_dim.abstract(reuse_last=True))
if not spaces.is_determined(self._stem_dim):
root_node.append("_embed_dim", self._embed_dim.abstract(reuse_last=True)) root_node.append("_embed_dim", self._embed_dim.abstract(reuse_last=True))
xdict = dict( xdict = dict(
input_embed=self.input_embed.abstract_search_space, input_embed=self.input_embed.abstract_search_space,
@ -155,13 +150,13 @@ class SuperTransformer(super_core.SuperModule):
def forward_candidate(self, input: torch.Tensor) -> torch.Tensor: def forward_candidate(self, input: torch.Tensor) -> torch.Tensor:
batch, flatten_size = input.shape batch, flatten_size = input.shape
feats = self.input_embed(input) # batch * 60 * 64 feats = self.input_embed(input) # batch * 60 * 64
if not spaces.is_determined(self._stem_dim): if not spaces.is_determined(self._embed_dim):
stem_dim = self.abstract_child["_stem_dim"].value embed_dim = self.abstract_child["_embed_dim"].value
else: else:
stem_dim = spaces.get_determined_value(self._stem_dim) embed_dim = spaces.get_determined_value(self._embed_dim)
cls_tokens = self.cls_token.expand(batch, -1, -1) cls_tokens = self.cls_token.expand(batch, -1, -1)
cls_tokens = F.interpolate( cls_tokens = F.interpolate(
cls_tokens, size=(stem_dim), mode="linear", align_corners=True cls_tokens, size=(embed_dim), mode="linear", align_corners=True
) )
feats_w_ct = torch.cat((cls_tokens, feats), dim=1) feats_w_ct = torch.cat((cls_tokens, feats), dim=1)
feats_w_tp = self.pos_embed(feats_w_ct) feats_w_tp = self.pos_embed(feats_w_ct)
@ -191,7 +186,6 @@ def get_transformer(config):
if name == "basic": if name == "basic":
model = SuperTransformer( model = SuperTransformer(
d_feat=config.get("d_feat"), d_feat=config.get("d_feat"),
stem_dim=config.get("stem_dim"),
embed_dim=config.get("embed_dim"), embed_dim=config.get("embed_dim"),
num_heads=config.get("num_heads"), num_heads=config.get("num_heads"),
mlp_hidden_multipliers=config.get("mlp_hidden_multipliers"), mlp_hidden_multipliers=config.get("mlp_hidden_multipliers"),

View File

@ -2,17 +2,42 @@
"cells": [ "cells": [
{ {
"cell_type": "code", "cell_type": "code",
"execution_count": 1, "execution_count": null,
"metadata": {}, "metadata": {},
"outputs": [ "outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"library path: /Users/xuanyidong/Desktop/XAutoDL/lib\n"
]
},
{ {
"name": "stderr", "name": "stderr",
"output_type": "stream", "output_type": "stream",
"text": [ "text": [
"[95290:MainThread](2021-03-03 12:14:32,922) INFO - qlib.Initialization - [config.py:276] - default_conf: client.\n", "[61704:MainThread](2021-03-22 13:56:38,104) INFO - qlib.Initialization - [config.py:276] - default_conf: client.\n",
"[95290:MainThread](2021-03-03 12:14:32,925) WARNING - qlib.Initialization - [config.py:291] - redis connection failed(host=127.0.0.1 port=6379), cache will not be used!\n", "[61704:MainThread](2021-03-22 13:56:38,106) WARNING - qlib.Initialization - [config.py:291] - redis connection failed(host=127.0.0.1 port=6379), cache will not be used!\n",
"[95290:MainThread](2021-03-03 12:14:33,203) INFO - qlib.Initialization - [__init__.py:46] - qlib successfully initialized based on client settings.\n", "[61704:MainThread](2021-03-22 13:56:38,680) INFO - qlib.Initialization - [__init__.py:46] - qlib successfully initialized based on client settings.\n",
"[95290:MainThread](2021-03-03 12:14:33,205) INFO - qlib.Initialization - [__init__.py:47] - data_path=/Users/xuanyidong/.qlib/qlib_data/cn_data\n" "[61704:MainThread](2021-03-22 13:56:38,681) INFO - qlib.Initialization - [__init__.py:47] - data_path=/Users/xuanyidong/.qlib/qlib_data/cn_data\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"{'class': 'DatasetH',\n",
" 'kwargs': {'handler': {'class': 'Alpha158',\n",
" 'kwargs': {'end_time': '2020-08-01',\n",
" 'fit_end_time': '2014-12-31',\n",
" 'fit_start_time': '2008-01-01',\n",
" 'instruments': 'csi100',\n",
" 'start_time': '2008-01-01'},\n",
" 'module_path': 'qlib.contrib.data.handler'},\n",
" 'segments': {'test': ('2017-01-01', '2020-08-01'),\n",
" 'train': ('2008-01-01', '2014-12-31'),\n",
" 'valid': ('2015-01-01', '2016-12-31')}},\n",
" 'module_path': 'qlib.data.dataset'}\n"
] ]
} }
], ],
@ -24,6 +49,16 @@
"import numpy as np\n", "import numpy as np\n",
"import pandas as pd\n", "import pandas as pd\n",
"\n", "\n",
"from pathlib import Path\n",
"\n",
"__file__ = os.path.dirname(os.path.realpath(\"__file__\"))\n",
"\n",
"lib_dir = (Path(__file__).parent / \"..\" / \"lib\").resolve()\n",
"print(\"library path: {:}\".format(lib_dir))\n",
"assert lib_dir.exists(), \"{:} does not exist\".format(lib_dir)\n",
"if str(lib_dir) not in sys.path:\n",
" sys.path.insert(0, str(lib_dir))\n",
"\n",
"from qlib import config as qconfig\n", "from qlib import config as qconfig\n",
"from qlib.utils import init_instance_by_config\n", "from qlib.utils import init_instance_by_config\n",
"\n", "\n",
@ -41,7 +76,7 @@
" \"end_time\": \"2020-08-01\",\n", " \"end_time\": \"2020-08-01\",\n",
" \"fit_start_time\": \"2008-01-01\",\n", " \"fit_start_time\": \"2008-01-01\",\n",
" \"fit_end_time\": \"2014-12-31\",\n", " \"fit_end_time\": \"2014-12-31\",\n",
" \"instruments\": \"csi300\",\n", " \"instruments\": \"csi100\",\n",
" },\n", " },\n",
" },\n", " },\n",
" \"segments\": {\n", " \"segments\": {\n",
@ -50,7 +85,15 @@
" \"test\": (\"2017-01-01\", \"2020-08-01\"),\n", " \"test\": (\"2017-01-01\", \"2020-08-01\"),\n",
" },\n", " },\n",
" },\n", " },\n",
" }" " }\n",
"pprint.pprint(dataset_config)\n",
"dataset = init_instance_by_config(dataset_config)\n",
"\n",
"df_train, df_valid, df_test = dataset.prepare(\n",
" [\"train\", \"valid\", \"test\"],\n",
" col_set=[\"feature\", \"label\"],\n",
" data_key=DataHandlerLP.DK_L,\n",
" )"
] ]
}, },
{ {
@ -89,39 +132,10 @@
} }
], ],
"source": [ "source": [
"pprint.pprint(dataset_config)\n", "from trade_models.transformations import get_transformer\n",
"dataset = init_instance_by_config(dataset_config)" "\n",
"model = get_transformer(None)"
] ]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{'class': 'MLflowExpManager', 'module_path': 'qlib.workflow.expm', 'kwargs': {'uri': 'file:/Users/xuanyidong/Desktop/AutoDL-Projects/notebooks/Q/mlruns', 'default_exp_name': 'Experiment'}}\n",
"Wrapper(provider=<qlib.workflow.QlibRecorder object at 0x7ff46b8a4850>)\n",
"<qlib.workflow.expm.MLflowExpManager object at 0x7ff46b8a4c10>\n"
]
}
],
"source": [
"from qlib.workflow import R\n",
"from qlib.config import C\n",
"print(C.exp_manager)\n",
"print(R)\n",
"print(R.exp_manager)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
} }
], ],
"metadata": { "metadata": {
@ -140,7 +154,7 @@
"name": "python", "name": "python",
"nbconvert_exporter": "python", "nbconvert_exporter": "python",
"pygments_lexer": "ipython3", "pygments_lexer": "ipython3",
"version": "3.8.3" "version": "3.8.8"
} }
}, },
"nbformat": 4, "nbformat": 4,