begin the model stuff

alphamind/model/data_preparing.py

@@ -61,7 +61,8 @@ def prepare_data(engine: SqlEngine,
     df = pd.merge(df, industry_df, on=['trade_date', 'code'])
     df['weight'] = df['weight'].fillna(0.)
 
-    return df[['trade_date', 'code', 'dx']], df[['trade_date', 'code', 'weight', 'isOpen', 'industry_code', 'industry'] + transformer.names]
+    return df[['trade_date', 'code', 'dx']], df[
+        ['trade_date', 'code', 'weight', 'isOpen', 'industry_code', 'industry'] + transformer.names]
 
 
 def batch_processing(x_values,
@@ -75,6 +76,7 @@ def batch_processing(x_values,
     train_x_buckets = {}
     train_y_buckets = {}
     predict_x_buckets = {}
+    predict_y_buckets = {}
 
     for i, start in enumerate(groups[:-batch]):
         end = groups[i + batch]
@@ -110,7 +112,16 @@ def batch_processing(x_values,
                                  risk_factors=this_risk_exp,
                                  post_process=post_process)
         predict_x_buckets[end] = ne_x[sub_dates == end]
-    return train_x_buckets, train_y_buckets, predict_x_buckets
+
+        this_raw_y = y_values[index]
+        if len(this_raw_y) > 0:
+            ne_y = factor_processing(this_raw_y,
+                                     pre_process=pre_process,
+                                     risk_factors=this_risk_exp,
+                                     post_process=post_process)
+            predict_y_buckets[end] = ne_y[sub_dates == end]
+
+    return train_x_buckets, train_y_buckets, predict_x_buckets, predict_y_buckets
 
 
 def fetch_data_package(engine: SqlEngine,
@@ -170,7 +181,7 @@ def fetch_data_package(engine: SqlEngine,
 
     alpha_logger.info("Loading data is finished")
 
-    train_x_buckets, train_y_buckets, predict_x_buckets = batch_processing(x_values,
+    train_x_buckets, train_y_buckets, predict_x_buckets, predict_y_buckets = batch_processing(x_values,
                                                                                               y_values,
                                                                                               dates,
                                                                                               date_label,
@@ -185,7 +196,7 @@ def fetch_data_package(engine: SqlEngine,
     ret['x_names'] = transformer.names
     ret['settlement'] = return_df
     ret['train'] = {'x': train_x_buckets, 'y': train_y_buckets}
-    ret['predict'] = {'x': predict_x_buckets}
+    ret['predict'] = {'x': predict_x_buckets, 'y': predict_y_buckets}
     return ret
 
 

alphamind/model/linearmodel.py

@@ -5,66 +5,49 @@ Created on 2017-5-10
 @author: cheng.li
 """
 
-from typing import Tuple
-from typing import Union
+import pickle
 import numpy as np
-import numba as nb
-from alphamind.utilities import groupby
-from alphamind.data.neutralize import ls_fit
+from distutils.version import LooseVersion
+from sklearn import __version__ as sklearn_version
+from sklearn.linear_model import LinearRegression as LinearRegressionImpl
+from alphamind.model.modelbase import ModelBase
+from alphamind.utilities import alpha_logger
 
 
-class LinearModel(object):
+class LinearRegression(ModelBase):
 
-    def __init__(self, init_param=None):
-        self.model_parameter = init_param
+    def __init__(self, features, fit_intercept: bool=False):
+        super().__init__(features)
+        self.impl = LinearRegressionImpl(fit_intercept=fit_intercept)
 
-    def calibrate(self, x, y, groups=None):
-        self.model_parameter = _train(x, y, groups)
+    def fit(self, x: np.ndarray, y: np.ndarray):
+        self.impl.fit(x, y)
 
-    def predict(self, x, groups=None):
-        if groups is not None and isinstance(self.model_parameter, tuple):
-            names = np.unique(groups)
-            return _prediction_impl(self.model_parameter[0], self.model_parameter[1], groups, names, x)
-        elif self.model_parameter is None:
-            raise ValueError("linear model is not calibrated yet")
-        elif groups is None:
-            return x @ self.model_parameter
-        else:
-            raise ValueError("grouped x value can't be used for vanilla linear model")
+    def predict(self, x: np.ndarray) -> np.ndarray:
+        return self.impl.predict(x)
 
+    def save(self) -> dict:
+        model_desc = super().save()
+        model_desc['desc'] = pickle.dumps(self.impl)
+        model_desc['sklearn_version'] = sklearn_version
+        return model_desc
 
-@nb.njit(nogil=True, cache=True)
-def _prediction_impl(calibrated_names, model_parameter, groups, names, x):
-    places = np.searchsorted(calibrated_names, names)
-    pred_v = np.zeros(x.shape[0])
-    for k, name in zip(places, names):
-        this_param = model_parameter[k]
-        idx = groups == name
-        pred_v[idx] = x[idx] @ this_param
-    return pred_v
+    def load(self, model_desc: dict):
+        super().load(model_desc)
 
+        if LooseVersion(sklearn_version) < LooseVersion(model_desc['sklearn_version']):
+            alpha_logger.warning('Current sklearn version {0} is lower than the model version {1}. '
+                                 'Loaded model may work incorrectly.'.format(
+                sklearn_version, model_desc['sklearn_version']))
 
-def _train(x: np.ndarray, y: np.ndarray, groups: np.ndarray=None) -> Union[np.ndarray, Tuple[np.ndarray, np.ndarray]]:
-    if groups is None:
-        return ls_fit(x, y)
-    else:
-        index_diff, order = groupby(groups)
-        res_beta = _train_loop(index_diff, order, x, y)
-        return np.unique(groups), res_beta
+        self.impl = pickle.loads(model_desc['desc'])
 
 
-@nb.njit(nogil=True, cache=True)
-def _train_loop(index_diff, order, x, y):
-    res_beta = np.zeros((len(index_diff)+1, x.shape[1]))
-    start = 0
-    for k, diff_loc in enumerate(index_diff):
-        res_beta[k] = _train_sub_group(x, y, order[start:diff_loc + 1])
-        start = diff_loc + 1
-    return res_beta
+if __name__ == '__main__':
 
+    import pprint
+    ls = LinearRegression(['a', 'b'])
 
-@nb.njit(nogil=True, cache=True)
-def _train_sub_group(x, y, curr_idx):
-    curr_x = x[curr_idx]
-    curr_y = y[curr_idx]
-    return ls_fit(curr_x, curr_y)
+    model_desc = ls.save()
+    new_model = ls.load(model_desc)
+    pprint.pprint(model_desc)

alphamind/model/modelbase.py

+# -*- coding: utf-8 -*-
+"""
+Created on 2017-9-4
+
+@author: cheng.li
+"""
+
+import abc
+import arrow
+import numpy as np
+from alphamind.utilities import alpha_logger
+
+
+class ModelBase(metaclass=abc.ABCMeta):
+
+    def __init__(self, features: list):
+        self.features = features
+
+    @abc.abstractmethod
+    def fit(self, x, y):
+        pass
+
+    @abc.abstractmethod
+    def predict(self, x) -> np.ndarray:
+        pass
+
+    @abc.abstractmethod
+    def save(self) -> dict:
+
+        if self.__class__.__module__ == '__main__':
+            alpha_logger.warning("model is defined in a main module. The model_name may not be correct.")
+
+        model_desc = dict(internal_model=self.impl.__class__.__module__ + "." + self.impl.__class__.__name__,
+                          model_name=self.__class__.__module__ + "." + self.__class__.__name__,
+                          language='python',
+                          timestamp=arrow.now().format(),
+                          features=self.features)
+        return model_desc
+
+    @abc.abstractmethod
+    def load(self, model_desc: dict):
+        self.features = model_desc['features']
+

requirements.txt

+arrow >= 0.10.0
 cython >= 0.25.2
 finance-python >= 0.5.7
 mysqlclient >= 1.3.10