import yfinance as yf   
from datetime import datetime
from datetime import timedelta
import time
import pathlib

import pandas as pd
import numpy as np

import tsai
from tsai.all import *
import wandb
from scipy.ndimage.interpolation import shift
from fastai.callback.wandb import *

project = 'ts-rnn'
# Get the data for the stock Apple by specifying the stock ticker, start date, and end date 


# Multi day dataframe
poll_freq = '1m'
end = datetime.today().date()-timedelta(days=4)
start = end-timedelta(days=7)
tickers = ['AAPL']
data_1m = yf.download(tickers=tickers, end=end, start=start, interval=poll_freq)
data_1m = data_1m.sort_index()
data_1m.index = data_1m.index.tz_convert(None)

data_1m['ts'] = data_1m.index
row_1 = data_1m.iloc[0]

data_1m['tsHour'] = data_1m.apply(lambda row: row['ts'].to_pydatetime().hour , axis=1)
data_1m['tsMinute'] = data_1m.apply(lambda row: row['ts'].to_pydatetime().minute , axis=1)
data_1m['tsSecond'] = data_1m.apply(lambda row: row['ts'].to_pydatetime().second , axis=1)

start_dt = pd.Timestamp(year=row_1['ts'].year, month=row_1['ts'].month, day=row_1['ts'].day, hour=9, minute=30, second=0)
data_1m = add_datepart(data_1m, 'ts')

[*********************100%***********************]  1 of 1 completed

# feature eng

# day changes
def is_new_day(this_day, last_day):
    if pd.isnull(this_day):
        return True
    
    if pd.isnull(last_day):
        return True
    
    if last_day != this_day:
        return True
    
    return False

def create_template_row(df, fixed_val=np.nan):
    return {col: fixed_val for col in df.columns}

def offset_dateindex(ts, interval, units):
    if interval == '1m':
        return ts + pd.DateOffset(minutes=units)
    if interval == '2m':
        return ts + pd.DateOffset(minutes=2*units)
    
def insert_end_of_day_rows(df, first_daily_obs, end_day_val):
    new_df = [df]
    for obs_idx in first_daily_obs:
        first_obs = df.iloc[obs_idx]
        new_line_idx = offset_dateindex(
            ts=first_obs.name, 
            interval=poll_freq,
            units=-1
        )
        new_line_template = create_template_row(
            df, fixed_val=end_day_val
        )
        new_df.append(
            pd.DataFrame(new_line_template, index=[new_line_idx])
        )
    
    return pd.concat(new_df, axis=0).sort_index()

days = data_1m[['tsYear', 'tsMonth', 'tsDay']].drop_duplicates()
days.iloc[-2]

tsYear     2020
tsMonth      12
tsDay        23
Name: 2020-12-23 14:30:00, dtype: int64

data_1m['tsLastDay'] = data_1m['tsDay'].shift(1)
data_1m['isNewDay'] = data_1m.apply(
    lambda x:
    is_new_day(this_day=x['tsDay'], last_day=x['tsLastDay']), axis=1
)

Notes: There were issues installing tsai. Both sktime and tsai were installed from forks. TODO: need to add upstreams to both of those forks to keep them up to date.

Plan

train self supervise the network by using the previous minutes to predict the next minutes average
obtain labeled segments

use cases

predict reversals in the next timestamp
predict future trend

class ModelRun:
    def __init__(self, checkpoint):
        self.checkpoint = checkpoint
        self.savetodir = None
        self.savetoname = None
        self.dls = self.get_dls()
        self.learner = self.load()
    
    
    def load(self):
        pass
    
    
    def save(self):
        pass
    
    
    def load_existing_checkpoint(self, name, fp, dls, cbs=None):
        """Loads and existing model phase"""

        model = InceptionTime(dls.vars, dls.c)
        learn = Learner(
            dls, model,
            metrics=[mae, mse], 
            cbs=cbs
        )

        learn.model_dir = learn.model_dir/name
        return learn.load(fp)
    
    
    def save_training_run(self, name, fp, dls, learn):
        learn.model_dir = learn.model_dir/name
        learn.save(fp)


def create_df(X, splits, window_length, horizon, tfms, train_bs, valid_bs):
    X1, y1 = SlidingWindow(window_length, horizon=horizon)(X)
    splits1 = (L(splits[0]), L(splits[1][:-window_length]))
    ds = TSDatasets(X=X1, y=y1, tfms=tfms, splits=splits1)
    dls = TSDataLoaders.from_dsets(ds.train, ds.valid, bs=[train_bs, valid_bs], shuffle_train=False)
    return ds, dls, X1, y1, splits1


def create_base_learner(fp, dls):
    """Initializes a model with randomized weights."""
    print('create_base_learner', fp)
    model = InceptionTime(dls.vars, dls.c)
    learn = Learner(
        dls, model, metrics=[mae, mse]
    )
    return learn.save(fp)


def load_existing_checkpoint(fp, dls, cbs=None):
    """Loads and existing model phase"""
    print('load_existing_checkpoint', fp)
    model = InceptionTime(dls.vars, dls.c)
    learn = Learner(
        dls, model,
        metrics=[mae, mse], 
        cbs=cbs
    )

    return learn.load(fp)


def save_training_run(fp, dls, learn):
    """Saves a trained model"""
    print('save_training_run', fp)
    dlss[fp] = dls
    learn.save(fp,)
    print('save_training_run success')
    

def save_dataloader(model_dlss, model_id, dls_fp, dls):
    """Saves data loader."""
    if Path(dls_fp).exists():
        return
    
    torch.save(dls, dls_fp)
    model_dlss[model_id] = dls_fp
    
def load_dataloader(model_dlss, dls_fp):
    """Loads dataloader"""
    if not Path(dls_fp)
    return torch.load(fp)

import torch
torch.save(obj=dls, f=Path('data/ts-rnn/dls.pk'))

# separate ds initialization
X_feature = 'Close'
y_feature = 'Close'

# y_offset=1 means X(t) :-> y=X(t+1), y_offset=2 means X(t) :-> y=X(t+2), ..
# 
y_offset, window_length, horizon = 1,1,1
tfms  = [None, [ToFloat(), ToNumpyTensor()]]
train_bs, valid_bs = 64, 128

training_end_ts = datetime(days.iloc[-2]['tsYear'],days.iloc[-2]['tsMonth'],days.iloc[-2]['tsDay'], 23, 59, 0) # less than or equal to this date
reqd_cols = set([X_feature, y_feature, 'isNewDay',])
end_day_val = 0. # TODO: Experiment with different values
df = data_1m[reqd_cols]

print('Training end timestamp', training_end_ts)
noise_reduction_threshold = 3 #decimals
first_daily_obs = list(reversed(np.where(df['isNewDay'])[0]))
df = insert_end_of_day_rows(
    df=df, 
    first_daily_obs=first_daily_obs,
    end_day_val=end_day_val
)

del df['isNewDay']
cond = (df.index <= training_end_ts)        
X = df[X_feature].values
y = shift(input=df[y_feature].values, shift=y_offset, cval=np.NaN)

train_idx = np.where( cond)[0]
valid_idx = np.where(~cond)[0]

splits = (list(train_idx),list(valid_idx))
print('training:\t', len(train_idx), '\ntesting:\t', len(valid_idx))

ds, dls, X1, y1, splits1 = create_df(
    X=X,
    splits=splits,
    window_length=window_length, 
    horizon=horizon,
    tfms=tfms,
    train_bs=train_bs, 
    valid_bs=valid_bs
)

config_log = {
    "poll_freq": poll_freq,
    "ds.request_end": end,
    "ds.request_start": start,
    "tickers": tickers,
    "ds.actual_start": start_dt,
    "X_feature": X_feature,
    "y_feature": y_feature,
    "training_end_ts": training_end_ts,
    "noise_reduction_threshold": noise_reduction_threshold,
    "first_daily_obs": first_daily_obs,
    "window_length": window_length,
    "horizon": horizon,
    "tfms": tfms,
    "train_bs": train_bs,
    "valid_bs": valid_bs,
    "y_offset": y_offset
}

Training end timestamp 2020-12-23 23:59:00
training:	 1171 
testing:	 212

Files:

best models can be found on wandb https://wandb.ai/soellingeraj/ts-rnn/artifacts/model/run-bgo7lcbw-model

# initialize phase
p0_name = "single-dsconfig-training-phase0_sweep"
next_phase_name = "single-dsconfig-training-phase1_sweep"
p = pathlib.Path('models')/next_phase_name
p.mkdir(parents=True, exist_ok=True)

N = 10
phase_0s = []
dlss = {}
logs = {}
for i in range(N):
    fp_in = os.path.join(p0_name, str(i))
    fp_out = create_base_learner(fp=fp_in, dls=dls)
    phase_0s.append(fp_in)
    dlss[fp_in] = dls
    logs[fp_in] = config_log

create_base_learner single-dsconfig-training-phase0_sweep/0
create_base_learner single-dsconfig-training-phase0_sweep/1
create_base_learner single-dsconfig-training-phase0_sweep/2
create_base_learner single-dsconfig-training-phase0_sweep/3
create_base_learner single-dsconfig-training-phase0_sweep/4
create_base_learner single-dsconfig-training-phase0_sweep/5
create_base_learner single-dsconfig-training-phase0_sweep/6
create_base_learner single-dsconfig-training-phase0_sweep/7
create_base_learner single-dsconfig-training-phase0_sweep/8
create_base_learner single-dsconfig-training-phase0_sweep/9

p0_config = {
    "name": p0_name,
    "method": "grid",
    "metric": {
        "name": "mse",
        "goal": "minimize"
    },
    "parameters": {
            "checkpoint": {
                "values": phase_0s
            },
            "n_epoch": {
                "values": [1, 3, 5, 10, 20, 40, 80]
            },
            "lr_max": {
                "values": [1e-1 ,1e-2, 1e-3, 1e-4, 1e-5, 1e-6, 1e-7]
            },
            "div": {
                "values": [10., 25., 50., 75.]
            }
    }
}

sweep_id = wandb.sweep(p0_config, project=project)

Failed to query for notebook name, you can set it manually with the WANDB_NOTEBOOK_NAME environment variable

Create sweep with ID: vbyp7u8r
Sweep URL: https://wandb.ai/soellingeraj/ts-rnn/sweeps/vbyp7u8r

def get_config(phase):
    name = "single-dsconfig-training-phase{}_sweep".format(phase)
    phasepath = Path('models')/name
    checkpoints = phasepath.ls()
    config = {
        "name": "single-dsconfig-training-phase_sweep",
        "method": "random",
        "metric": {
            "name": "mse",
            "goal": "minimize"
        },
        "parameters": {
                "checkpoint": {
                    "values": checkpoints
                },
                "n_epoch": {
                    "values": [3, 5, 10, 20]
                },
                "lr_max": {
                    "values": [1e-1 ,1e-2, 1e-3, 1e-4, 1e-5, 1e-6, 1e-7]
                },
                "div": {
                    "values": [10., 25., 50., 75.]
                }
        }
    }
    
    return config

def train():
    default_config = {
        "checkpoint": phase_1s[0],
        "n_epoch": 3,
        "lr_max": 1e-06,
        "div": 50.
    }
    
    wandb.init(config={**default_config, **config_log})
    dls = dlss[default_config["checkpoint"]]
    
    # TODO: find a better way to manage next phase dirname
    learn = load_existing_checkpoint(
        fp=default_config["checkpoint"], 
        dls=dls,
        cbs=[WandbCallback(
            log_model=True, 
            log_preds=True,
            n_preds=len(valid_idx),
        ), SaveModelCallback()]
    )
    
    config = wandb.config
    learn.fit_one_cycle(
        config.n_epoch, 
        lr_max=config.lr_max,
        div=config.div
    )  
    outfp = os.path.join(next_phase_name, wandb.run.name)  
    save_training_run(
        fp=outfp,
        learn=learn,
        dls=dls
    )
    
    preds,targs = learn.get_preds()
    preds_df = pd.DataFrame(preds.flatten(), columns=['preds'])
    targs_df = pd.DataFrame(targs.flatten(), columns=['targs'])
    results = pd.concat([preds_df, targs_df], axis=1)
    wandb.log(
        {
            "predictions": results.plot()
        }
    )

wandb.agent(sweep_id=sweep_id, function=train, )

wandb: Agent Starting Run: enscz6k1 with config:
wandb: 	checkpoint: single-dsconfig-training-phase0_sweep/0
wandb: 	div: 10
wandb: 	lr_max: 0.0001
wandb: 	n_epoch: 5
Failed to query for notebook name, you can set it manually with the WANDB_NOTEBOOK_NAME environment variable

load_existing_checkpoint single-dsconfig-training-phase0_sweep/0
WandbCallback was not able to prepare a DataLoader for logging prediction samples -> tuple index out of range

Better model found at epoch 0 with valid_loss value: 17094.232421875.

wandb: Adding directory to artifact (/tmp/tmp9oah3rba)... Done. 0.1s

save_training_run single-dsconfig-training-phase1_sweep/driven-sweep-24
save_training_run success

/opt/conda/lib/python3.6/site-packages/plotly/matplotlylib/renderer.py:410: UserWarning:

Bummer! Plotly can currently only draw Line2D objects from matplotlib that are in 'data' coordinates!

/opt/conda/lib/python3.6/site-packages/plotly/matplotlylib/renderer.py:512: UserWarning:

I found a path object that I don't think is part of a bar chart. Ignoring.

wandb: Sweep Agent: Waiting for job.
wandb: Job received.
wandb: Agent Starting Run: 59z9gu7r with config:
wandb: 	checkpoint: single-dsconfig-training-phase0_sweep/0
wandb: 	div: 10
wandb: 	lr_max: 0.0001
wandb: 	n_epoch: 10
Failed to query for notebook name, you can set it manually with the WANDB_NOTEBOOK_NAME environment variable

load_existing_checkpoint single-dsconfig-training-phase0_sweep/0
WandbCallback was not able to prepare a DataLoader for logging prediction samples -> tuple index out of range

/opt/conda/lib/python3.6/site-packages/fastai/learner.py:55: UserWarning:

Saved filed doesn't contain an optimizer state.

Better model found at epoch 0 with valid_loss value: 17122.673828125.

learn.fit_one_cycle(10, lr_max=1e-3, div=50.,)
learn.fit_one_cycle(10, lr_max=1e-3, div=50.,)
learn.fit_one_cycle(10, lr_max=1e-3, div=50.,)
learn.fit_one_cycle(10, lr_max=1e-3, div=50.,)
learn.fit_one_cycle(10, lr_max=1e-3, div=50.,)
learn.fit_one_cycle(10, lr_max=0.1, div=50.,)
learn.fit_one_cycle(20, lr_max=0.1, div=50.,)
learn.fit_one_cycle(20, lr_max=0.1, div=50.,)
learn.fit_one_cycle(40, lr_max=slice(1e-07, 1e-06), div=50.,)
learn.fit_one_cycle(40, lr_max=slice(1e-07, 1e-06), div=50.,)
learn.fit_one_cycle(40, lr_max=slice(1e-04, 1e-1), div=50.,)
learn.fit_one_cycle(20,lr_max=slice(3e-07,9e-07), div=50.,)
learn.fit_one_cycle(20, lr_max=slice(1e-03, 1e-01), div=50.,)

learn.fit_one_cycle(10, lr_max=1e-3, div=50.,)
learn.fit_one_cycle(10, lr_max=1e-3, div=50.,)
learn.fit_one_cycle(10, lr_max=1e-3, div=50.,)
learn.fit_one_cycle(10, lr_max=1e-3, div=50.,)
learn.fit_one_cycle(10, lr_max=1e-3, div=50.,)
learn.fit_one_cycle(10, lr_max=0.1, div=50.,)
learn.fit_one_cycle(20, lr_max=0.1, div=50.,)
learn.fit_one_cycle(20, lr_max=0.1, div=50.,)
learn.fit_one_cycle(40, lr_max=slice(1e-07, 1e-06), div=50.,)
learn.fit_one_cycle(40, lr_max=slice(1e-07, 1e-06), div=50.,)
learn.fit_one_cycle(40, lr_max=slice(1e-04, 1e-1), div=50.,)
learn.fit_one_cycle(20,lr_max=slice(3e-07,9e-07), div=50.,)
learn.fit_one_cycle(20, lr_max=slice(1e-03, 1e-01), div=50.,)

WandbCallback requires use of "SaveModelCallback" to log best model
WandbCallback was not able to prepare a DataLoader for logging prediction samples -> tuple index out of range

wandb: WARNING Step must only increase in log calls.  Step 0 < 1; dropping {'epoch': 0.16666666666666666, 'train_loss': TensorBase(16534.7344), 'raw_loss': TensorBase(16534.7500), 'wd_0': 0.01, 'sqr_mom_0': 0.99, 'lr_0': 2e-05, 'mom_0': 0.95, 'eps_0': 1e-05}.

learn.save('best-Ar{}'.format(window_length))
learn.export('best-Ar{}-envoker'.format(window_length))

learn.save('best-Ar{}'.format(window_length))
learn.export('best-Ar{}-envoker'.format(window_length))

# 50 epochs
learn.lr_find(stop_div=False)

# 50 epochs
learn.lr_find(stop_div=False)

SuggestedLRs(lr_min=3.981071586167673e-07, lr_steep=9.12010818865383e-07)

learn.save('100epochs')
learn.export('100epochs-envoker')
learn = learn.load('100epochs')

learn.save('100epochs')
learn.export('100epochs-envoker')
learn = learn.load('100epochs')

learn_e = load_learner('100epochs-envoker')

learn_e = load_learner('100epochs-envoker')

preds,targs = learn.get_preds()
preds_df = pd.DataFrame(preds.flatten(), columns=['preds'])
targs_df = pd.DataFrame(targs.flatten(), columns=['targs'])
results = pd.concat([preds_df, targs_df], axis=1)

preds,targs = learn.get_preds()
preds_df = pd.DataFrame(preds.flatten(), columns=['preds'])
targs_df = pd.DataFrame(targs.flatten(), columns=['targs'])
results = pd.concat([preds_df, targs_df], axis=1)

preds.flatten()

preds.flatten()

tensor([253.2895, 127.3790, 128.3668,  ..., 125.5647, 125.4898, 125.4906])

itemify(X1, y1)[1][0]

itemify(X1, y1)[1][0]

array([[127.67810059]])

epoch	train_loss	valid_loss	mae	mse	time
0	16851.765625	17094.232422	130.742371	17094.232422	00:01
1	16785.460938	17414.597656	131.963181	17414.597656	00:01
2	16717.101562	17452.841797	132.108109	17452.841797	00:01
3	16660.125000	17421.761719	131.988907	17421.761719	00:01
4	16621.132812	17386.763672	131.854050	17386.763672	00:01

epoch	▁▁▁▂▂▂▂▂▂▃▃▃▃▃▃▄▄▄▄▄▅▅▅▅▅▅▆▆▆▆▆▇▇▇▇▇▇███
train_loss	▂▁▂▅▆▇███▇▆▇▇▇▇█▇▇▆▇▇▇▇▇▇▆▆▆▇▇▇▇▇▆▆▆▆▇▇▇
raw_loss	▃▃▄██▇▇▇▃▃▄██▇▇▇▂▂▃▇▇▆▇▇▂▃▃▆▇▆▇▇▁▂▃▆▇▆▇▄
wd_0	▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
sqr_mom_0	▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
lr_0	▂▂▂▃▄▅▆▆▇███████▇▇▇▇▆▆▅▅▅▄▄▄▃▃▃▂▂▂▂▁▁▁▁▁
mom_0	███▇▆▄▄▃▂▁▁▁▁▁▁▁▂▂▂▂▃▃▄▄▄▅▅▅▆▆▆▇▇▇▇█████
eps_0	▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
_step	▁▁▁▂▂▂▂▂▂▃▃▃▃▃▄▄▄▄▄▄▅▅▅▅▅▅▆▆▆▆▆▇▇▇▇▇▇███
_runtime	▁▂▂▂▂▂▂▂▂▂▃▃▃▃▃▃▄▄▄▅▅▅▅▅▅▅▅▆▆▆▆▇▇▇▇▇▇▇▇█
_timestamp	▁▂▂▂▂▂▂▂▂▂▃▃▃▃▃▃▄▄▄▅▅▅▅▅▅▅▅▆▆▆▆▇▇▇▇▇▇▇▇█
valid_loss	▁▇█▇▇
mae	▁▇█▇▇
mse	▁▇█▇▇

epoch	train_loss	valid_loss	mae	mse	time
0	16863.748047	17122.673828	130.850800	17122.673828	00:01
1	16818.023438	17425.871094	132.005035	17425.871094	00:01
2	16754.675781	17453.398438	132.109451	17453.398438	00:01
3	16683.044922	17422.099609	131.991074	17422.099609	00:01
4	16613.001953	17367.019531	131.782394	17367.019531	00:01
5	16550.369141	17261.785156	131.382187	17261.785156	00:01
6	16496.580078	17283.605469	131.465286	17283.605469	00:01
7	16453.351562	17280.964844	131.455246	17280.964844	00:01
8	16420.773438	17260.287109	131.376450	17260.287109	00:01
9	16397.923828	17260.642578	131.377579	17260.642578	00:01

epoch	▁▁▁▂▂▂▂▂▂▃▃▃▃▃▃▄▄▄▄▄▅▅▅▅▅▅▆▆▆▆▆▇▇▇▇▇▇███
train_loss	▁▂▆██▇▇█▇▇▇█▇▇▇▇▇▆▇▇▆▆▆▇▆▆▆▆▆▅▆▆▅▅▆▆▅▅▅▆
raw_loss	▂▄██▃▇█▇▂▄▇▇▃▇▇▇▁▃▇▇▂▆▆▆▁▇▆▆▂▆▆▆▂▆▆▆▂▆▆▄
wd_0	▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
sqr_mom_0	▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
lr_0	▂▂▂▃▄▅▆▇███████▇▇▇▇▇▆▆▅▅▅▄▄▄▃▃▃▂▂▂▂▁▁▁▁▁
mom_0	██▇▆▆▄▃▂▂▁▁▁▁▁▁▂▂▂▂▂▃▃▄▄▄▅▅▅▆▆▆▇▇▇▇█████
eps_0	▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
_step	▁▁▁▂▂▂▂▂▂▃▃▃▃▃▃▄▄▄▄▄▅▅▅▅▅▅▆▆▆▆▆▇▇▇▇▇▇███
_runtime	▁▁▁▁▂▂▂▂▂▂▂▂▃▃▃▃▄▄▄▄▄▅▅▅▅▅▅▅▆▆▆▆▇▇▇▇▇▇██
_timestamp	▁▁▁▁▂▂▂▂▂▂▂▂▃▃▃▃▄▄▄▄▄▅▅▅▅▅▅▅▆▆▆▆▇▇▇▇▇▇██
valid_loss	▁▇█▇▆▄▄▄▄▄
mae	▁▇█▇▆▄▄▄▄▄
mse	▁▇█▇▆▄▄▄▄▄

epoch	train_loss	valid_loss	mae	mse	time
0	16867.824219	17132.888672	130.889725	17132.888672	00:01
1	16839.199219	17433.878906	132.034576	17433.878906	00:01
2	16801.939453	17463.720703	132.146637	17463.720703	00:01
3	16752.289062	17461.597656	132.138443	17461.597656	00:01
4	16690.041016	17409.464844	131.942780	17409.464844	00:01
5	16621.437500	17337.548828	131.670731	17337.548828	00:01
6	16552.236328	17287.386719	131.480652	17287.386719	00:01
7	16486.255859	17292.101562	131.497726	17292.101562	00:01
8	16424.000000	17217.488281	131.211563	17217.488281	00:01
9	16366.322266	16441.832031	128.220978	16441.832031	00:01
10	16314.892578	17180.570312	131.063293	17180.570312	00:01
11	16267.355469	17145.031250	130.924408	17145.031250	00:01
12	16224.870117	17115.468750	130.808517	17115.468750	00:01
13	16187.346680	17111.597656	130.790070	17111.597656	00:01
14	16155.184570	17069.980469	130.627670	17069.980469	00:01
15	16128.502930	17074.386719	130.641922	17074.386719	00:01
16	16107.181641	17059.001953	130.581757	17059.001953	00:02
17	16090.777344	17060.320312	130.586044	17060.320312	00:01
18	16078.731445	17060.767578	130.587402	17060.767578	00:01
19	16070.275391	17061.562500	130.590424	17061.562500	00:01

Parameter Sweep for One Step Ahead Timeseries Prediction (using Weights and Biases Sweeps)

Plan

use cases

Run summary:

Run history:

Run summary:

Run history:

Run summary:

Run history:

Run summary:

Run history:

Run summary:

Run history:

Run summary:

Run history:

Run summary:

Run history:

Run summary:

Run history:

Run summary:

Run history:

Run summary:

Run history:

Run summary:

Run history:

epoch	5
train_loss	16621.13281
raw_loss	16067.28613
wd_0	0.01
sqr_mom_0	0.99
lr_0	0.0
mom_0	0.94995
eps_0	1e-05
_step	95
_runtime	12
_timestamp	1609380473
valid_loss	17386.76367
mae	131.85405
mse	17386.76367

epoch	10
train_loss	16397.92383
raw_loss	15865.48047
wd_0	0.01
sqr_mom_0	0.99
lr_0	0.0
mom_0	0.94999
eps_0	1e-05
_step	190
_runtime	21
_timestamp	1609380515
valid_loss	17260.64258
mae	131.37758
mse	17260.64258

epoch	20
train_loss	16070.27539
raw_loss	15582.51172
wd_0	0.01
sqr_mom_0	0.99
lr_0	0.0
mom_0	0.95
eps_0	1e-05
_step	380
_runtime	41
_timestamp	1609380563
valid_loss	17061.5625
mae	130.59042
mse	17061.5625

epoch	train_loss	valid_loss	mae	mse	time
0	16868.968750	17136.835938	130.904785	17136.835938	00:01
1	16848.242188	17440.337891	132.058838	17440.337891	00:01
2	16824.509766	17474.201172	132.184906	17474.201172	00:01
3	16797.003906	17489.435547	132.232437	17489.435547	00:01
4	16762.984375	17436.039062	132.027267	17436.039062	00:01
5	16720.851562	17305.435547	131.537766	17305.435547	00:01
6	16670.619141	17322.958984	131.612320	17322.958984	00:01
7	16614.996094	17159.644531	130.992523	17159.644531	00:01
8	16554.523438	17280.388672	131.454147	17280.388672	00:01
9	16493.986328	17240.910156	131.302536	17240.910156	00:01
10	16432.507812	17218.324219	131.215179	17218.324219	00:01
11	16371.637695	17195.798828	131.124664	17195.798828	00:01
12	16311.836914	17155.773438	130.968170	17155.773438	00:01
13	16253.294922	17144.005859	130.913879	17144.005859	00:01
14	16196.915039	16842.351562	129.750626	16842.351562	00:01
15	16141.201172	17083.568359	130.662430	17083.568359	00:01
16	16086.743164	16850.421875	129.759445	16850.421875	00:01
17	16034.440430	17008.308594	130.355576	17008.308594	00:01
18	15984.509766	16981.875000	130.236404	16981.875000	00:01
19	15935.457031	17011.630859	130.336441	17011.630859	00:01
20	15888.824219	16945.287109	130.072052	16945.287109	00:01
21	15844.999023	16983.332031	130.198898	16983.332031	00:01
22	15803.041016	16874.208984	129.768845	16874.208984	00:01
23	15763.515625	16790.880859	129.435135	16790.880859	00:01
24	15726.163086	16898.494141	129.841171	16898.494141	00:01
25	15691.458008	16870.443359	129.722198	16870.443359	00:01
26	15659.495117	16820.820312	129.534393	16820.820312	00:01
27	15630.248047	16775.466797	129.350357	16775.466797	00:01
28	15603.407227	16809.470703	129.479416	16809.470703	00:01
29	15579.450195	16743.490234	129.220825	16743.490234	00:01
30	15558.423828	16818.669922	129.505325	16818.669922	00:01
31	15540.255859	16741.880859	129.205276	16741.880859	00:01
32	15524.796875	16785.113281	129.369064	16785.113281	00:01
33	15511.831055	16712.759766	129.088699	16712.759766	00:01
34	15501.209961	16698.632812	129.032715	16698.632812	00:01
35	15492.711914	16695.896484	129.020630	16695.896484	00:01
36	15486.111328	16691.992188	129.004913	16691.992188	00:01
37	15481.150391	16688.164062	128.989838	16688.164062	00:01
38	15477.551758	16686.958984	128.985092	16686.958984	00:01
39	15475.038086	16686.474609	128.983246	16686.474609	00:01

epoch	40
train_loss	15475.03809
raw_loss	15026.24512
wd_0	0.01
sqr_mom_0	0.99
lr_0	0.0
mom_0	0.95
eps_0	1e-05
_step	760
_runtime	79
_timestamp	1609380654
valid_loss	16686.47461
mae	128.98325
mse	16686.47461

epoch	train_loss	valid_loss	mae	mse	time
0	16869.257812	17137.013672	130.905457	17137.013672	00:01
1	16850.804688	17441.402344	132.062790	17441.402344	00:01
2	16832.667969	17477.330078	132.196335	17477.330078	00:01
3	16813.412109	17503.720703	132.281219	17503.720703	00:01
4	16792.462891	17490.791016	132.222321	17490.791016	00:01
5	16769.287109	17451.078125	132.078705	17451.078125	00:01
6	16743.125000	17436.800781	132.027786	17436.800781	00:01
7	16713.744141	17420.609375	131.975861	17420.609375	00:01
8	16679.734375	17381.044922	131.829483	17381.044922	00:01
9	16642.068359	17375.595703	131.811722	17375.595703	00:01
10	16599.986328	17334.763672	131.659607	17334.763672	00:01
11	16554.986328	17331.441406	131.647339	17331.441406	00:01
12	16508.843750	17324.857422	131.622665	17324.857422	00:01
13	16461.917969	17276.437500	131.437561	17276.437500	00:01
14	16413.794922	17250.800781	131.340363	17250.800781	00:01
15	16364.436523	17204.869141	131.163635	17204.869141	00:01
16	16313.495117	17174.109375	131.044647	17174.109375	00:01
17	16261.653320	17224.951172	131.232269	17224.951172	00:01
18	16209.461914	17011.435547	130.402603	17011.435547	00:01
19	16154.955078	17082.398438	130.667343	17082.398438	00:01
20	16100.580078	17011.433594	130.377579	17011.433594	00:01
21	16045.573242	17017.248047	130.387314	17017.248047	00:01
22	15990.035156	16976.095703	130.221558	16976.095703	00:01
23	15934.102539	16887.833984	129.874191	16887.833984	00:01
24	15877.906250	16911.458984	129.940872	16911.458984	00:01
25	15822.751953	16914.113281	129.936508	16914.113281	00:01
26	15768.279297	16959.685547	130.088623	16959.685547	00:01
27	15713.018555	16998.724609	130.212753	16998.724609	00:01
28	15657.997070	16798.298828	129.425400	16798.298828	00:01
29	15603.409180	16786.078125	129.357727	16786.078125	00:01
30	15548.836914	16703.484375	129.030258	16703.484375	00:01
31	15494.398438	16708.650391	129.022079	16708.650391	00:01
32	15440.585938	16807.248047	129.390701	16807.248047	00:01
33	15388.107422	16435.195312	127.934128	16435.195312	00:01
34	15335.736328	16462.425781	128.016846	16462.425781	00:01
35	15284.082031	16694.603516	128.915863	16694.603516	00:01
36	15233.023438	15922.021484	125.921989	15922.021484	00:01
37	15182.677734	16471.378906	128.111618	16471.378906	00:01
38	15133.390625	16446.744141	128.022141	16446.744141	00:01
39	15084.399414	16454.222656	128.066132	16454.222656	00:01
40	15036.383789	16221.137695	127.166962	16221.137695	00:01
41	14989.958008	16126.601562	126.818680	16126.601562	00:01
42	14944.378906	16356.822266	127.715233	16356.822266	00:01
43	14899.675781	16018.284180	126.385941	16018.284180	00:01
44	14856.560547	15812.881836	125.601227	15812.881836	00:01
45	14814.570312	16360.370117	127.758057	16360.370117	00:01
46	14773.966797	15624.066406	124.857079	15624.066406	00:01
47	14734.716797	15811.144531	125.600220	15811.144531	00:01
48	14696.546875	15703.860352	125.172295	15703.860352	00:01
49	14659.550781	15820.836914	125.642372	15820.836914	00:01
50	14624.221680	16087.708984	126.694527	16087.708984	00:01
51	14590.236328	16414.070312	127.973030	16414.070312	00:01
52	14557.952148	15773.204102	125.446091	15773.204102	00:01
53	14527.762695	15752.934570	125.371223	15752.934570	00:01
54	14499.076172	15874.183594	125.853745	15874.183594	00:01
55	14471.650391	15869.708984	125.839699	15869.708984	00:01
56	14445.521484	15778.150391	125.471039	15778.150391	00:01
57	14420.866211	15710.672852	125.212860	15710.672852	00:01
58	14397.514648	15934.793945	126.101578	15934.793945	00:01
59	14375.751953	15700.195312	125.164146	15700.195312	00:02
60	14355.648438	15536.324219	124.518661	15536.324219	00:01
61	14337.143555	16217.024414	127.220970	16217.024414	00:01
62	14320.227539	15664.918945	125.032066	15664.918945	00:01
63	14304.718750	15597.861328	124.767197	15597.861328	00:01
64	14290.696289	15615.498047	124.836067	15615.498047	00:01
65	14277.873047	15618.549805	124.847084	15618.549805	00:01
66	14266.379883	15637.005859	124.922401	15637.005859	00:01
67	14256.192383	15610.528320	124.817131	15610.528320	00:01
68	14247.232422	15589.907227	124.736023	15589.907227	00:01
69	14239.430664	15618.666016	124.850433	15618.666016	00:01
70	14232.706055	15602.001953	124.784561	15602.001953	00:01
71	14226.975586	15599.855469	124.775902	15599.855469	00:01
72	14222.163086	15606.209961	124.801208	15606.209961	00:01
73	14218.184570	15606.207031	124.801468	15606.207031	00:01
74	14214.954102	15607.133789	124.805130	15607.133789	00:01
75	14212.393555	15610.329102	124.817856	15610.329102	00:01
76	14210.413086	15612.440430	124.826271	15612.440430	00:01
77	14208.930664	15613.489258	124.830475	15613.489258	00:01
78	14207.857422	15614.208984	124.833344	15614.208984	00:01
79	14207.110352	15614.526367	124.834625	15614.526367	00:01

epoch	80
train_loss	14207.11035
raw_loss	13846.10547
wd_0	0.01
sqr_mom_0	0.99
lr_0	0.0
mom_0	0.95
eps_0	1e-05
_step	1520
_runtime	158
_timestamp	1609380826
valid_loss	15614.52637
mae	124.83463
mse	15614.52637

epoch	1
train_loss	16874.99414
raw_loss	16366.72559
wd_0	0.01
sqr_mom_0	0.99
lr_0	0.0
mom_0	0.94879
eps_0	1e-05
_step	19
_runtime	5
_timestamp	1609380865
valid_loss	17144.14062
mae	130.93263
mse	17144.14062

epoch	▁▁▂▂▃▃▃▄▄▅▅▅▆▆▆▇▇██
train_loss	▂▁▁▂▂▂▄▅▅▆▆▇▇▇▇████
raw_loss	▂▁▂▃▃▃▇█▇██▇▆▇▇▇▇▇▄
wd_0	▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
sqr_mom_0	▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
lr_0	▂▂▄▆████▇▇▆▅▄▄▃▂▂▁▁
mom_0	█▇▅▃▁▁▁▁▂▂▃▄▅▅▆▇▇██
eps_0	▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
_step	▁▁▂▂▂▃▃▄▄▄▅▅▅▆▆▇▇▇██
_runtime	▁▁▁▃▃▃▃▃▃▃▃▃▃▃▃▃▃▃▆█
_timestamp	▁▁▁▃▃▃▃▃▃▃▃▃▃▃▃▃▃▃▆█
valid_loss	▁
mae	▁
mse	▁

epoch	train_loss	valid_loss	mae	mse	time
0	16876.732422	17141.242188	130.921616	17141.242188	00:01
1	16862.562500	17441.089844	132.061646	17441.089844	00:01
2	16851.972656	17481.369141	132.211151	17481.369141	00:01

epoch	train_loss	valid_loss	mae	mse	time
0	16878.900391	17145.847656	130.939163	17145.847656	00:01
1	16865.958984	17442.156250	132.065613	17442.156250	00:01
2	16850.605469	17480.759766	132.208832	17480.759766	00:01
3	16838.125000	17508.949219	132.301270	17508.949219	00:01
4	16829.718750	17522.125000	132.333313	17522.125000	00:01

epoch	train_loss	valid_loss	mae	mse	time
0	16880.458984	17149.529297	130.953201	17149.529297	00:01
1	16873.070312	17444.021484	132.072510	17444.021484	00:01
2	16859.412109	17481.353516	132.211044	17481.353516	00:01
3	16843.810547	17505.541016	132.288071	17505.541016	00:01
4	16828.460938	17524.246094	132.338242	17524.246094	00:01
5	16814.417969	17506.425781	132.268372	17506.425781	00:02
6	16802.154297	17502.203125	132.252838	17502.203125	00:01
7	16792.259766	17501.347656	132.250580	17501.347656	00:01
8	16784.847656	17502.875000	132.256836	17502.875000	00:01
9	16779.656250	17504.652344	132.263718	17504.652344	00:01

epoch	train_loss	valid_loss	mae	mse	time
0	16880.945312	17150.736328	130.957809	17150.736328	00:01
1	16877.128906	17446.263672	132.080872	17446.263672	00:01
2	16870.166016	17484.619141	132.222946	17484.619141	00:01
3	16859.164062	17517.058594	132.332047	17517.058594	00:01
4	16845.341797	17536.710938	132.387405	17536.710938	00:01
5	16830.156250	17517.066406	132.307373	17517.066406	00:01
6	16814.980469	17503.625000	132.256241	17503.625000	00:01
7	16799.580078	17491.130859	132.209869	17491.130859	00:01
8	16784.806641	17481.343750	132.177948	17481.343750	00:01
9	16770.957031	17477.386719	132.164749	17477.386719	00:01
10	16758.294922	17472.787109	132.151215	17472.787109	00:01
11	16746.857422	17469.808594	132.140518	17469.808594	00:01
12	16736.849609	17466.707031	132.129959	17466.707031	00:01
13	16728.126953	17465.542969	132.126495	17465.542969	00:01
14	16720.771484	17463.621094	132.119690	17463.621094	00:01
15	16714.699219	17462.357422	132.115433	17462.357422	00:01
16	16709.875000	17462.132812	132.114914	17462.132812	00:01
17	16706.187500	17462.005859	132.114578	17462.005859	00:01
18	16703.482422	17461.890625	132.114212	17461.890625	00:01
19	16701.585938	17461.886719	132.114212	17461.886719	00:01

epoch	train_loss	valid_loss	mae	mse	time
0	16881.080078	17151.042969	130.958969	17151.042969	00:01
1	16878.455078	17446.785156	132.082809	17446.785156	00:01
2	16874.832031	17485.751953	132.226974	17485.751953	00:01
3	16869.564453	17525.572266	132.364563	17525.572266	00:01
4	16862.164062	17551.330078	132.447708	17551.330078	00:01
5	16852.714844	17542.667969	132.409348	17542.667969	00:01
6	16841.560547	17535.132812	132.377441	17535.132812	00:01
7	16828.923828	17515.777344	132.302673	17515.777344	00:01
8	16815.234375	17508.582031	132.276382	17508.582031	00:01
9	16800.615234	17495.964844	132.230637	17495.964844	00:01
10	16786.003906	17484.613281	132.192505	17484.613281	00:01
11	16771.501953	17472.380859	132.149429	17472.380859	00:01
12	16757.445312	17466.017578	132.127762	17466.017578	00:01
13	16743.726562	17458.269531	132.101227	17458.269531	00:01
14	16730.601562	17453.017578	132.082962	17453.017578	00:01
15	16718.101562	17449.166016	132.069824	17449.166016	00:01
16	16706.189453	17442.000000	132.044708	17442.000000	00:01
17	16694.677734	17441.875000	132.045258	17441.875000	00:01
18	16683.830078	17435.544922	132.022278	17435.544922	00:01
19	16673.498047	17428.867188	131.998352	17428.867188	00:02
20	16663.734375	17423.363281	131.978271	17423.363281	00:01
21	16654.568359	17422.289062	131.974945	17422.289062	00:01
22	16645.996094	17415.966797	131.951950	17415.966797	00:01
23	16637.943359	17416.625000	131.955109	17416.625000	00:01
24	16630.496094	17410.580078	131.933075	17410.580078	00:01
25	16623.593750	17409.496094	131.929550	17409.496094	00:01
26	16617.287109	17410.453125	131.933716	17410.453125	00:01
27	16611.566406	17413.537109	131.946030	17413.537109	00:01
28	16606.435547	17412.398438	131.941986	17412.398438	00:01
29	16601.904297	17410.878906	131.936615	17410.878906	00:01
30	16597.939453	17411.490234	131.938995	17411.490234	00:01
31	16594.519531	17410.371094	131.935089	17410.371094	00:01
32	16591.611328	17410.490234	131.935699	17410.490234	00:01
33	16589.177734	17410.302734	131.935104	17410.302734	00:01
34	16587.187500	17410.230469	131.934921	17410.230469	00:01
35	16585.595703	17410.242188	131.935013	17410.242188	00:01
36	16584.357422	17410.320312	131.935364	17410.320312	00:01
37	16583.423828	17410.513672	131.936142	17410.513672	00:01
38	16582.751953	17410.679688	131.936752	17410.679688	00:01
39	16582.281250	17410.781250	131.937164	17410.781250	00:01

epoch	3
train_loss	16851.97266
raw_loss	16339.2666
wd_0	0.01
sqr_mom_0	0.99
lr_0	0.0
mom_0	0.94987
eps_0	1e-05
_step	57
_runtime	8
_timestamp	1609380891
valid_loss	17481.36914
mae	132.21115
mse	17481.36914

epoch	5
train_loss	16829.71875
raw_loss	16319.44043
wd_0	0.01
sqr_mom_0	0.99
lr_0	0.0
mom_0	0.94995
eps_0	1e-05
_step	95
_runtime	12
_timestamp	1609380921
valid_loss	17522.125
mae	132.33331
mse	17522.125

epoch	10
train_loss	16779.65625
raw_loss	16275.87793
wd_0	0.01
sqr_mom_0	0.99
lr_0	0.0
mom_0	0.94999
eps_0	1e-05
_step	190
_runtime	22
_timestamp	1609380949
valid_loss	17504.65234
mae	132.26372
mse	17504.65234

epoch	20
train_loss	16701.58594
raw_loss	16201.1748
wd_0	0.01
sqr_mom_0	0.99
lr_0	0.0
mom_0	0.95
eps_0	1e-05
_step	380
_runtime	42
_timestamp	1609380998
valid_loss	17461.88672
mae	132.11421
mse	17461.88672

epoch	40
train_loss	16582.28125
raw_loss	16091.3418
wd_0	0.01
sqr_mom_0	0.99
lr_0	0.0
mom_0	0.95
eps_0	1e-05
_step	760
_runtime	81
_timestamp	1609381087
valid_loss	17410.78125
mae	131.93716
mse	17410.78125

epoch	train_loss	valid_loss	mae	mse	time
0	16881.113281	17151.136719	130.959320	17151.136719	00:01
1	16878.820312	17446.955078	132.083435	17446.955078	00:01
2	16876.208984	17486.324219	132.229065	17486.324219	00:01
3	16873.048828	17526.808594	132.369080	17526.808594	00:01
4	16869.144531	17551.029297	132.448166	17551.029297	00:01
5	16864.316406	17553.990234	132.454880	17553.990234	00:01
6	16858.488281	17550.400391	132.439590	17550.400391	00:01
7	16851.685547	17542.806641	132.409210	17542.806641	00:01
8	16844.027344	17533.001953	132.370819	17533.001953	00:01
9	16835.484375	17523.636719	132.332626	17523.636719	00:01
10	16826.144531	17516.369141	132.304886	17516.369141	00:01
11	16816.181641	17510.248047	132.282364	17510.248047	00:01

epoch	train_loss	valid_loss	mae	mse	time
0	16500.857422	17058.966797	130.454163	17058.966797	00:01
1	16435.394531	16334.189453	127.655075	16334.189453	00:01
2	16343.923828	15895.993164	125.941399	15895.993164	00:01
3	16242.678711	15843.721680	125.743126	15843.721680	00:01
4	16139.079102	16056.191406	126.584099	16056.191406	00:01
5	16039.528320	15964.089844	126.224045	15964.089844	00:01
6	15948.698242	15934.723633	126.103943	15934.723633	00:01
7	15871.872070	15880.710938	125.885681	15880.710938	00:01
8	15807.691406	15862.152344	125.809784	15862.152344	00:01
9	15753.535156	15828.784180	125.677429	15828.784180	00:01

epoch	train_loss	valid_loss	mae	mse	time
0	15470.530273	15807.198242	125.590584	15807.198242	00:01
1	15450.349609	15748.249023	125.350815	15748.249023	00:01
2	15397.788086	15369.669922	123.832458	15369.669922	00:01
3	15323.241211	14496.727539	120.258461	14496.727539	00:01
4	15241.073242	14810.000000	121.528625	14810.000000	00:01
5	15157.637695	14561.288086	120.511833	14561.288086	00:01
6	15073.031250	14484.727539	120.184296	14484.727539	00:01
7	15001.869141	14237.068359	119.154198	14237.068359	00:01
8	14941.365234	14286.781250	119.361732	14286.781250	00:01
9	14890.690430	14274.882812	119.311737	14274.882812	00:01

epoch	train_loss	valid_loss	mae	mse	time
0	14604.194336	14319.080078	119.494308	14319.080078	00:01
1	14572.034180	14023.852539	118.248299	14023.852539	00:01
2	14517.897461	11494.937500	107.068962	11494.937500	00:01
3	14436.357422	11822.633789	108.572411	11822.633789	00:01
4	14340.881836	13002.192383	113.824600	13002.192383	00:01
5	14232.510742	13026.773438	113.895866	13026.773438	00:01
6	14134.999023	13047.250000	113.979614	13047.250000	00:01
7	14039.472656	12630.025391	112.135170	12630.025391	00:01
8	13951.371094	12370.928711	110.939743	12370.928711	00:01
9	13885.535156	12396.886719	111.074158	12396.886719	00:01

epoch	train_loss	valid_loss	mae	mse	time
0	13545.776367	12378.215820	110.988564	12378.215820	00:01
1	13506.545898	11457.753906	106.783272	11457.753906	00:01
2	13425.323242	9944.270508	99.477539	9944.270508	00:01
3	13313.986328	9618.911133	97.676544	9618.911133	00:01
4	13187.502930	9595.845703	97.499611	9595.845703	00:01
5	13072.986328	9962.482422	99.214157	9962.482422	00:01
6	12960.182617	10250.379883	100.751534	10250.379883	00:01
7	12849.628906	9783.357422	98.511055	9783.357422	00:01
8	12765.044922	10108.954102	100.156601	10108.954102	00:01
9	12695.324219	10490.105469	102.018394	10490.105469	00:01

epoch	train_loss	valid_loss	mae	mse	time
0	12330.806641	10802.211914	103.498978	10802.211914	00:01
1	12262.324219	10862.475586	103.793655	10862.475586	00:01
2	12160.761719	6844.447266	82.174149	6844.447266	00:01
3	12057.375977	8600.846680	91.995308	8600.846680	00:01
4	11919.283203	9083.191406	94.626465	9083.191406	00:01
5	11797.177734	8949.733398	93.899994	8949.733398	00:01
6	11677.922852	9467.136719	96.650116	9467.136719	00:01
7	11573.804688	9596.978516	97.406593	9596.978516	00:01
8	11483.841797	9797.712891	98.405647	9797.712891	00:01
9	11410.873047	9862.557617	98.694275	9862.557617	00:01