Learner

nbeats_learner[source]

nbeats_learner(dbunch:TSDataLoaders, output_channels=None, metrics=None, cbs=None, b_loss=0.0, loss_func=None, opt_func=None, stack_types=('trend', 'seasonality'), nb_blocks_per_stack=3, horizon=5, lookback=10, thetas_dim=None, share_weights_in_layers=True, layers=[1024, 512])

Build a N-Beats style learner

Example

horizon, lookback = 10,20
items = dummy_data_generator(75, 10, nrows=10)
data = TSDataLoaders.from_items(items, horizon = horizon, lookback=lookback, step=3, valid_pct=.5
                               )
data.show_batch()

(1, 85)
Train:110; Valid: 20; Test 10

learn = nbeats_learner(data, metrics = NBeatsLossPart(0,-9,'Last'))

learn.fit_flat_cos(5, 3e-2)
learn.recorder.plot_loss()
learn.recorder.plot_sched()

epoch	train_loss	valid_loss	Last	mae	smape	theta	b_loss	f_loss	f_smape	time
0	4.503861	4.276122	0.198660	1.066485	1.702529	0.090910	nan	nan	0.172096	00:01
1	3.882679	3.882362	0.181528	1.020999	1.547903	0.132729	nan	nan	0.154726	00:00
2	3.579792	3.691950	0.132820	0.982745	1.431832	0.306930	nan	nan	0.150932	00:00
3	3.466791	2.177217	0.084399	0.745303	0.977235	0.987222	nan	nan	0.097492	00:00
4	3.283105	2.354932	0.079065	0.751035	0.950528	1.304956	nan	nan	0.097970	00:00

learn.show_results()

learn.show_results(0)

learn.n_beats_attention.means()

	trend0_0	trend0_1	seasonality1_0	seasonality1_1	seasonality1_2	seasonality1_3
theta_0_mean	-0.29973587	-0.20294055	-0.03109547	-0.11714697	-0.00836199	0.11202081
theta_0_std	0.61486304	0.8000404	0.34559253	0.33142605	0.32781237	0.3299975
theta_1_mean	-0.013373043	-0.00935214	-0.01582453	0.011355982	-0.0012693154	-0.012632762
theta_1_std	0.019088455	0.023187334	0.14136028	0.086675294	0.11071833	0.073259145
theta_2_mean	1.7291142e-05	-0.0008753341	0.36068255	-0.044498563	-0.005808751	0.0013584372
theta_2_std	0.0017255846	0.0018285766	0.8355213	0.8352149	0.01987379	0.01662062
theta_3_mean	0.0024652497	-7.876513e-05	NaN	-0.0310319	-0.39758366	0.15243284
theta_3_std	0.03500661	0.0002167299	NaN	0.06423591	0.790334	0.67478925
theta_4_mean	0.01688141	0.010520181	NaN	NaN	0.032521922	0.0023262694
theta_4_std	0.20457159	0.1036169	NaN	NaN	0.052365586	0.038209
theta_5_mean	0.021814847	-0.091876894	NaN	NaN	NaN	0.0032276318
theta_5_std	0.1277609	0.24249457	NaN	NaN	NaN	0.0056538703
att_mean	0.509504	0.506815	0.481883	0.4948	0.501913	0.521625
att_std	0.19548	0.252551	0.274901	0.20862	0.221856	0.197302
theta_6_mean	NaN	0.06229969	NaN	NaN	NaN	NaN
theta_6_std	NaN	0.13474756	NaN	NaN	NaN	NaN
theta_7_mean	NaN	0.27522308	NaN	NaN	NaN	NaN
theta_7_std	NaN	0.4069584	NaN	NaN	NaN	NaN