classifier.load_state_dict(weight, strict=False) learn = Learner( data_bunch, classifier, loss_func=Loss_combine_weighted_v2(), opt_func=SGD, metrics=[Metric_grapheme(), Metric_vowel(), Metric_consonant(), Metric_tot()] ) name = 'pytorch_model_simple50' logger = CSVLogger(learn, name) # ========================================================================================================================= learn.fit_one_cycle( cyc_len=160, max_lr=.01, wd=0.0, moms=0.5, div_factor=25, final_div=100, pct_start=0.015, callbacks=[logger, SaveModelCallback(learn, monitor='metric_tot', mode='max', name=name), MixUpCallback(learn)], ) # learn.fit( # 160, # lr=.01, wd=0.00001, # callbacks=[ # logger, # SaveModelCallback(learn, monitor='metric_tot', mode='max', name=name), # MixUpCallback(learn), # ReduceLROnPlateauCallback(learn, patience=10, factor=0.5, min_lr=.0001) # ] # )
learn.clip_grad = 1.0 # learn.split([classifier.predictor.lin_layers]) learn.unfreeze() # In[9]: learn.fit_one_cycle(96, max_lr=1e-2, wd=0., pct_start=0.0, div_factor=100, callbacks=[ logger, SaveModelCallback(learn, monitor='metric_tot', mode='max', name=logging_name), MixUpCallback(learn) ]) # learn.fit( # 160, # lr=.01, # wd=0., # callbacks=[ # logger, # SaveModelCallback(learn, monitor='metric_tot', mode='max', name=logging_name), # ReduceLROnPlateauCallback(learn, patience=10, factor=.1, min_lr=1e-5), # MixUpCallback(learn, alpha=.8), # ] # )
# ================================================================== classifier = Seresnext50MishFrac() logging_name = 'new_baseline_seresnext50_Mish_Frac' learn = Learner( data_bunch, classifier, loss_func=Loss_combine_weighted_v2(), opt_func=Over9000, metrics=[Metric_grapheme(), Metric_vowel(), Metric_consonant(), Metric_tot()] ) logger = CSVLogger(learn, logging_name) learn.clip_grad = 1.0 # ================================================================== learn.fit_one_cycle( 120, max_lr=1e-2, wd=0., pct_start=0.0, div_factor=100, callbacks=[logger, SaveModelCallback(learn, monitor='metric_tot', mode='max', name=logging_name), MixUpCallback(learn, alpha=1)] )
learn = Learner(data_bunch, classifier, loss_func=Loss_combine_weighted(), opt_func=Over9000, metrics=[ Metric_grapheme(), Metric_vowel(), Metric_consonant(), Metric_tot() ]) logger = CSVLogger(learn, logging_name) learn.clip_grad = 1.0 # ================================================================== learn.fit_one_cycle(64, max_lr=1e-2, wd=0., pct_start=0.0, div_factor=100, callbacks=[ logger, SaveModelCallback(learn, monitor='metric_tot', mode='max', name=logging_name), MixUpCallback(learn, alpha=1.) ])