how to avoid cross lines?

[GodsDusk]

graph TD

candidate_opinion[/"candidate_opinion(batch_size, seq_len)"/] --"sum(-1)"--> candidate_opinion_sum
candidate_opinion_sum --> opinion_denominator

candidate_opinion --> B["(batch_size, 1, feature_size)"]

B --> candidate_opinion_rep
opinion_denominator --> candidate_opinion_rep["candidate_opinion_rep(batch_size, feature_size)"]

aspect_input_ids[/aspect_input_ids/] --> Bert
aspect_token_type_ids[/aspect_token_type_ids/] --> Bert
aspect_attention_mask[/aspect_attention_mask/] --> Bert
Bert --> pooled_outputs["pooled_outputs (batch_size,seq_len,feature_size)"]

candidate_aspect[/"candidate_aspect(batch_size, seq_len)"/] --"sum(-1)"--> candidate_aspect_sum["candidate_aspect_sum(batch_size, )"]
candidate_aspect_sum --> aspect_denominator

candidate_aspect --> A["(batch_size, 1, feature_size)"]
A --> candidate_aspect_rep
aspect_denominator --> candidate_aspect_rep["candidate_opinion_rep(batch_size, feature_size)"]

pooled_outputs --> A
pooled_outputs --> B

Loading

there are two lines crossed left, how to avoid them?

here is the raw code:

graph TD

candidate_opinion[/"candidate_opinion(batch_size, seq_len)"/] --"sum(-1)"--> candidate_opinion_sum
candidate_opinion_sum --> opinion_denominator

candidate_opinion --> B["(batch_size, 1, feature_size)"]

B --> candidate_opinion_rep
opinion_denominator --> candidate_opinion_rep["candidate_opinion_rep(batch_size, feature_size)"]

aspect_input_ids[/aspect_input_ids/] --> Bert
aspect_token_type_ids[/aspect_token_type_ids/] --> Bert
aspect_attention_mask[/aspect_attention_mask/] --> Bert
Bert --> pooled_outputs["pooled_outputs (batch_size,seq_len,feature_size)"]

candidate_aspect[/"candidate_aspect(batch_size, seq_len)"/] --"sum(-1)"--> candidate_aspect_sum["candidate_aspect_sum(batch_size, )"]
candidate_aspect_sum --> aspect_denominator

candidate_aspect --> A["(batch_size, 1, feature_size)"]
A --> candidate_aspect_rep
aspect_denominator --> candidate_aspect_rep["candidate_opinion_rep(batch_size, feature_size)"]

pooled_outputs --> A
pooled_outputs --> B

[Yankie]

Change order of definitions :)

graph TD

candidate_opinion --> B["(batch_size, 1, feature_size)"]
opinion_denominator --> candidate_opinion_rep["candidate_opinion_rep(batch_size, feature_size)"]
candidate_opinion[/"candidate_opinion(batch_size, seq_len)"/] --"sum(-1)"--> candidate_opinion_sum
candidate_opinion_sum --> opinion_denominator


B --> candidate_opinion_rep

aspect_input_ids[/aspect_input_ids/] --> Bert
aspect_token_type_ids[/aspect_token_type_ids/] --> Bert
aspect_attention_mask[/aspect_attention_mask/] --> Bert
Bert --> pooled_outputs["pooled_outputs (batch_size,seq_len,feature_size)"]

candidate_aspect[/"candidate_aspect(batch_size, seq_len)"/] --"sum(-1)"--> candidate_aspect_sum["candidate_aspect_sum(batch_size, )"]
candidate_aspect_sum --> aspect_denominator

candidate_aspect --> A["(batch_size, 1, feature_size)"]
A --> candidate_aspect_rep
aspect_denominator --> candidate_aspect_rep["candidate_opinion_rep(batch_size, feature_size)"]

pooled_outputs --> A
pooled_outputs --> B

Loading

[GodsDusk]

is there any principles? or just keep trying manually?

[Yankie]

I'm not sure, but I keep changing the order of statements in the hope it fixes itself :) чт, 16 мая 2024 г. в 04:57, GodsDusk ***@***.***>:

…

is there any principles? or just keep trying manually? — Reply to this email directly, view it on GitHub <#4045 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGGWLQTFTKRZIWEI3YL24DZCQG7HAVCNFSM6AAAAAAUKUMM66VHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4TINJSGI2DK> . You are receiving this because you commented.Message ID: ***@***.***>